CN106682736A - Image identification method and apparatus - Google Patents
Image identification method and apparatus Download PDFInfo
- Publication number
- CN106682736A CN106682736A CN201710034836.7A CN201710034836A CN106682736A CN 106682736 A CN106682736 A CN 106682736A CN 201710034836 A CN201710034836 A CN 201710034836A CN 106682736 A CN106682736 A CN 106682736A
- Authority
- CN
- China
- Prior art keywords
- convolutional
- convolution kernel
- convolutional layer
- group
- neural networks
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses an image identification method and apparatus. The image identification method includes the steps: inputting an image to be identified into a trained first convolution neural network which includes P groups of convolution layers, wherein each group of convolution layers includes two convolution layers; the sizes of the convolution cores of the two convolution layers are respectively a 1*N first convolution core and an N*1 second convolution core; by means of each group of convolution layers in the first convolution neural network, extracting the characteristics of the image to be identified and obtaining the characteristics, of each group of convolution layers, to be extracted; and according to the characteristics, of each group of convolution layers, to be extracted, determining the identifying result of the image to be identified. The technical scheme of the image identification method and apparatus solves the problem that in the related technologies, a square convolution core causes high computational complexity and high computing cost.
Description
Technical field
It relates to convolution technique field, more particularly to a kind of image-recognizing method and device.
Background technology
Convolutional neural networks (Convolutional Neural Network, CNN) are a kind of efficient image recognition sides
Method.In CNN, image is obtained through a series of convolutional layer, active coating, pond layer, the feature extraction of full articulamentum and process
The recognition result of images to be recognized.In correlation technique, the Computing Principle of convolutional layer is with sliding window using a square convolution kernel
The form of mouth is slided on the rectangular area of input, often slides into a new position, just calculates convolution kernel and the value of the position
Product, the amount of calculation of square convolution kernel is O (N2), amount of calculation is calculated relatively costly than larger.
The content of the invention
To overcome problem present in correlation technique, the embodiment of the present disclosure to provide a kind of image-recognizing method and device, use
To reduce the amount of calculation of convolution, reduce calculating cost.
According to the first aspect of the embodiment of the present disclosure, there is provided a kind of image-recognizing method, it may include：
The first convolutional neural networks that images to be recognized input has been trained, first convolutional neural networks include P groups
Convolutional layer, each group of convolutional layer includes two convolutional layers, and the convolution kernel of described two convolutional layers is respectively size is 1*N first
Convolution kernel and the second convolution kernel that size is N*1；
Feature extraction is carried out to the images to be recognized by each group of convolutional layer of first convolutional neural networks, is obtained
To the feature to be extracted of each group of convolutional layer；
The recognition result of the images to be recognized is determined according to the feature to be extracted of each group of convolutional layer.
In one embodiment, method also includes：
Each convolutional layer in second convolutional neural networks is decomposed into into described two convolutional layers, the first volume is obtained
Product neutral net, each convolutional layer in second convolutional neural networks includes a square convolution kernel.
In one embodiment, the convolution kernel of each convolutional layer in the second convolutional neural networks is the side that size is N*N
Shape convolution kernel, second convolutional neural networks include P convolutional layer；
Described each convolutional layer by the second convolutional neural networks is decomposed into described two convolutional layers, including：
Square convolution kernel of the size of each convolutional layer in second convolutional neural networks for N*N is decomposed into into size
For 1*N the first convolution kernel and size for N*1 the second convolution kernel, the product of first convolution kernel and second convolution kernel
It is worth for the square convolution kernel；
Correspondence in second convolutional neural networks is replaced respectively using first convolution kernel and second convolution kernel
Convolutional layer square convolution kernel, obtain one group of convolutional layer of first convolutional neural networks.
In one embodiment, the images to be recognized is entered by each group of convolutional layer of first convolutional neural networks
Row feature extraction, including：
Spy is carried out to input information using a convolutional layer in each group of convolutional layer of first convolutional neural networks
Extraction is levied, the intermediate value of the feature to be extracted of each group of convolutional layer is obtained；
Another convolutional layer that the intermediate value of the feature to be extracted is input in each group of convolutional layer is carried out into feature
Extract, obtain the feature to be extracted of each group of convolutional layer.
