CA2383760A1 - Fish detection method using sonar data - Google Patents
Fish detection method using sonar data Download PDFInfo
- Publication number
- CA2383760A1 CA2383760A1 CA002383760A CA2383760A CA2383760A1 CA 2383760 A1 CA2383760 A1 CA 2383760A1 CA 002383760 A CA002383760 A CA 002383760A CA 2383760 A CA2383760 A CA 2383760A CA 2383760 A1 CA2383760 A1 CA 2383760A1
- Authority
- CA
- Canada
- Prior art keywords
- echogram
- echo
- filter
- fish
- objects
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Abandoned
Links
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
- G01S7/52—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S15/00
- G01S7/523—Details of pulse systems
- G01S7/526—Receivers
- G01S7/527—Extracting wanted echo signals
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S15/00—Systems using the reflection or reradiation of acoustic waves, e.g. sonar systems
- G01S15/02—Systems using the reflection or reradiation of acoustic waves, e.g. sonar systems using reflection of acoustic waves
- G01S15/06—Systems determining the position data of a target
- G01S15/46—Indirect determination of position data
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S15/00—Systems using the reflection or reradiation of acoustic waves, e.g. sonar systems
- G01S15/88—Sonar systems specially adapted for specific applications
- G01S15/96—Sonar systems specially adapted for specific applications for locating fish
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
- G01S7/52—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S15/00
- G01S7/539—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S15/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
Landscapes
- Engineering & Computer Science (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Physics & Mathematics (AREA)
- Computer Networks & Wireless Communication (AREA)
- General Physics & Mathematics (AREA)
- Acoustics & Sound (AREA)
- Measurement Of Velocity Or Position Using Acoustic Or Ultrasonic Waves (AREA)
Abstract
The invention relates to a method of detection of objects, preferably fish, from sonar data. Echo signals from repeated sonar measurements are digitised and registered in a two-dimensional data presentation where echo distance, alternatively the duration of the signal, constitutes one dimension, and the measurement number, alternatively the time of measurement, constitutes the other dimension. Once the desired size of the data presentation is obtained, two-dimensional filtration, adaptive thresholding and morphological processing are carried out to enable an echogram having less noise and more reliable detection of desired objects to be produced. Additional shape analysis further enhances the detection.
Description
FISH DETECTION METHOD USING SONAR DATA
The invention relates to a method for analysing data from single-beam, double-beam and split-beam echo sounders, preferably for the detection of fish. In the analysis s echoes from individual objects are detected and their origin determined.
Data in this case means digitised echo information recorded using a vertical or horizontal, stationary or mobile echo sounder in a river, a watercourse, a lake or the sea.
Individual objects may be fish swimming alone, but may also be objects of any other type that are the target of a study.
io Introduction Pollution, the spread of new diseases and overfishing are some of the reasons why many fish stocks today have been reduced or even threatened. For the bodies or agencies responsible for the management of these resources, it is therefore important to find is methods that can measure the size and development of the stocks.
Traditionally, fish catch data has been used in measurements of this kind, but catch data is an unreliable and selective method which also has an impact on the stocks. For marine species the use of echo sounders has long been a recognised method and a great deal of equipment and many theories have been developed for the purpose of estimating fish through the zo use of echo sounders mounted in boats. In shallow rivers and lakes.
however, the situation is different. Here, it is often not expedient to use vertical echo sounders because the range of coverage would then be too small. Recently, however, field studies have been carried out involving echo sounders having horizontal, narrow beams adapted to water depth and bottom profile.
z~
In particular in connection with the recording of salmon migration in larger rivers, tests involving the use of echo sounders have been carried out. In Norn~av the rivers Numedalslagen and Tana have been tested. In Finland some tests have been conducted in the river Torneo. Other countries that have carried out tests are England, Scotland, so the USA and Canada, to mention just a few.
In such tests it has been usual to find places in the river where the conical beam profile can follow close to the bottom and the surface. This is to give as wide a coverage of the river cross-section as possible without interference from echoes from stones on the 3s bottom. The echo data collected has then been studied to determine the number and size of the fish that have passed by. However, the use of echo sounders iz rivers is not without its problems because the signal to noise ratio here is offer, r;,uch poorer than in open water. Th is due to the proximity of the sound beam to the bottom and the surface, the fact that fish have poorer reflection properties from the flank, and the fact that there are far more different echo sources such as drifting objects and air bubbles.
Often the use of traditional analysis methods on data from rivers does not work in a very s satisfactory manner.
In small rivers where the fish can be guided past narrow counting stations, it is possible to use alternative methods such as mechanical counters, video, light or conductivity recorders, but for wide rivers echo sounders are often the only alternative to catch data.
io Therefore it is important to improve the method.
Description of the existing method Traditionally, an echo sounder is only capable of measuring distance to an object in the water. A short sound pulse is emitted, and the echo sounder measures the time interval is between the emission of the pulse and the return of its echo. With knowledge of the speed of sound, the distance can then be calculated. In practice, a continuous echo signal will be generated after the emission of a sound pulse because all changes of density in bodies of water will reflect some energy back to the echo sounder.
After some time the signal emitted will be so weak that it will not be possible to measure any zo more echoes, and then a new sound pulse can be emitted.
To be able to measure the movement of an object and not just the distance, echo sounders equipped with several listening elements having different geometric orientations are used. By using three or more elements, the three-dimensional position as of an object can be determined on the basis of time differences in the echo. This gives a total of five dimensions: time, intensity, distance and two angular positions.
Echo sounders equipped with means for determining the position of objects in the beam are often called SoNaR (Sound Navigation and Ranging), and not echo sounders. A
much used five-dimensional echo sounder is Simrad's EY500 split-beam sounder, which 3o employs four listening elements.
There is one exception when it comes to the possibility of determining the angular position of an object. This is when two or more objects are located at the same distance but at different points in the beam (multiple echo). The echo from the objects will then ss reach the echo sounder simultaneously, and it becomes difficult to find clearly defined angle measurements which indicate the location of the two objects. The same is true of the echo from objects that are large in relation to the width/height of the sound beam.
Bottom echoes are one example of such objects. By putting together the echo signals from many subsequent sound pulses (pings), a mufti-dimensional image or echogram can be generated that shows the development of the echo intensity over time within the range ofthe sounder. This image contains information about passing objects as well as s information about the general level of background reflection of the waters, and it is therefore important to be able to distinguish between echoes from background and echoes from rigid objects.
Traditionally, this separation is made using a single echo detector (SED).
This detector io works with echoes from individual pings. Echoes from a single object of the size of a fish will usually reflect a sound pulse having approximately the same duration as the emitted sound pulse. The shape will grow monotonously to a maximum and will then diminish monotonously. When using a split-beam sounder, the angle measurements will also be relatively stable. However, echoes from multiple objects will often have is longer duration, be mufti-peaked and have great variations in the angle measurements.
A SED uses one or more of these properties to detect single echoes.
Acknowledged pulses are traditionally presented as "dots" in an SED echogram. The difference from an ordinary echogram is that the SED echogram only shows echoes from single detections. To avoid confusion between ordinary amplitude echograms and echograms ao based on SED information, we will in the rest of the text use the designations amp echogram and SED echogram respectively.
By joining together all individual detections that originate from one and the same object to form tracks it is possible to calculate the number of objects, their movement and size.
zs Joining together echoes to form tracks is called tracking. Automatic tracking algorithms often use forms of neighbourhood algorithms. These assess and join together echoes that are close in time and space.
In cases where there are several different object types that produce single echoes or 3o where it is desirable to separate different sizes of objects, a form of classification of the generated tracks is required. Here, properties such as size, direction of travel and velocity are used. In a river it is usual to distinguish objects that move with and against the current. The method can be summarised as follows:
~ Collecting and storing sonar data ss ~ SED detection ~ Combination of echoes to form tracks ~ Classification of tracks Problems associated with the existing method When the signal to noise ratio is poor, two problems arise. The single echo detector _ may start to detect echoes based on the background noise, and it may start to reject s echoes from target objects. This makes its difficult for both humans and automatic methods to identify the target objects in the SED echograms. Automatic methods will tend to generate quantities of tracks based on noise echoes whilst tracks from target objects are overlooked, divided into many smaller tracks or mixed with the noise echoes. Examples of sonar echograms with good and poor noise to sound ratio are io shown in Figure 8 and Figure 9. Sonar data with low signal to noise ratio is found in particular in shallow rivers where noise production is great, but also when measuring small objects and when measuring at a great distance because the sound intensity decreases by 1/R~4 where R is distance in metres.
is The reason echoes from target objects are rejected in the SED detector at a low signal to noise ratio is that they fail to satisfy one or more of the demands made by the detector.
For fish in rivers, we have shown that the echo signal from fish can be rejected both because of multi-peaks, unduly long pulse duration and excessive spread in the angle measurements.
The object of the invention It is therefore an object of the invention to overcome the disadvantages of the known methods in such manner that a more reliable detection of desired objects in rivers, watercourses, lakes and seas is obtained on the basis of echograms recorded by echo zs sounder or sonar measurements.
Brief description of the invention The aforementioned object is achieved in that the invention provides a method characterised by the features set forth in patent claim 1 below.
The method of the invention is also characterised by the advantageous features set forth in the attached dependent patent claims.
Brief description of the drawings Fig. 1 is a schematic illustration of the principle of a standard hydroacoustic measuring method. From the echo sounder sound passes through a Time Variable Gain amplifier (TVG) which compensates for geometric loss of echo intensity. An amplitude detector is then used to remove the carrier frequency before the amp echogram can be shown. In split-beam echo, sounders, a four-channel amplifier is used and also a phase detector.
s Fig. 2 shows a SED echogram recorded using a horizontal, stationary echo sounder in a river. The SED echogram shows many echoes from background noise. Some echoes may originate from passing fish. The relatively even horizontal lines are produced as a result of stationary io bottom structures. (The vertical axis is distance from the echo sounder in metres whilst the horizontal axis indicates time.) Fig. 3 is a block diagram of the main elements of the new method of analysis.
is Fig. 4 shows the principle of thresholding on cross-filtration. The result of filtrations and subsequent thresholding is also shown as 3D echograms in the attachment.