According to the second aspect of the embodiment of the present disclosure, there is provided a kind of pattern recognition device, it may include：
Input module, is configured to the first convolutional neural networks for having trained images to be recognized input, the first volume
Product neutral net includes P group convolutional layers, and each group of convolutional layer includes two convolutional layers, and the convolution kernel of described two convolutional layers is distinguished
For the second convolution kernel that the first convolution kernel and size that size is 1*N are N*1；
Characteristic extracting module, each group of convolutional layer for being configured to first convolutional neural networks is waited to know to described
Other image carries out feature extraction, obtains the feature to be extracted of each group of convolutional layer；
Determining module, is configured to determine the images to be recognized according to the feature to be extracted of each group of convolutional layer
Recognition result.
In one embodiment, device also includes：
Decomposing module, is configured to for each convolutional layer in the second convolutional neural networks to be decomposed into described two convolution
Layer, obtains first convolutional neural networks, and each convolutional layer in second convolutional neural networks is square including one
Convolution kernel.
In one embodiment, the convolution kernel of each convolutional layer in the second convolutional neural networks is the side that size is N*N
Shape convolution kernel, second convolutional neural networks include P convolutional layer；
The decomposing module includes：
Decompose submodule, be configured to the size of each convolutional layer in second convolutional neural networks as N*N's
Square convolution kernel be decomposed into size be 1*N the first convolution kernel and size for N*1 the second convolution kernel, first convolution kernel with
The product value of second convolution kernel is the square convolution kernel；
Determination sub-module, is configured with first convolution kernel and second convolution kernel replaces respectively described second
The square convolution kernel of corresponding convolutional layer in convolutional neural networks, obtains one group of convolutional layer of first convolutional neural networks.
In one embodiment, characteristic extracting module includes：
First extracting sub-module, be configured to, with each group of convolutional layer of first convolutional neural networks
Convolutional layer carries out feature extraction to input information, obtains the intermediate value of the feature to be extracted of each group of convolutional layer；
Second extracting sub-module, is configured to that the intermediate value of the feature to be extracted is input in each group of convolutional layer
Another convolutional layer carry out feature extraction, obtain the feature to be extracted of each group of convolutional layer.
According to the third aspect of the embodiment of the present disclosure, there is provided a kind of pattern recognition device, including：
Processor；
For storing the memorizer of processor executable；
Wherein, the processor is configured to：
The first convolutional neural networks that images to be recognized input has been trained, first convolutional neural networks include P groups
Convolutional layer, each group of convolutional layer includes two convolutional layers, and the convolution kernel of described two convolutional layers is respectively size is 1*N first
Convolution kernel and the second convolution kernel that size is N*1；
Feature extraction is carried out to the images to be recognized by each group of convolutional layer of first convolutional neural networks, is obtained
To the feature to be extracted of each group of convolutional layer；
The recognition result of the images to be recognized is determined according to the feature to be extracted of each group of convolutional layer.
The technical scheme that embodiment of the disclosure is provided can include following beneficial effect：It is every in first convolutional neural networks
The convolution kernel of two convolutional layers included by one group of convolutional layer is that respectively size is N*1 for first convolution kernel and size of 1*N
The second convolution kernel, the convolution kernel amount of calculation of each group of convolutional layer is 2*O (N), in convolution kernel than in the case of larger, the disclosure
In each group of convolutional layer amount of calculation 2*O (N) will be far below correlation technique in square convolution kernel amount of calculation O (N2), solve
Using computationally intensive, calculating high cost problem caused by square convolution kernel institute in correlation technique.