Fig. 5 shows example of echograms including small regions containing echoes 2o from noise and larger regions containing echoes from fish and drifting objects.
Fig. 6 is a block diagram showing combined use of pulse peak and single echo detection in order to detect and evaluate tracks in a detected region.
Zs Pulse peaks are indicated by round dark dots. Single echo detection uses angle and amplitude data from the detected regions on the basis of detected pulse peaks. In this example the tracking algorithm finds that the region contains echoes from one object. The two pulse peaks that could have formed an additional track are rejected by the demand for a so minimum number of echoes.
Fig. 7 shows an example of a detected region with pulse peaks. Only one region is shown, but it has two branches. In the middle of each branch there are pulse peaks from objects which upon subsequent analysis show 3s a movement against the current. It is therefore highly probable that the tracks represent salmon. Original data was recorded using a horizontal split-beam sounder in the river Tana in the summer of 1999. The objects passed the sounder on 19 July at 1325 hours at a distance of 31 metres.
Classification to determine the origin of the objects.
Fig. 8 shows an example of angle data for a detected object. The ellipse s indicates the cross-section of the emitted ultrasonic beam, whilst the axes show angle in relation to beam centre. Calculation of direction of travel and velocity shows that this object is moving against the current at a speed of 0.59 m/sec. It is then easy to classify the object as a fish migrating upstream. +alo indicates angle to the surface whilst +ath ~o indicates angle to the current direction.
Fig. 9 is an example of an echogram resulting from vertical recording of small fish in calm water. The recording was made under ice in a Norwegian lake called Semsvannet in March 1999, and is an illustration of good is signal to noise ratio with calm and weak background noise.
Fig. 10 is an example of an echogram resulting from a horizontal recording of large fish in a river. The figure illustrates low signal to noise ratio with great variations in the background intensity.
Fig. 11 is an example of an echogram of two fish in the river Tana. Original echogram. Horizontal echo sounder.
Fig. 12 is an example of an echogram of two fish in the river Tana. Data has 2s been filtered using a median 3x7 filter to highlight objects and remove noise.
Fig. 13 is an example of an echogram of two fish in the river Tana. Data has been filtered using a 55x7 median filter to detect background.
Fig. 14 is an example of an echogram of two fish in the river Tana. The threshold with the background image as threshold value.
Fig. 15 is an example of the processing of an echogram having a particularly 3s poor signal/noise ratio of: A) an SED echogram; B) an amp echogram;
C) result of thresholding using a 1-3 foreground median filter and a 1x99 background median filter; D) result of a growth operation and a 3x3 median filtration (note here the reduction in the number of missing echoes in relation to the preceding image); E) result after removing regions having short perimeters; F) contour detection of the region. The area inside the contour will now be analysed using a pulse peak analyser s or a single echo detector, and detected pulses will then be joined together to form tracks and be classified. This is not shown in the figure.
Fig. 16 shows an echogram as a result of median pass filtration using different filter dimensions. Dimensions are given as height times width. It can be to seen how long, narrow filters suppress target objects, whilst medium width, short filters suppress noise whilst the useful signal is highlighted.
Fig. 17 illustrates the principle for selection of critical frequencies in the filter, in this case shown only for one dimension.
Basis for development of a new method of analysis After working for some considerable time using existing methods, we were able to make the following important observations:
1. Missing echoes in tracks from fish combined with many noise-based echoes zo cause traditional tracking algorithms to fail.
2. An amp echogram contains more information than the SED echogram. More echoes from fish, pulse width, pulse shape and surroundings.
The invention relates to a method for analysing data from single-beam, double-beam and split-beam echo sounders, preferably for the detection of fish. In the analysis s echoes from individual objects are detected and their origin determined.
Data in this case means digitised echo information recorded using a vertical or horizontal, stationary or mobile echo sounder in a river, a watercourse, a lake or the sea.
Individual objects may be fish swimming alone, but may also be objects of any other type that are the target of a study.
io Introduction Pollution, the spread of new diseases and overfishing are some of the reasons why many fish stocks today have been reduced or even threatened. For the bodies or agencies responsible for the management of these resources, it is therefore important to find is methods that can measure the size and development of the stocks.
Traditionally, fish catch data has been used in measurements of this kind, but catch data is an unreliable and selective method which also has an impact on the stocks. For marine species the use of echo sounders has long been a recognised method and a great deal of equipment and many theories have been developed for the purpose of estimating fish through the zo use of echo sounders mounted in boats. In shallow rivers and lakes.
however, the situation is different. Here, it is often not expedient to use vertical echo sounders because the range of coverage would then be too small. Recently, however, field studies have been carried out involving echo sounders having horizontal, narrow beams adapted to water depth and bottom profile.
z~
In particular in connection with the recording of salmon migration in larger rivers, tests involving the use of echo sounders have been carried out. In Norn~av the rivers Numedalslagen and Tana have been tested. In Finland some tests have been conducted in the river Torneo. Other countries that have carried out tests are England, Scotland, so the USA and Canada, to mention just a few.
In such tests it has been usual to find places in the river where the conical beam profile can follow close to the bottom and the surface. This is to give as wide a coverage of the river cross-section as possible without interference from echoes from stones on the 3s bottom. The echo data collected has then been studied to determine the number and size of the fish that have passed by. However, the use of echo sounders iz rivers is not without its problems because the signal to noise ratio here is offer, r;,uch poorer than in open water. Th is due to the proximity of the sound beam to the bottom and the surface, the fact that fish have poorer reflection properties from the flank, and the fact that there are far more different echo sources such as drifting objects and air bubbles.
Often the use of traditional analysis methods on data from rivers does not work in a very s satisfactory manner.
In small rivers where the fish can be guided past narrow counting stations, it is possible to use alternative methods such as mechanical counters, video, light or conductivity recorders, but for wide rivers echo sounders are often the only alternative to catch data.
io Therefore it is important to improve the method.
Description of the existing method Traditionally, an echo sounder is only capable of measuring distance to an object in the water. A short sound pulse is emitted, and the echo sounder measures the time interval is between the emission of the pulse and the return of its echo. With knowledge of the speed of sound, the distance can then be calculated. In practice, a continuous echo signal will be generated after the emission of a sound pulse because all changes of density in bodies of water will reflect some energy back to the echo sounder.
After some time the signal emitted will be so weak that it will not be possible to measure any zo more echoes, and then a new sound pulse can be emitted.
To be able to measure the movement of an object and not just the distance, echo sounders equipped with several listening elements having different geometric orientations are used. By using three or more elements, the three-dimensional position as of an object can be determined on the basis of time differences in the echo. This gives a total of five dimensions: time, intensity, distance and two angular positions.
Echo sounders equipped with means for determining the position of objects in the beam are often called SoNaR (Sound Navigation and Ranging), and not echo sounders. A
much used five-dimensional echo sounder is Simrad's EY500 split-beam sounder, which 3o employs four listening elements.
There is one exception when it comes to the possibility of determining the angular position of an object. This is when two or more objects are located at the same distance but at different points in the beam (multiple echo). The echo from the objects will then ss reach the echo sounder simultaneously, and it becomes difficult to find clearly defined angle measurements which indicate the location of the two objects. The same is true of the echo from objects that are large in relation to the width/height of the sound beam.
Bottom echoes are one example of such objects. By putting together the echo signals from many subsequent sound pulses (pings), a mufti-dimensional image or echogram can be generated that shows the development of the echo intensity over time within the range ofthe sounder. This image contains information about passing objects as well as s information about the general level of background reflection of the waters, and it is therefore important to be able to distinguish between echoes from background and echoes from rigid objects.
Traditionally, this separation is made using a single echo detector (SED).
This detector io works with echoes from individual pings. Echoes from a single object of the size of a fish will usually reflect a sound pulse having approximately the same duration as the emitted sound pulse. The shape will grow monotonously to a maximum and will then diminish monotonously. When using a split-beam sounder, the angle measurements will also be relatively stable. However, echoes from multiple objects will often have is longer duration, be mufti-peaked and have great variations in the angle measurements.
A SED uses one or more of these properties to detect single echoes.
Acknowledged pulses are traditionally presented as "dots" in an SED echogram. The difference from an ordinary echogram is that the SED echogram only shows echoes from single detections. To avoid confusion between ordinary amplitude echograms and echograms ao based on SED information, we will in the rest of the text use the designations amp echogram and SED echogram respectively.
By joining together all individual detections that originate from one and the same object to form tracks it is possible to calculate the number of objects, their movement and size.
zs Joining together echoes to form tracks is called tracking. Automatic tracking algorithms often use forms of neighbourhood algorithms. These assess and join together echoes that are close in time and space.
In cases where there are several different object types that produce single echoes or 3o where it is desirable to separate different sizes of objects, a form of classification of the generated tracks is required. Here, properties such as size, direction of travel and velocity are used. In a river it is usual to distinguish objects that move with and against the current. The method can be summarised as follows:
~ Collecting and storing sonar data ss ~ SED detection ~ Combination of echoes to form tracks ~ Classification of tracks Problems associated with the existing method When the signal to noise ratio is poor, two problems arise. The single echo detector _ may start to detect echoes based on the background noise, and it may start to reject s echoes from target objects. This makes its difficult for both humans and automatic methods to identify the target objects in the SED echograms. Automatic methods will tend to generate quantities of tracks based on noise echoes whilst tracks from target objects are overlooked, divided into many smaller tracks or mixed with the noise echoes. Examples of sonar echograms with good and poor noise to sound ratio are io shown in Figure 8 and Figure 9. Sonar data with low signal to noise ratio is found in particular in shallow rivers where noise production is great, but also when measuring small objects and when measuring at a great distance because the sound intensity decreases by 1/R~4 where R is distance in metres.
is The reason echoes from target objects are rejected in the SED detector at a low signal to noise ratio is that they fail to satisfy one or more of the demands made by the detector.
For fish in rivers, we have shown that the echo signal from fish can be rejected both because of multi-peaks, unduly long pulse duration and excessive spread in the angle measurements.
The object of the invention It is therefore an object of the invention to overcome the disadvantages of the known methods in such manner that a more reliable detection of desired objects in rivers, watercourses, lakes and seas is obtained on the basis of echograms recorded by echo zs sounder or sonar measurements.
Brief description of the invention The aforementioned object is achieved in that the invention provides a method characterised by the features set forth in patent claim 1 below.
The method of the invention is also characterised by the advantageous features set forth in the attached dependent patent claims.
Brief description of the drawings Fig. 1 is a schematic illustration of the principle of a standard hydroacoustic measuring method. From the echo sounder sound passes through a Time Variable Gain amplifier (TVG) which compensates for geometric loss of echo intensity. An amplitude detector is then used to remove the carrier frequency before the amp echogram can be shown. In split-beam echo, sounders, a four-channel amplifier is used and also a phase detector.
s Fig. 2 shows a SED echogram recorded using a horizontal, stationary echo sounder in a river. The SED echogram shows many echoes from background noise. Some echoes may originate from passing fish. The relatively even horizontal lines are produced as a result of stationary io bottom structures. (The vertical axis is distance from the echo sounder in metres whilst the horizontal axis indicates time.) Fig. 3 is a block diagram of the main elements of the new method of analysis.
is Fig. 4 shows the principle of thresholding on cross-filtration. The result of filtrations and subsequent thresholding is also shown as 3D echograms in the attachment.
Fig. 5 shows example of echograms including small regions containing echoes 2o from noise and larger regions containing echoes from fish and drifting objects.
Fig. 6 is a block diagram showing combined use of pulse peak and single echo detection in order to detect and evaluate tracks in a detected region.
Zs Pulse peaks are indicated by round dark dots. Single echo detection uses angle and amplitude data from the detected regions on the basis of detected pulse peaks. In this example the tracking algorithm finds that the region contains echoes from one object. The two pulse peaks that could have formed an additional track are rejected by the demand for a so minimum number of echoes.
Fig. 7 shows an example of a detected region with pulse peaks. Only one region is shown, but it has two branches. In the middle of each branch there are pulse peaks from objects which upon subsequent analysis show 3s a movement against the current. It is therefore highly probable that the tracks represent salmon. Original data was recorded using a horizontal split-beam sounder in the river Tana in the summer of 1999. The objects passed the sounder on 19 July at 1325 hours at a distance of 31 metres.
Classification to determine the origin of the objects.
Fig. 8 shows an example of angle data for a detected object. The ellipse s indicates the cross-section of the emitted ultrasonic beam, whilst the axes show angle in relation to beam centre. Calculation of direction of travel and velocity shows that this object is moving against the current at a speed of 0.59 m/sec. It is then easy to classify the object as a fish migrating upstream. +alo indicates angle to the surface whilst +ath ~o indicates angle to the current direction.
Fig. 9 is an example of an echogram resulting from vertical recording of small fish in calm water. The recording was made under ice in a Norwegian lake called Semsvannet in March 1999, and is an illustration of good is signal to noise ratio with calm and weak background noise.
Fig. 10 is an example of an echogram resulting from a horizontal recording of large fish in a river. The figure illustrates low signal to noise ratio with great variations in the background intensity.
Fig. 11 is an example of an echogram of two fish in the river Tana. Original echogram. Horizontal echo sounder.
Fig. 12 is an example of an echogram of two fish in the river Tana. Data has 2s been filtered using a median 3x7 filter to highlight objects and remove noise.
Fig. 13 is an example of an echogram of two fish in the river Tana. Data has been filtered using a 55x7 median filter to detect background.
Fig. 14 is an example of an echogram of two fish in the river Tana. The threshold with the background image as threshold value.
Fig. 15 is an example of the processing of an echogram having a particularly 3s poor signal/noise ratio of: A) an SED echogram; B) an amp echogram;
C) result of thresholding using a 1-3 foreground median filter and a 1x99 background median filter; D) result of a growth operation and a 3x3 median filtration (note here the reduction in the number of missing echoes in relation to the preceding image); E) result after removing regions having short perimeters; F) contour detection of the region. The area inside the contour will now be analysed using a pulse peak analyser s or a single echo detector, and detected pulses will then be joined together to form tracks and be classified. This is not shown in the figure.
Fig. 16 shows an echogram as a result of median pass filtration using different filter dimensions. Dimensions are given as height times width. It can be to seen how long, narrow filters suppress target objects, whilst medium width, short filters suppress noise whilst the useful signal is highlighted.
Fig. 17 illustrates the principle for selection of critical frequencies in the filter, in this case shown only for one dimension.
Basis for development of a new method of analysis After working for some considerable time using existing methods, we were able to make the following important observations:
1. Missing echoes in tracks from fish combined with many noise-based echoes zo cause traditional tracking algorithms to fail.
2. An amp echogram contains more information than the SED echogram. More echoes from fish, pulse width, pulse shape and surroundings.
3. Data in an echogram is more or less horizontally oriented because of the time factor along the x-axis.
zs 4. Angle estimates from multibeam echo sounders are highly unreliable in relation to the distance estimate.
Point 1 indicates that we must reduce noise and increase the number of echoes in tracks from fish in order to be able to improve the method.
3o Point 2 indicates that we should use the amplitude echogram rather than the SED
echogram.
Point 3 indicates that we can make use of certain properties in the echogram for the detection target objects.
Point 4 indicates that we should focus on distance estimates when combining echoes.
On the basis of these observations, we have developed a new method of analysis based on the processing of the amplitude echogram. We make use of elements from the field of image processing to analyse the amplitude information. First, noise is removed by using a low-pass filter. Then we separate regions containing echoes from solid objects from regions containing background noise. With the aid of shape analysis we then remove regions where there is a low probability of their containing target objects. The s remaining regions are now those where there is a high probability of their containing echoes from one or more target objects. Individual objects are now separated out using a region analysis and the separated objects are then classified as fish, drifting objects and the like on the basis of properties that are calculated during the process.
io ~ Collection of sonar data. The same as for the traditional method.
~ Low-pass filtration. To remove noise in amplitude echogram.
~ Segmentation. To distinguish foreground from background and to find key regions in the echogram.
~ Shape analysis. To remove regions where there is a low probability is of their being a useful signal.
~ Area analysis To find the number of objects in one and the same region. E.g., individual fish in a small shoal.
~ Classification To classify individual objects such as fish, drifting objects etc.
zs The method can be used for single-beam and multi-beam echo sounders. (Single-beam, dual-beam and split-beam). The use of mufti-beam echo sounders, where relative position in the beam can be calculated, provides a greater basis for making the object analysis, but otherwise the analysis is the same for all types of sounders.
Elements that affect the method are distance and time resolution in the amplitude echogram. Le., the number of emitted sound pulses (pings) per second and the number of samples per metre. This affects the choice of filter dimensions that are to be used. In our tests we have mainly used about 5 shots per second and 9 cm per sample.
With 3o EY500 from Simrad we can obtain as much as 3 cm depth resolution per sample. This has been used in some tests, but results in slow data processing and vast amounts of data.
More detailed description of steps in the new method Recording and registration of sonar data Recording and registration of sonar data is done in the traditional manner in that ultrasonic signals in the volume that is to be measured are repeatedly emitted, and that-echoes from reflecting objects and ultrasonic signals in the volume being measured are s received, amplified and registered. Upon reception of the signals, the distance to the source of the echo can also be calculated as an alternative to the time interval between the emission time of the ultrasonic signal and the reception time of the echo signal.
Upon reception and registration, the signals are digitised. The registration is preferably made in a two-dimensional data presentation, where distance, alternatively the echo io time, constitutes one dimension, and the sequence number of the measurement, alternatively the time of measurement, constitutes the other dimension. In the case of single-beam sonars, data that is stored in the elements in this two-dimensional presentation will only comprise the intensity of the received signal, but in the case of other sonar types it will also include indications of direction of the source of the is received signal.
Low-pass filtration Low-pass filtration is important for two reasons. The filtration removes noise and it reduces the problems of missing echoes in tracks.
The amplitude in an echogram can vary greatly from ping to ping and from sample to sample. Rapid variations are often associated with noise, and we remove this noise by filtering. A second but important reason is that we wish to fill in the missing echoes in tracks from target objects. The selection of filters with dimensions that cover several zs pings will ensure that a little of the echo energy from pings before and after the missing echo will be transferred to the missing echo.
The filter dimensions are important for a number of reasons. Because of the time aspect in the echograms, tracks are more or less horizontally oriented. Le., vertically narrow so and horizontally broad. By using filters that are broad and narrow, we will therefore be able to highlight and smooth useful tracks at the same time as we reduce background noise. Filters with the opposite orientation will reduce both useful signals and sound.
(See filter examples in figs.) The best result is achieved using filter dimensions that are somewhat narrower and shorter than the tracks from the useful signals.
Segmentation The next step in the process is the segmentation. Segmentation involves separating signals from background noise. Several techniques are common in the field of image s processing, but we have found that segmentation based on thresholding works best for sonar data. Because echograms often do not have a constant background level, we cannot use a constant threshold value. The echo intensity varies both with time and with distance from the transducer and we have therefore developed a special adaptive thresholding method (adaptable thresholding).
io New thresholding method developed for segmentation of echograms As described in the section on filtration, the filter dimension is important.
A two-dimensional filter has a dimension along two normal axes in a Cartesian coordinate system. The size of the filter is described by the dimensioning or the number of cells is along the two normal axes that are referred to. The filter dimension 3 x S
will thus mean that we have a two-dimensional filter with three elements along one of the axes of dimension (y axis) and five elements along the other axis of dimension (x axis). In the case of digital sonar data, the x axis will measure increasing ping number, whilst the y axis will measure sample number for echoes received after each individual ping. One zo dimension will highlight useful tracks, whilst another dimension will reduce the tracks.