By the way that the square convolution kernel of each convolutional layer in the second convolutional neural networks is split as into two orthogonal convolution
Core, you can obtain the first convolutional neural networks, because the product of two orthogonal convolution kernels is corresponding square convolution kernel, because
This, using the first convolutional neural networks in each group of convolutional layer to the feature extraction effect of images to be recognized and the second convolution god
The feature extraction effect of the square property convolution kernel of each corresponding convolutional layer is identical in Jing networks, therefore the disclosure is realized
Convolutional calculation amount is reduced on the basis of convolution effect identical, calculating cost is reduced.
By directly using the fisrt feature figure of the extraction of a convolutional layer in each group of convolutional layer as another convolution
The input of layer, it is possible to achieve the feature extraction effect of an each group of convolutional layer convolution corresponding with the second convolutional neural networks
The feature extraction effect of layer is identical.
It should be appreciated that the general description of the above and detailed description hereinafter are only exemplary and explanatory, not
The disclosure can be limited.
Description of the drawings
Accompanying drawing herein is merged in description and constitutes the part of this specification, shows the enforcement for meeting the present invention
Example, and be used to explain the principle of the present invention together with description.
Fig. 1 is the flow chart of the image-recognizing method according to an exemplary embodiment.
Fig. 2 is to obtain the first convolution nerve net by the second convolutional neural networks according to an exemplary embodiment one
The flow chart of network.
Fig. 3 A are each group of convolutional layers pair by the first convolutional neural networks according to an exemplary embodiment two
Images to be recognized carries out the flow chart of feature extraction.
Fig. 3 B are a square convolution kernels pair in convolutional neural networks of use second according to an exemplary embodiment
Input information carries out the schematic diagram of convolution.
Fig. 3 C are the square of corresponding diagram 3B in the convolutional neural networks of use first according to an exemplary embodiment
One convolution kernel of convolution kernel carries out the schematic diagram of convolution to input information.
Fig. 3 D are the square of corresponding diagram 3B in the convolutional neural networks of use first according to an exemplary embodiment
Another convolution kernel of convolution kernel carries out the schematic diagram of convolution to the output information of Fig. 3 C.
Fig. 4 is a kind of block diagram of the pattern recognition device according to an exemplary embodiment.
Fig. 5 is the block diagram of another kind of pattern recognition device according to an exemplary embodiment.
Fig. 6 is a kind of block diagram suitable for pattern recognition device according to an exemplary embodiment.
Specific embodiment
Here exemplary embodiment will be illustrated in detail, its example is illustrated in the accompanying drawings.Explained below is related to
During accompanying drawing, unless otherwise indicated, the same numbers in different accompanying drawings represent same or analogous key element.Following exemplary embodiment
Described in embodiment do not represent and the consistent all embodiments of the present invention.Conversely, they be only with it is such as appended
The example of the consistent apparatus and method of some aspects described in detail in claims, the present invention.
Fig. 1 is the flow chart of the image-recognizing method according to an exemplary embodiment；The image-recognizing method can be with
Apply in electronic equipment (for example：Smart mobile phone, panel computer, personal computer) on, as shown in figure 1, the image-recognizing method
Comprise the following steps：
In a step 101, the first convolutional neural networks images to be recognized input trained, the first convolutional neural networks
Including P group convolutional layers, each group of convolutional layer includes two convolutional layers, and the convolution kernel of two convolutional layers is respectively size for 1*N's
First convolution kernel and the second convolution kernel that size is N*1.
In one embodiment, the first convolutional neural networks can be obtained by carrying out process to the second convolutional neural networks, can
The first convolutional neural networks are obtained by carrying out decomposition to each convolutional layer in the second convolutional neural networks, Fig. 2 is can be found in
Illustrated embodiment, does not first describe in detail here.
In one embodiment, the convolutional layer in the first convolutional neural networks occurs in groups, two volumes of each group of convolutional layer
The convolution kernel of lamination is the first convolution kernel and size that a size is 1*N for second convolution kernel of N*1.
In one embodiment, the second convolutional neural networks can related training method training be obtained in correlation technique according to
Convolutional neural networks, the convolution kernel in convolutional layer is square convolution kernel, as N*N convolution kernels.