A foreground filter is composed of a low-pass filter that removes high-frequency noise without removing the useful signal that exists at lower frequencies. A
background filter removes both high-frequency noise and useful signals, but nevertheless maintains the "background intensity" that consists of unwanted signals at frequencies below the is frequencies of the useful signal, and any unwanted objects. By allowing two different filters to process the echograms, one that reduces the useful tracks and one that highlights them, we obtain two different result echograms. The filtration of the two-dimensional echogram using a two-dimensional filter is done by convolution. In a median filter all the filter coefficients will be alike. The break frequency or the filter 3o frequency is determined by the dimension of the filter. When the filter dimension (number of elements/size) increases, the critical frequency of the filter will decrease.
Two-dimensional filters have two critical frequencies. A filter consisting of more elements along the y axis than along the x axis will have a lower critical frequency along the y axis than along the x axis.
On subtracting the two echograms that are produced after filtration using the two filters, we will be left with one differential echogram. Subtraction in this context entails using a quantity operator, so that a difference between objects in two images, or two two-dimensional tables, is produced. Because it is primarily the useful tracks that are different, it is these that will remain after the subtraction, whilst the background noise has been eliminated.
In practice we use the following method for all x, y positions in the echograms, represented by a two-dimensional data presentation or image matrix for a sampled echogram. Here, x = ping number, y = sample number and it is the echo amplitude that is compared.
io If Foreground-filtered echogram 1 [x,y] > (Background-filtered echogram 2 [x,y]+6dB) then Threshold echogram [x,y]: = Original echogram [x,y]
or else Threshold echogram [x, y]: = 0.
In the example above the background is lifted by 6dB in order to provide a 6dB
is distinction between signal and noise. In special cases it may be necessary to use other values.
In the threshold method simple rules are applied for the choice of filter dimensions:
a) the foreground filter must not be larger than the target object;
zo b) the background filter must be larger than the target object.
As the size of the target objects depends upon the set-up (ping rate, sample rate, emitted pulse length, choice of recording site and location of recording site), there is no guarantee that pre-defined filter sizes will always give optimum segmentation.
The size of the target objects and optimum filter dimensions are now found by a manual method.
zs In this method a small, but representative set of the objects that are to be detected or removed is chosen manually from the collected echograms. The size of the foreground filter is adjusted so that width and height are less than or equal to width and height of the smallest of these objects. The size of the background filter is chosen so that it is larger than the largest of the desired objects. The term "larger than" in this case is used 3o to mean higher or wider, or higher and wider. For echograms from horizontally placed echo sounders in a shallow river, we have obtained particularly good results when we have used a foreground filter of 1 x 3 (height x width) and a background filter of 1 x 99.
The echogram resolution was then 9 cm per sample and 5 pings per second. These values are used in the counting test shown in Table 1 and Table 2.
Taking as a basis sonar data collected using a transducer with an aperture angle of 10°, a shot rate of 3-6 pings per second, emitted sonar pulse length of 0.3 ms and a sample distance of 9 cm, typical filter sizes could be: foreground filter: height - 1 to 5 samples, width - 1 to 11 pings; background filter: height - 1 to 55 samples, width - 1 to 99 pings.
As mentioned above, a two-dimensional filter has two critical frequencies, one along s each coordinate axis. This provides plenty of scope for finding filter sizes that give critical frequencies capable of maintaining background and unwanted echo signals at the same time as it is able to remove useful signals. In the example from the river Tana, which is also illustrated in Figure 15, a 1 x 99 filter is used as background filter. This filter removes echoes from fish because it is considerably wider than the fish echo io tracks. At the same time, background and unwanted echoes from stones and other stationary objects are kept because these echo tracks are considerably shorter and wider than the height and width of the background filter.
Post-processing in order to reduce problems with split regions is The preliminary filtration reduces the possibilities of missing echoes in tracks from target objects. There will nevertheless be some splitting up of regions found by thresholding owing to erratic echo intensity. To join together such regions we have used morphological methods and filters. We have found growing particularly suitable for this purpose. By allowing all detected regions to grow or spread out so that the area Zo increases, small cracks between separated regions will close. Split regions thus merge together. We have also found a similar effect by using different low-pass filters.
Because of the more or less horizontal orientation of the useful tracks, we have found that a larger growth in the horizontal direction than in the vertical direction is favourable (e.g., growth 3 pings horizontally and 1 sample vertically outwards from all side edges Zs in the detected regions).
Shape analysis We have now found a set of regions which may contain echoes from target objects.
Besides the target objects, we may still also have regions containing noise and regions so containing echoes from unwanted objects such as the bottom, boats, wakes or the like.
By studying the appearance of the detected regions, we are able to say a great deal about the origin of the regions and thus remove regions where there is a low probability of their containing echoes from target objects. Many properties of a region's shape can be calculated; e.g., area, height/width ratio, centre of gravity, branches, rotation and ss frequency ranges for the outline. For horizontal sonar recordings using a stationary echo sounder in a river, four main types of regions are found. These are: a) regions containing residual background noise; b) regions containing echoes from moving objects such as fish and drifting objects; c) regions containing echoes from stationary objects such as stones on the bottom; and d) regions containing echoes from the wake of boats. These differ greatly from one another as regards many of the said parameters, and when fish are of interest we can in a relatively easy manner remove regions of the s type a, c and d. In this way we increase the probability of the remaining regions containing fish or drifting objects.
In our test we used the perimeter of the objects measured in a plurality of edge samples.
By means of the echogram resolution used, we set an upper limit of 500 and a lower io limit of 20. Regions of less than 20 were considered as group a, whilst regions of more than 500 were taken as belonging to groups c and d.
Region analysis for the purpose of finding the number of objects After the aforementioned steps have been carried out, we are left with a small number of is regions where there is a high probability of their containing echoes for target objects.
However, these regions may contain echoes from more than one object, and therefore it is necessary to separate out individual objects from these regions. We find examples of such multi-object regions when two or more salmon swim close together in a shoal. It is also possible that a region empty of fish has "survived" until this stage.
zo Many of the same properties found by shape analysis can also be used in this analysis.
Regions containing a number of fish will often be wider and have a more branched contour than regions containing echoes from a single object.
is To date, we have only tested methods that evaluate pulse peaks, single echoes and the number of branches within the detected regions. A region with a high percentage of single-peak pulses will in all probability contain echoes from one individual object, whilst regions having a large percentage of two or multi-peaked pulses will contain several objects. By using a contour detector to find the contour of each individual so region and then using a pulse peak detector on the echogram within each contour, we produce a plurality of dense, narrow tracks consisting of a series of pulse peaks. These are then joined together to give tracks by using a traditional tracking algorithm within each detected contour.
ss For each region:
Low-pass filtration Pulse peak detection to form dense tracks suitable for tracking Single echo detection to find the angular position and measurement intensity correction tracking.
Joining together echoes close in time and space to find a number of objects/tracks.
In cases where there are many objects in a region, an automatic analysis may cause problems owing to interference and masking effects between objects. The object closest to the echo sounder can more or less shield objects located behind it. An alternative strategy would then be to store the position of such multiple object regions and then io evaluate them manually afterwards.
Owing to the high degree of variation in the amplitude signal, pulse peak detection should be combined with a low-pass filter that is vertically oriented. Le., that it has a greater extension in the distance domain that in the time domain. This reduces the is number of pulse peaks from random objects. For tracks that are not horizontally oriented (changes distance to the sonar), pulse peak detection should normally be carried out on the track and not on the basis of single pings. In the case of certain types of data, pulse peak detection has been found to give too many peaks within a detected region. One possible solution may be to draw in criteria associated with single echo zo detection such as pulse duration.
Pulse peak detection gives denser tracks and therefore a better basis for joining echoes to give tracks than single echo detection. In a river we will often find tracks from fish where not a single echo is accepted by the single echo detector. Nevertheless we obtain zs excellent tracks with the aid of pulse peak detection. When using the traditional method these tracks would be overlooked.
However, single echo detection is necessary in order to reduce unreliability as regards size estimation and angular position. We combine these two techniques in our method.
3o First we use pulse peak detection to obtain dense tracks suitable for tracking. A single echo detection is then carried out for each of the pulse peaks in order to find angular position and correct amplitude. In addition to angular position, the deviation of the pulse peaks from the "ideal" single echo criteria is noted. This deviation is used as a character for each echo, and we can later choose the most "reliable" echoes when the ss echo intensity, position and velocity of the track is to be evaluated. This is shown in Figure 6. Figure 7 shows what pulse peaks in a region detected by image processing may look like.
Classification of objects Depending on the recording site and what objects are to be studied, we can now make a final classification of the detected tracks. The classification involves an evaluation of s the objects as regards one or more calculated properties in order to determine the object type. In addition to the properties that are found by calculation in shape and region analysis, other data relating to the objects, such as, for example, data indicating the movement pattern of the objects in the volume covered by the echo sounder or sonar, will be evaluated against criteria of a similar type during the classification. All ~o properties that are calculated in the shape and region analysis can be used to determine the origin of the detected individual objects. In rivers in particular, the direction of travel and velocity in relation to water current will be important criteria.
Two examples of systems to provide information about the movement pattern of the is objects are given below:
Single beam: Calculation of time/distance angle to find velocity and direction;
Split beam: Analysis of angles of the pulse peaks to find velocity and direction.
2o If the echo sounder is slightly angled relative to the objects that pass by, the objects that pass one way will give tracks at positive angles, whilst objects that pass the other way will give tracks at negative angles. By calculating the angles of the tracks we can thus determine both velocity and direction. This method can be used for all types of echo sounders.