In a step 102, carry out feature to images to be recognized by each group of convolutional layer of the first convolutional neural networks to carry
Take, obtain the feature to be extracted of each group of convolutional layer.
In one embodiment, carry out feature to images to be recognized by each group of convolutional layer of the first convolutional neural networks to carry
The step of taking can be found in Fig. 3 A illustrated embodiments, first not describe in detail here.
In step 103, the recognition result of images to be recognized is determined according to the feature to be extracted of each group of convolutional layer.
In one embodiment, each group of convolutional layer is carried in the way of convolutional neural networks used according to correlation technique
The feature to be extracted for taking is processed, and is identified result, such as using pond layer, full articulamentum etc. to each group of convolutional layer
Feature to be extracted carries out process and is identified result, and I will not elaborate.
In the present embodiment, the convolution kernel point of two convolutional layers in the first convolutional neural networks included by each group of convolutional layer
Not Wei size for 1*N the first convolution kernel and size for N*1 the second convolution kernel, the convolution kernel amount of calculation of each group of convolutional layer is
2*O (N), in convolution kernel than in the case of larger, amount of calculation 2*O (N) of each group of convolutional layer will be far below correlation in the disclosure
Amount of calculation O (the N of the square convolution kernel in technology2), solve in correlation technique using the caused amount of calculation of square convolution kernel institute
Greatly, the problem of high cost is calculated.
In one embodiment, method also includes：
Each convolutional layer in second convolutional neural networks is decomposed into into two convolutional layers, the first convolution nerve net is obtained
Network, each convolutional layer in the second convolutional neural networks includes a square convolution kernel.
In one embodiment, the convolution kernel of each convolutional layer in the second convolutional neural networks is the side that size is N*N
Shape convolution kernel, the second convolutional neural networks include P convolutional layer；
Each convolutional layer in second convolutional neural networks is decomposed into into two convolutional layers, including：
The size of each convolutional layer in the second convolutional neural networks is decomposed into into size for 1* for the square convolution kernel of N*N
The product value of first convolution kernel of N and the second convolution kernel that size is N*1, the first convolution kernel and the second convolution kernel is square convolution
Core；
The side of corresponding convolutional layer in the second convolutional neural networks is replaced respectively using the first convolution kernel and the second convolution kernel
Shape convolution kernel, obtains one group of convolutional layer of the first convolutional neural networks.
In one embodiment, carry out feature to images to be recognized by each group of convolutional layer of the first convolutional neural networks to carry
Take, including：
Feature is carried out using a convolutional layer in each group of convolutional layer of the first convolutional neural networks to input information to carry
Take, obtain the intermediate value of the feature to be extracted of each group of convolutional layer；
Another convolutional layer that the intermediate value of feature to be extracted is input in each group of convolutional layer is carried out into feature extraction, is obtained
The feature to be extracted of each group of convolutional layer.
Specifically how image recognition is carried out, refer to subsequent embodiment.
Illustrate the technical scheme that the embodiment of the present disclosure is provided with a specific embodiment below.
Fig. 2 is the flow chart of the image-recognizing method according to an exemplary embodiment one；The present embodiment utilizes this public affairs
The said method of embodiment offer is provided, how to be decomposed into according to each convolutional layer of the second convolutional neural networks with electronic equipment
It is illustrative as a example by two convolutional layers, as shown in Fig. 2 comprising the steps：
In step 201, the square convolution kernel by the size of each convolutional layer in the second convolutional neural networks for N*N divides
Solve the product for the second convolution kernel that first convolution kernel and size of 1*N are N*1, the first convolution kernel and the second convolution kernel for size
It is worth for square convolution kernel.
In one embodiment, it is illustrated how to decompose convolutional layer so that square convolution kernel is for 3*3 convolution kernels as an example.Example
Such as, square convolution kernel isFirst convolution kernel [110] and a volume Two for 3*1 of a 1*3 can be split as
Product coreIn the present embodiment, the first convolution kernel is row vector, and the second convolution kernel is column vector.