For split beam echo sounders, we have angle estimates for all echoes and can therefore find out how an object has moved through the beam irrespective of whether the sounder is angled or not. However, the angle measurement is likely to be more unreliable than the distance measurement, both because of the way in which the echo sounder detects 3o the angles and because of physical conditions around one of the sonar recordings. In vertical measurements from a boat the echo sounder will be apt to change angle so that passing fish will "jump and dance" around in the beam. In a river the echo sound will be affected by the turbulence of the water at the same time as the actual echo sounder will vibrate in the water current.
In the pulse peak analysis we found peaks for each ping within a detected region. (High track quality, no missing echoes.) However, many of the angle measurements for these peaks will be unreliable. For one object we therefore calculate the mean value and standard deviation for the closest angle samples and reject the result if unreliability is low. In this way, we find a trajectory in both time and space for most tracks.
Velocity and direction can be easily calculated, and this can in turn be used to distinguish, e.g., s fish swimming upstream in a river from a twig drifting downstream.
Test of the method In this test we have first used sonar recordings made using a horizontally positioned split-beam echo sounder in the river Tana near Polmak, Norway, in the summer of 1999 io and in the Norwegian river Numedalslagen in 1996. The equipment used was Simrad's EY500 with a 120kHz 4x10 degree transducer of the type ES 120-4. The echogram resolution was chosen at 9 cm/ sample 5 pings per second. We have also used files from vertical recordings.
is The data from the river Tana had a vertical resolution of 9 cm/sample, whilst the resolution in the other data was 20 cm/sample. In the horizontal plane the resolution varied from 3 to 6 pings per second.
Fish were first counted manually, then with a traditional SED/tracking algorithm and 20 lastly using our new method. The results from the counts were compared.
For the tests we first chose sonar files containing echoes from fish. These were then studied closely and analysed manually to find the number of fish. The manually detected fish tracks were then analysed to find parameters for a traditional neighbour is counting routine. This requires a figure for the highest number of missing echoes and the smallest number of echoes in a single track in order to work. The figures 11 and 7 were used respectively, and the algorithm was set to analyse the selected data.
Subsequently, our method of analysis was set as follows: First, the original echograms 3o were processed using a median filter having a height = 1 and a width = 3.
Then the original echograms were filtered using a similar median filter of the same height, but width = 99. The intensity in the result from this filtration was then reduced using 6dB
and used to threshold the result from the first filtration. Then regions with perimeters shorter than 20 and perimeters longer than 1000 were removed. Next, regions having 3s perimeters longer than 20 and shorter than 500 were extended by 3 samples horizontally and 1 sample vertically. The regions that still had a perimeter of between 20 and 500 were then "region-analysed", and individual objects found here were then classified as fish or drifting objects on the basis of direction of travel.
The results from the two automatic methods were then compared with the result from s the manual method. For horizontal recordings our method obtained an accuracy of 93%
before manual sorting and 97% after manual sorting compared with the manual results.
The SED/neighbour counting method achieved 2% accuracy. For vertical recordings with better signal/noise ratio, out method resulted in an accuracy of 98%
against 63%
for the SED/neighbourhood method.
Sonar Duration TransducerRecordingManual Image SED and file in angle site counting counting neighbour_ name mins. counting method 06242218.dg640 Hor. RiverNumedal,26 28 360 Norwa 07191222.d25 Hor. RiverNorwa 7 8 408 9 , Tana 07191248.d25 Hor. RiverNorwa 1 2 384 9 , Tans 07191313.d25 Hor. RiverNorwa 7 6 406 9 , Tans 07191339.d25 Hor. RiverNorway, 2 3 400 9 Tans 07191405.d25 Hor. RiverNorwa 5 6 400 9 , Tana 07191430.d25 Hor. RiverNorwa 4 4 356 9 , Tans 07191456.d25 Hor. RiverNorwa 8 8 361 9 , Tana 07191548.d25 Hor. RiverNorwa 2 3 391 9 , Tana 07191614.d25 Hor. RiverNorwa I I 13 413 9 , Tana 07191639.d25 Hor. RiverNorwa 9 9 368 9 , Tans 07191705.d25 Hor. RiverNorwa 9 11 323 9 , Tana 07191731.dg925 Hor. RiverNorway, 3 5 352 Tana 07191756.d25 Hor. RiverNorwa 10 10 323 9 , Tana 07191822.d25 Hor. RiverNorwa 12 11 352 9 , Tana 07191848.d25 Hor. RiverNorwa 7 9 346 9 , Tans 07191913.d25 Hor. RiverNorwa 4 4 366 9 , Tans 07191939.d25 Hor. RiverNorwa 14 12 344 , Tana Total 465 141 152 6653 Overall 100% 93% 2%
accurac Table 1: Counting results for horizontally positioned echo sounder in a shallow river Sonar Duration TransducerRecordingManual Image SED and file in name mins. angle site counting counting neighbour_ counting method 08141313.dg911 Vert. Czech 7 8 37 water Re ublic 09292220.dg72 Vert. Annecy, 50 50 53 water France Total 13 57 58 90 Overall 100% 98% 63%
accurac Table 2: Counting results for vertically positioned echo sounder
zs 4. Angle estimates from multibeam echo sounders are highly unreliable in relation to the distance estimate.
Point 1 indicates that we must reduce noise and increase the number of echoes in tracks from fish in order to be able to improve the method.
3o Point 2 indicates that we should use the amplitude echogram rather than the SED
echogram.
Point 3 indicates that we can make use of certain properties in the echogram for the detection target objects.
Point 4 indicates that we should focus on distance estimates when combining echoes.
On the basis of these observations, we have developed a new method of analysis based on the processing of the amplitude echogram. We make use of elements from the field of image processing to analyse the amplitude information. First, noise is removed by using a low-pass filter. Then we separate regions containing echoes from solid objects from regions containing background noise. With the aid of shape analysis we then remove regions where there is a low probability of their containing target objects. The s remaining regions are now those where there is a high probability of their containing echoes from one or more target objects. Individual objects are now separated out using a region analysis and the separated objects are then classified as fish, drifting objects and the like on the basis of properties that are calculated during the process.
io ~ Collection of sonar data. The same as for the traditional method.
~ Low-pass filtration. To remove noise in amplitude echogram.
~ Segmentation. To distinguish foreground from background and to find key regions in the echogram.
~ Shape analysis. To remove regions where there is a low probability is of their being a useful signal.
~ Area analysis To find the number of objects in one and the same region. E.g., individual fish in a small shoal.
~ Classification To classify individual objects such as fish, drifting objects etc.
zs The method can be used for single-beam and multi-beam echo sounders. (Single-beam, dual-beam and split-beam). The use of mufti-beam echo sounders, where relative position in the beam can be calculated, provides a greater basis for making the object analysis, but otherwise the analysis is the same for all types of sounders.
Elements that affect the method are distance and time resolution in the amplitude echogram. Le., the number of emitted sound pulses (pings) per second and the number of samples per metre. This affects the choice of filter dimensions that are to be used. In our tests we have mainly used about 5 shots per second and 9 cm per sample.
With 3o EY500 from Simrad we can obtain as much as 3 cm depth resolution per sample. This has been used in some tests, but results in slow data processing and vast amounts of data.
More detailed description of steps in the new method Recording and registration of sonar data Recording and registration of sonar data is done in the traditional manner in that ultrasonic signals in the volume that is to be measured are repeatedly emitted, and that-echoes from reflecting objects and ultrasonic signals in the volume being measured are s received, amplified and registered. Upon reception of the signals, the distance to the source of the echo can also be calculated as an alternative to the time interval between the emission time of the ultrasonic signal and the reception time of the echo signal.
Upon reception and registration, the signals are digitised. The registration is preferably made in a two-dimensional data presentation, where distance, alternatively the echo io time, constitutes one dimension, and the sequence number of the measurement, alternatively the time of measurement, constitutes the other dimension. In the case of single-beam sonars, data that is stored in the elements in this two-dimensional presentation will only comprise the intensity of the received signal, but in the case of other sonar types it will also include indications of direction of the source of the is received signal.
Low-pass filtration Low-pass filtration is important for two reasons. The filtration removes noise and it reduces the problems of missing echoes in tracks.
The amplitude in an echogram can vary greatly from ping to ping and from sample to sample. Rapid variations are often associated with noise, and we remove this noise by filtering. A second but important reason is that we wish to fill in the missing echoes in tracks from target objects. The selection of filters with dimensions that cover several zs pings will ensure that a little of the echo energy from pings before and after the missing echo will be transferred to the missing echo.
The filter dimensions are important for a number of reasons. Because of the time aspect in the echograms, tracks are more or less horizontally oriented. Le., vertically narrow so and horizontally broad. By using filters that are broad and narrow, we will therefore be able to highlight and smooth useful tracks at the same time as we reduce background noise. Filters with the opposite orientation will reduce both useful signals and sound.
(See filter examples in figs.) The best result is achieved using filter dimensions that are somewhat narrower and shorter than the tracks from the useful signals.
Segmentation The next step in the process is the segmentation. Segmentation involves separating signals from background noise. Several techniques are common in the field of image s processing, but we have found that segmentation based on thresholding works best for sonar data. Because echograms often do not have a constant background level, we cannot use a constant threshold value. The echo intensity varies both with time and with distance from the transducer and we have therefore developed a special adaptive thresholding method (adaptable thresholding).
io New thresholding method developed for segmentation of echograms As described in the section on filtration, the filter dimension is important.
A two-dimensional filter has a dimension along two normal axes in a Cartesian coordinate system. The size of the filter is described by the dimensioning or the number of cells is along the two normal axes that are referred to. The filter dimension 3 x S
will thus mean that we have a two-dimensional filter with three elements along one of the axes of dimension (y axis) and five elements along the other axis of dimension (x axis). In the case of digital sonar data, the x axis will measure increasing ping number, whilst the y axis will measure sample number for echoes received after each individual ping. One zo dimension will highlight useful tracks, whilst another dimension will reduce the tracks.
A foreground filter is composed of a low-pass filter that removes high-frequency noise without removing the useful signal that exists at lower frequencies. A
background filter removes both high-frequency noise and useful signals, but nevertheless maintains the "background intensity" that consists of unwanted signals at frequencies below the is frequencies of the useful signal, and any unwanted objects. By allowing two different filters to process the echograms, one that reduces the useful tracks and one that highlights them, we obtain two different result echograms. The filtration of the two-dimensional echogram using a two-dimensional filter is done by convolution. In a median filter all the filter coefficients will be alike. The break frequency or the filter 3o frequency is determined by the dimension of the filter. When the filter dimension (number of elements/size) increases, the critical frequency of the filter will decrease.
Two-dimensional filters have two critical frequencies. A filter consisting of more elements along the y axis than along the x axis will have a lower critical frequency along the y axis than along the x axis.
On subtracting the two echograms that are produced after filtration using the two filters, we will be left with one differential echogram. Subtraction in this context entails using a quantity operator, so that a difference between objects in two images, or two two-dimensional tables, is produced. Because it is primarily the useful tracks that are different, it is these that will remain after the subtraction, whilst the background noise has been eliminated.
In practice we use the following method for all x, y positions in the echograms, represented by a two-dimensional data presentation or image matrix for a sampled echogram. Here, x = ping number, y = sample number and it is the echo amplitude that is compared.
io If Foreground-filtered echogram 1 [x,y] > (Background-filtered echogram 2 [x,y]+6dB) then Threshold echogram [x,y]: = Original echogram [x,y]
or else Threshold echogram [x, y]: = 0.
In the example above the background is lifted by 6dB in order to provide a 6dB
is distinction between signal and noise. In special cases it may be necessary to use other values.
In the threshold method simple rules are applied for the choice of filter dimensions:
a) the foreground filter must not be larger than the target object;
zo b) the background filter must be larger than the target object.
As the size of the target objects depends upon the set-up (ping rate, sample rate, emitted pulse length, choice of recording site and location of recording site), there is no guarantee that pre-defined filter sizes will always give optimum segmentation.
The size of the target objects and optimum filter dimensions are now found by a manual method.
zs In this method a small, but representative set of the objects that are to be detected or removed is chosen manually from the collected echograms. The size of the foreground filter is adjusted so that width and height are less than or equal to width and height of the smallest of these objects. The size of the background filter is chosen so that it is larger than the largest of the desired objects. The term "larger than" in this case is used 3o to mean higher or wider, or higher and wider. For echograms from horizontally placed echo sounders in a shallow river, we have obtained particularly good results when we have used a foreground filter of 1 x 3 (height x width) and a background filter of 1 x 99.
The echogram resolution was then 9 cm per sample and 5 pings per second. These values are used in the counting test shown in Table 1 and Table 2.
Taking as a basis sonar data collected using a transducer with an aperture angle of 10°, a shot rate of 3-6 pings per second, emitted sonar pulse length of 0.3 ms and a sample distance of 9 cm, typical filter sizes could be: foreground filter: height - 1 to 5 samples, width - 1 to 11 pings; background filter: height - 1 to 55 samples, width - 1 to 99 pings.
As mentioned above, a two-dimensional filter has two critical frequencies, one along s each coordinate axis. This provides plenty of scope for finding filter sizes that give critical frequencies capable of maintaining background and unwanted echo signals at the same time as it is able to remove useful signals. In the example from the river Tana, which is also illustrated in Figure 15, a 1 x 99 filter is used as background filter. This filter removes echoes from fish because it is considerably wider than the fish echo io tracks. At the same time, background and unwanted echoes from stones and other stationary objects are kept because these echo tracks are considerably shorter and wider than the height and width of the background filter.
Post-processing in order to reduce problems with split regions is The preliminary filtration reduces the possibilities of missing echoes in tracks from target objects. There will nevertheless be some splitting up of regions found by thresholding owing to erratic echo intensity. To join together such regions we have used morphological methods and filters. We have found growing particularly suitable for this purpose. By allowing all detected regions to grow or spread out so that the area Zo increases, small cracks between separated regions will close. Split regions thus merge together. We have also found a similar effect by using different low-pass filters.
Because of the more or less horizontal orientation of the useful tracks, we have found that a larger growth in the horizontal direction than in the vertical direction is favourable (e.g., growth 3 pings horizontally and 1 sample vertically outwards from all side edges Zs in the detected regions).
Shape analysis We have now found a set of regions which may contain echoes from target objects.
Besides the target objects, we may still also have regions containing noise and regions so containing echoes from unwanted objects such as the bottom, boats, wakes or the like.
By studying the appearance of the detected regions, we are able to say a great deal about the origin of the regions and thus remove regions where there is a low probability of their containing echoes from target objects. Many properties of a region's shape can be calculated; e.g., area, height/width ratio, centre of gravity, branches, rotation and ss frequency ranges for the outline. For horizontal sonar recordings using a stationary echo sounder in a river, four main types of regions are found. These are: a) regions containing residual background noise; b) regions containing echoes from moving objects such as fish and drifting objects; c) regions containing echoes from stationary objects such as stones on the bottom; and d) regions containing echoes from the wake of boats. These differ greatly from one another as regards many of the said parameters, and when fish are of interest we can in a relatively easy manner remove regions of the s type a, c and d. In this way we increase the probability of the remaining regions containing fish or drifting objects.
In our test we used the perimeter of the objects measured in a plurality of edge samples.
By means of the echogram resolution used, we set an upper limit of 500 and a lower io limit of 20. Regions of less than 20 were considered as group a, whilst regions of more than 500 were taken as belonging to groups c and d.
Region analysis for the purpose of finding the number of objects After the aforementioned steps have been carried out, we are left with a small number of is regions where there is a high probability of their containing echoes for target objects.
However, these regions may contain echoes from more than one object, and therefore it is necessary to separate out individual objects from these regions. We find examples of such multi-object regions when two or more salmon swim close together in a shoal. It is also possible that a region empty of fish has "survived" until this stage.
zo Many of the same properties found by shape analysis can also be used in this analysis.
Regions containing a number of fish will often be wider and have a more branched contour than regions containing echoes from a single object.
is To date, we have only tested methods that evaluate pulse peaks, single echoes and the number of branches within the detected regions. A region with a high percentage of single-peak pulses will in all probability contain echoes from one individual object, whilst regions having a large percentage of two or multi-peaked pulses will contain several objects. By using a contour detector to find the contour of each individual so region and then using a pulse peak detector on the echogram within each contour, we produce a plurality of dense, narrow tracks consisting of a series of pulse peaks. These are then joined together to give tracks by using a traditional tracking algorithm within each detected contour.
ss For each region:
Low-pass filtration Pulse peak detection to form dense tracks suitable for tracking Single echo detection to find the angular position and measurement intensity correction tracking.
Joining together echoes close in time and space to find a number of objects/tracks.
In cases where there are many objects in a region, an automatic analysis may cause problems owing to interference and masking effects between objects. The object closest to the echo sounder can more or less shield objects located behind it. An alternative strategy would then be to store the position of such multiple object regions and then io evaluate them manually afterwards.
Owing to the high degree of variation in the amplitude signal, pulse peak detection should be combined with a low-pass filter that is vertically oriented. Le., that it has a greater extension in the distance domain that in the time domain. This reduces the is number of pulse peaks from random objects. For tracks that are not horizontally oriented (changes distance to the sonar), pulse peak detection should normally be carried out on the track and not on the basis of single pings. In the case of certain types of data, pulse peak detection has been found to give too many peaks within a detected region. One possible solution may be to draw in criteria associated with single echo zo detection such as pulse duration.
Pulse peak detection gives denser tracks and therefore a better basis for joining echoes to give tracks than single echo detection. In a river we will often find tracks from fish where not a single echo is accepted by the single echo detector. Nevertheless we obtain zs excellent tracks with the aid of pulse peak detection. When using the traditional method these tracks would be overlooked.
However, single echo detection is necessary in order to reduce unreliability as regards size estimation and angular position. We combine these two techniques in our method.
3o First we use pulse peak detection to obtain dense tracks suitable for tracking. A single echo detection is then carried out for each of the pulse peaks in order to find angular position and correct amplitude. In addition to angular position, the deviation of the pulse peaks from the "ideal" single echo criteria is noted. This deviation is used as a character for each echo, and we can later choose the most "reliable" echoes when the ss echo intensity, position and velocity of the track is to be evaluated. This is shown in Figure 6. Figure 7 shows what pulse peaks in a region detected by image processing may look like.
Classification of objects Depending on the recording site and what objects are to be studied, we can now make a final classification of the detected tracks. The classification involves an evaluation of s the objects as regards one or more calculated properties in order to determine the object type. In addition to the properties that are found by calculation in shape and region analysis, other data relating to the objects, such as, for example, data indicating the movement pattern of the objects in the volume covered by the echo sounder or sonar, will be evaluated against criteria of a similar type during the classification. All ~o properties that are calculated in the shape and region analysis can be used to determine the origin of the detected individual objects. In rivers in particular, the direction of travel and velocity in relation to water current will be important criteria.
Two examples of systems to provide information about the movement pattern of the is objects are given below:
Single beam: Calculation of time/distance angle to find velocity and direction;
Split beam: Analysis of angles of the pulse peaks to find velocity and direction.
2o If the echo sounder is slightly angled relative to the objects that pass by, the objects that pass one way will give tracks at positive angles, whilst objects that pass the other way will give tracks at negative angles. By calculating the angles of the tracks we can thus determine both velocity and direction. This method can be used for all types of echo sounders.
For split beam echo sounders, we have angle estimates for all echoes and can therefore find out how an object has moved through the beam irrespective of whether the sounder is angled or not. However, the angle measurement is likely to be more unreliable than the distance measurement, both because of the way in which the echo sounder detects 3o the angles and because of physical conditions around one of the sonar recordings. In vertical measurements from a boat the echo sounder will be apt to change angle so that passing fish will "jump and dance" around in the beam. In a river the echo sound will be affected by the turbulence of the water at the same time as the actual echo sounder will vibrate in the water current.
In the pulse peak analysis we found peaks for each ping within a detected region. (High track quality, no missing echoes.) However, many of the angle measurements for these peaks will be unreliable. For one object we therefore calculate the mean value and standard deviation for the closest angle samples and reject the result if unreliability is low. In this way, we find a trajectory in both time and space for most tracks.
Velocity and direction can be easily calculated, and this can in turn be used to distinguish, e.g., s fish swimming upstream in a river from a twig drifting downstream.
Test of the method In this test we have first used sonar recordings made using a horizontally positioned split-beam echo sounder in the river Tana near Polmak, Norway, in the summer of 1999 io and in the Norwegian river Numedalslagen in 1996. The equipment used was Simrad's EY500 with a 120kHz 4x10 degree transducer of the type ES 120-4. The echogram resolution was chosen at 9 cm/ sample 5 pings per second. We have also used files from vertical recordings.
is The data from the river Tana had a vertical resolution of 9 cm/sample, whilst the resolution in the other data was 20 cm/sample. In the horizontal plane the resolution varied from 3 to 6 pings per second.
Fish were first counted manually, then with a traditional SED/tracking algorithm and 20 lastly using our new method. The results from the counts were compared.
For the tests we first chose sonar files containing echoes from fish. These were then studied closely and analysed manually to find the number of fish. The manually detected fish tracks were then analysed to find parameters for a traditional neighbour is counting routine. This requires a figure for the highest number of missing echoes and the smallest number of echoes in a single track in order to work. The figures 11 and 7 were used respectively, and the algorithm was set to analyse the selected data.
Subsequently, our method of analysis was set as follows: First, the original echograms 3o were processed using a median filter having a height = 1 and a width = 3.
Then the original echograms were filtered using a similar median filter of the same height, but width = 99. The intensity in the result from this filtration was then reduced using 6dB
and used to threshold the result from the first filtration. Then regions with perimeters shorter than 20 and perimeters longer than 1000 were removed. Next, regions having 3s perimeters longer than 20 and shorter than 500 were extended by 3 samples horizontally and 1 sample vertically. The regions that still had a perimeter of between 20 and 500 were then "region-analysed", and individual objects found here were then classified as fish or drifting objects on the basis of direction of travel.
The results from the two automatic methods were then compared with the result from s the manual method. For horizontal recordings our method obtained an accuracy of 93%
before manual sorting and 97% after manual sorting compared with the manual results.
The SED/neighbour counting method achieved 2% accuracy. For vertical recordings with better signal/noise ratio, out method resulted in an accuracy of 98%
against 63%
for the SED/neighbourhood method.
Sonar Duration TransducerRecordingManual Image SED and file in angle site counting counting neighbour_ name mins. counting method 06242218.dg640 Hor. RiverNumedal,26 28 360 Norwa 07191222.d25 Hor. RiverNorwa 7 8 408 9 , Tana 07191248.d25 Hor. RiverNorwa 1 2 384 9 , Tans 07191313.d25 Hor. RiverNorwa 7 6 406 9 , Tans 07191339.d25 Hor. RiverNorway, 2 3 400 9 Tans 07191405.d25 Hor. RiverNorwa 5 6 400 9 , Tana 07191430.d25 Hor. RiverNorwa 4 4 356 9 , Tans 07191456.d25 Hor. RiverNorwa 8 8 361 9 , Tana 07191548.d25 Hor. RiverNorwa 2 3 391 9 , Tana 07191614.d25 Hor. RiverNorwa I I 13 413 9 , Tana 07191639.d25 Hor. RiverNorwa 9 9 368 9 , Tans 07191705.d25 Hor. RiverNorwa 9 11 323 9 , Tana 07191731.dg925 Hor. RiverNorway, 3 5 352 Tana 07191756.d25 Hor. RiverNorwa 10 10 323 9 , Tana 07191822.d25 Hor. RiverNorwa 12 11 352 9 , Tana 07191848.d25 Hor. RiverNorwa 7 9 346 9 , Tans 07191913.d25 Hor. RiverNorwa 4 4 366 9 , Tans 07191939.d25 Hor. RiverNorwa 14 12 344 , Tana Total 465 141 152 6653 Overall 100% 93% 2%
accurac Table 1: Counting results for horizontally positioned echo sounder in a shallow river Sonar Duration TransducerRecordingManual Image SED and file in name mins. angle site counting counting neighbour_ counting method 08141313.dg911 Vert. Czech 7 8 37 water Re ublic 09292220.dg72 Vert. Annecy, 50 50 53 water France Total 13 57 58 90 Overall 100% 98% 63%
accurac Table 2: Counting results for vertically positioned echo sounder
Claims
claims 1.
A method using ultrasonic signals and ultrasonic signal echo measurements for detecting fish or fish like objects in a river, watercourse, lake or the sea, from a two-dimensional raw echogram recorded by means of an ultrasonic apparatus such as an echo sounder or sonar apparatus, which raw echogram is represented by a two-dimensional array of raw echo data elements, such as a raw image matrix, where distance, alternatively the duration of the ultrasonic signal, constitutes the first dimension, and time, alternatively the sequence number of the echo measurement if measurement is repeated at certain intervals, constitutes the second dimension, and the individual data element at least comprises an echo signal intensity sample, the method comprising a step of segmenting the echogram using adaptive thresholding for reducing noise signals in the echogram relative to echo signals from fish, characterised in that the adaptive thresholding includes the steps of:
producing a foreground-filtered echogram by filtering the raw echogram with regard to echogram sample value using a first filter dimensioned to highlight useful echo tracks;
producing a background-filtered echogram by filtering the raw echogram with regard to echogram sample value using a second filter dimensioned to reduce useful echo tracks;
and producing a thresholded echogram by setting the echo sample value of one element in the thresholded echogram at the same value as that of the corresponding element in the raw echogram if the corresponding element in the foreground-filtered echogram is larger than the corresponding element in the background-filtered echogram with a constant addition of a predetermined value and otherwise setting the element at zero.
2.
A method as disclosed in claim 1, characterised in that it further comprises:
low-pass filtering the echogram with regard to the echo signal intensity samples, prior to the segmentation step, using a one or two-dimensional low-pass filter.
3.
A method as disclosed in claim 2, characterised in that it further comprises:
filling in, in the segmented echogram, small segment spaces between segments with the aid of contour detection and morphological processing for merging adjacent segments to form one segment.
4.
A method as disclosed in claim 3, characterised in that it further comprises:
shape-analysing each segment by evaluating the segment's area, perimeter, height/width ratio, centre of gravity, branches, rotation and/or frequency range of the segment outline.
5.
A method as disclosed in claim 2, 3 or 4, characterised in that the value of the constant addition is about 6db relative to the respective element in the background-filtered echogram.
6.
A method as disclosed in any one of the previous claims, characterised in that the dimension of the first filter is in the range of 1 to 5 sample units in the first dimension and in the range of 1 to 11 measurement units in the second dimension.
7.
A method as disclosed in any one of the previous claims, characterised in that the dimension of the second filter is in the range of 1 to 55 sample units in the first dimension and in the range of 99 measurement units in the second dimension.
8.
A method as disclosed in claim 1, characterised in that it further includes the step of:
carrying out an area analysis of the segmented echogram by means of pulse peak detection in order to form dense tracks suitable for tracking and/or single echo detection in order to find angular position and measurement intensity correction tracking.
9.
A method as disclosed in claim 1, characterised in that it further includes the step of:
carrying out classification of the segments in the segmented echogram on the basis of calculated and/or registered properties of segments and/or segment elements, whereby segments are classified as fish in motion, stationary elements such as the riverbed or riverbed plants, other floating objects such as twigs or bubbles and so forth.
10.
A method as disclosed in claim 1, characterised in that the low-pass filter is a median filter.
11.
A method as disclosed in any one of the previous claims, characterised in that the first filter is a median filter.
12.
A method as disclosed in any one of the previous claims, characterised in that the second filter is a median filter.
13.
A method for detecting fish or fish like objects in a river, watercourse, lake or the sea, using ultrasonic signals and ultrasonic signal echo measurements, the method comprising:
obtaining a two-dimensional echogram recorded by means of an ultrasonic apparatus such as an echo sounder or sonar apparatus, and segmenting the two-dimensional echogram using adaptive thresholding to cosiderably reduce noise signals in the echogram relative to echo signals from the fish or the fish like objects, characterised in that the two-dimensional echogram is a raw echo two-dimensional echogram represented by a two-dimensional array, such as an image matrix, of raw echo data elements, where distance, alternatively duration of the ultrasonic signal echo, constitutes the first dimension, and time, alternatively the sequence number of the echo measurement if echo measurement is repeated at defined intervals, constitutes the second dimension, and the individual raw echo data element at least comprises an echo signal intensity sample.
14.
A method as disclosed in claim 13, characterised in that the adaptive thresholding includes the steps of:
producing a foreground-filtered echogram by filtering the raw echogram with regard to echogram sample value using a first filter dimensioned to highlight useful echo tracks;
producing a background-filtered echogram by filtering the raw echogram with regard to echogram sample value using a second filter dimensioned to reduce useful echo tracks;
producing a thresholded echogram by setting the echo sample value of one element in the thresholded echogram at the same value as that of the corresponding element in the raw echogram if the corresponding element in the foreground-filtered echogram is larger than the corresponding element in the background-filtered echogram with a constant addition of a predetermined value and otherwise setting the element at zero.
A method using ultrasonic signals and ultrasonic signal echo measurements for detecting fish or fish like objects in a river, watercourse, lake or the sea, from a two-dimensional raw echogram recorded by means of an ultrasonic apparatus such as an echo sounder or sonar apparatus, which raw echogram is represented by a two-dimensional array of raw echo data elements, such as a raw image matrix, where distance, alternatively the duration of the ultrasonic signal, constitutes the first dimension, and time, alternatively the sequence number of the echo measurement if measurement is repeated at certain intervals, constitutes the second dimension, and the individual data element at least comprises an echo signal intensity sample, the method comprising a step of segmenting the echogram using adaptive thresholding for reducing noise signals in the echogram relative to echo signals from fish, characterised in that the adaptive thresholding includes the steps of:
producing a foreground-filtered echogram by filtering the raw echogram with regard to echogram sample value using a first filter dimensioned to highlight useful echo tracks;
producing a background-filtered echogram by filtering the raw echogram with regard to echogram sample value using a second filter dimensioned to reduce useful echo tracks;
and producing a thresholded echogram by setting the echo sample value of one element in the thresholded echogram at the same value as that of the corresponding element in the raw echogram if the corresponding element in the foreground-filtered echogram is larger than the corresponding element in the background-filtered echogram with a constant addition of a predetermined value and otherwise setting the element at zero.
2.
A method as disclosed in claim 1, characterised in that it further comprises:
low-pass filtering the echogram with regard to the echo signal intensity samples, prior to the segmentation step, using a one or two-dimensional low-pass filter.
3.
A method as disclosed in claim 2, characterised in that it further comprises:
filling in, in the segmented echogram, small segment spaces between segments with the aid of contour detection and morphological processing for merging adjacent segments to form one segment.
4.
A method as disclosed in claim 3, characterised in that it further comprises:
shape-analysing each segment by evaluating the segment's area, perimeter, height/width ratio, centre of gravity, branches, rotation and/or frequency range of the segment outline.
5.
A method as disclosed in claim 2, 3 or 4, characterised in that the value of the constant addition is about 6db relative to the respective element in the background-filtered echogram.
6.
A method as disclosed in any one of the previous claims, characterised in that the dimension of the first filter is in the range of 1 to 5 sample units in the first dimension and in the range of 1 to 11 measurement units in the second dimension.
7.
A method as disclosed in any one of the previous claims, characterised in that the dimension of the second filter is in the range of 1 to 55 sample units in the first dimension and in the range of 99 measurement units in the second dimension.
8.
A method as disclosed in claim 1, characterised in that it further includes the step of:
carrying out an area analysis of the segmented echogram by means of pulse peak detection in order to form dense tracks suitable for tracking and/or single echo detection in order to find angular position and measurement intensity correction tracking.
9.
A method as disclosed in claim 1, characterised in that it further includes the step of:
carrying out classification of the segments in the segmented echogram on the basis of calculated and/or registered properties of segments and/or segment elements, whereby segments are classified as fish in motion, stationary elements such as the riverbed or riverbed plants, other floating objects such as twigs or bubbles and so forth.
10.
A method as disclosed in claim 1, characterised in that the low-pass filter is a median filter.
11.
A method as disclosed in any one of the previous claims, characterised in that the first filter is a median filter.
12.
A method as disclosed in any one of the previous claims, characterised in that the second filter is a median filter.
13.
A method for detecting fish or fish like objects in a river, watercourse, lake or the sea, using ultrasonic signals and ultrasonic signal echo measurements, the method comprising:
obtaining a two-dimensional echogram recorded by means of an ultrasonic apparatus such as an echo sounder or sonar apparatus, and segmenting the two-dimensional echogram using adaptive thresholding to cosiderably reduce noise signals in the echogram relative to echo signals from the fish or the fish like objects, characterised in that the two-dimensional echogram is a raw echo two-dimensional echogram represented by a two-dimensional array, such as an image matrix, of raw echo data elements, where distance, alternatively duration of the ultrasonic signal echo, constitutes the first dimension, and time, alternatively the sequence number of the echo measurement if echo measurement is repeated at defined intervals, constitutes the second dimension, and the individual raw echo data element at least comprises an echo signal intensity sample.
14.
A method as disclosed in claim 13, characterised in that the adaptive thresholding includes the steps of:
producing a foreground-filtered echogram by filtering the raw echogram with regard to echogram sample value using a first filter dimensioned to highlight useful echo tracks;
producing a background-filtered echogram by filtering the raw echogram with regard to echogram sample value using a second filter dimensioned to reduce useful echo tracks;
producing a thresholded echogram by setting the echo sample value of one element in the thresholded echogram at the same value as that of the corresponding element in the raw echogram if the corresponding element in the foreground-filtered echogram is larger than the corresponding element in the background-filtered echogram with a constant addition of a predetermined value and otherwise setting the element at zero.
Applications Claiming Priority (5)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| NO994327A NO994327D0 (en) | 1999-09-06 | 1999-09-06 | Improved fish detection probability in split beam sonar data |
| NO19994327 | 1999-09-06 | ||
| NO20003543 | 2000-07-10 | ||
| NO20003543A NO20003543L (en) | 1999-09-06 | 2000-07-10 | Procedure for fish detection from sonar data |
| PCT/NO2000/000288 WO2001018562A1 (en) | 1999-09-06 | 2000-09-05 | Fish detection method using sonar data |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CA2383760A1 true CA2383760A1 (en) | 2001-03-15 |
Family
ID=26648997
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CA002383760A Abandoned CA2383760A1 (en) | 1999-09-06 | 2000-09-05 | Fish detection method using sonar data |
Country Status (5)
| Country | Link |
|---|---|
| EP (1) | EP1210618A1 (en) |
| AU (1) | AU7043200A (en) |
| CA (1) | CA2383760A1 (en) |
| NO (1) | NO20003543L (en) |
| WO (1) | WO2001018562A1 (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2093546A1 (en) * | 2008-02-22 | 2009-08-26 | Siemens Milltronics Process Instruments Inc. | A method for locating a wanted echo from unwanted echoes in a time-of-flight level measurement system |
| US8164983B2 (en) | 2009-03-06 | 2012-04-24 | Johnson David A | Fish finder |
| DE102012007979A1 (en) * | 2012-04-24 | 2013-10-24 | Krohne Messtechnik Gmbh | Method for determining the level of a medium and corresponding device |
| GB2523561B (en) * | 2014-02-27 | 2016-03-02 | Sonardyne Internat Ltd | Underwater Environment Reference Map Enhancement and Intruder Detection |
| CN106019263B (en) * | 2016-07-13 | 2018-03-20 | 东南大学 | Target radial speed measuring method based on more bright spot echo models |
| CN113484867B (en) * | 2021-06-25 | 2023-10-20 | 山东航天电子技术研究所 | Method for detecting density of fish shoal in closed space based on imaging sonar |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US4074264A (en) * | 1976-01-07 | 1978-02-14 | Hughes Aircraft Company | Adaptive threshold clutter processor |
| US5930200A (en) * | 1998-05-08 | 1999-07-27 | Garmin Corporation | Depth sounder with object identification feature |
-
2000
- 2000-07-10 NO NO20003543A patent/NO20003543L/en unknown
- 2000-09-05 AU AU70432/00A patent/AU7043200A/en not_active Abandoned
- 2000-09-05 WO PCT/NO2000/000288 patent/WO2001018562A1/en not_active Application Discontinuation
- 2000-09-05 CA CA002383760A patent/CA2383760A1/en not_active Abandoned
- 2000-09-05 EP EP00959043A patent/EP1210618A1/en not_active Withdrawn
Also Published As
| Publication number | Publication date |
|---|---|
| EP1210618A1 (en) | 2002-06-05 |
| NO20003543L (en) | 2001-03-07 |
| NO20003543D0 (en) | 2000-07-10 |
| AU7043200A (en) | 2001-04-10 |
| WO2001018562A1 (en) | 2001-03-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP3573783B2 (en) | Sonar system | |
| Balk et al. | Improved fish detection in data from split-beam sonar | |
| CN102692627B (en) | For processing instrument and the method for Doppler measurement signal | |
| Kimura et al. | Comparison of stationary acoustic monitoring and visual observation of finless porpoises | |
| Trenkel et al. | Overview of recent progress in fisheries acoustics made by Ifremer with examples from the Bay of Biscay | |
| CN111077536B (en) | Marine laser radar sounding echo classification method and device | |
| CN111077537B (en) | Sounding echo optimization classification method and device | |
| Marques | Automatic mid-water target detection using multibeam water column | |
| EP1609001B1 (en) | Signal-processing method and active sonar implementing same | |
| CA2383760A1 (en) | Fish detection method using sonar data | |
| Atallah et al. | Wavelet analysis of bathymetric sidescan sonar data for the classification of seafloor sediments in Hopvågen Bay-Norway | |
| Yang et al. | Multibeam sonar bottom detection using multiple subarrays | |
| XU et al. | 3D-view generation and species classification of aquatic plants using acoustic images | |
| US7483336B2 (en) | Device for avoiding obstacles for high-speed multi-hulled watercraft | |
| Brehmer et al. | Adaptation of fisheries sonar for monitoring schools of large pelagic fish: dependence of schooling behaviour on fish finding efficiency | |
| Prager et al. | Bottom classification: operational results from QTC view | |
| Liu et al. | A method of sidelobe effect suppression for multibeam water column images based on an adaptive soft threshold | |
| Guenther et al. | Obstruction detection and data decimation for airborne laser hydrography | |
| Stanton | Density estimates of biological sound scatterers using sonar echo peak PDFs | |
| Tao et al. | Species identification of Chinese sturgeon using acoustic descriptors and ascertaining their spatial distribution in the spawning ground of Gezhouba Dam | |
| McClatchie et al. | Target strength of southern blue whiting (Micromesistius australis) using swimbladder modelling, split beam and deconvolution | |
| Kirsch et al. | Acoustic estimates of zooplankton distributions in Prince William Sound, spring 1996 | |
| Lee et al. | Application of acoustic Doppler current profiler combined with a scientific echo sounder for krill Euphausia pacifica density estimation | |
| Qi et al. | A real-time adaptive bottom tracking method for bathymetric side-scan sonar | |
| WO2003093868A1 (en) | Compensation of sonar image data primarily for seabed classification |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| FZDE | Discontinued |