In one embodiment, by the way that each square convolution kernel is split as into the first convolution kernel and the second convolution kernel, afterwards
One convolutional layer is split as one group of convolutional layer by realization.
In step 202., correspondence in the second convolutional neural networks is replaced respectively using the first convolution kernel and the second convolution kernel
Convolutional layer square convolution kernel, obtain one group of convolutional layer of the first convolutional neural networks.
In the present embodiment, by the way that the square convolution kernel of each convolutional layer in the second convolutional neural networks is split as into two
Individual orthogonal convolution kernel, you can obtain the first convolutional neural networks, due to the product of two orthogonal convolution kernels be it is corresponding square
Convolution kernel, therefore, using the first convolutional neural networks in each group of convolutional layer to the feature extraction effect of images to be recognized with
The feature extraction effect of the square property convolution kernel of each corresponding convolutional layer is identical in second convolutional neural networks, therefore
The disclosure is realized and reduces convolutional calculation amount on the basis of convolution effect identical, reduces calculating cost.
Fig. 3 A are each group of convolutional layers pair by the first convolutional neural networks according to an exemplary embodiment two
Images to be recognized carries out the flow chart of feature extraction, and Fig. 3 B are the convolutional Neurals of use second according to an exemplary embodiment
Square convolution kernel carries out the schematic diagram of convolution to input information in network, and Fig. 3 C are illustrated according to an exemplary embodiment
Convolution is carried out to input information using a convolution kernel of the square convolution kernel of corresponding diagram 3B in the first convolutional neural networks
Schematic diagram, Fig. 3 D are the square volumes of corresponding diagram 3B in the convolutional neural networks of use first according to an exemplary embodiment
Another convolution kernel of product core carries out the schematic diagram of convolution to the output information of Fig. 3 C；The present embodiment is carried using the embodiment of the present disclosure
For said method, with electronic equipment how by each group of convolutional layer images to be recognized is carried out feature extraction carry out it is exemplary
Illustrate, as shown in Figure 3A, comprise the steps：
In step 301, using a convolutional layer in each group of convolutional layer of the first convolutional neural networks to input letter
Breath carries out feature extraction, obtains the intermediate value of the feature to be extracted of each group of convolutional layer.
In step 302, another convolutional layer that the intermediate value of feature to be extracted is input in each group of convolutional layer is carried out
Feature extraction, obtains the feature to be extracted of each group of convolutional layer.
In one embodiment, referring to Fig. 3 B, the images to be recognized 31 with input information 31 as 5*5 here, by volume Two
Source convolution kernel in product neutral net, that is, be numbered 36 square convolution kernel and carry out feature extraction and obtain being numbered 35 feature, the
It is [110] that square convolution kernel 36 can be decomposed into the first convolution kernel 32 in two convolutional neural networks, and the second convolution kernel 34 is
In one embodiment, in step 301, referring to Fig. 3 C, input information 31 is input in the first convolutional neural networks
One group of convolutional layer a convolutional layer, process of convolution is carried out by the first convolution kernel 32, obtain the to be extracted of this group of convolutional layer
The intermediate value of feature, namely it is numbered 33 feature；In step 302, referring to Fig. 3 D, by the way that the feature for being numbered 33 is input into
Another convolutional layer in first convolutional neural networks, and process of convolution is carried out using the second convolution kernel 34, obtain being numbered 35
Feature.By Fig. 3 B- Fig. 3 D, the disclosure is using the convolution kernel of a 1*N and the convolution kernel of a N*1 to figure to be identified
As the effect for carrying out the effect of convolution with square convolution kernel used in correlation technique carries out convolution it is identical.
In the present embodiment, by directly using the fisrt feature figure of the extraction of a convolutional layer in each group of convolutional layer as
The input of another convolutional layer, it is possible to achieve the feature extraction effect of each group of convolutional layer is corresponding with the second convolutional neural networks
A convolutional layer feature extraction effect it is identical.
Corresponding with the embodiment of aforementioned image-recognizing method, the disclosure additionally provides pattern recognition device and its is applied
Electronic equipment embodiment.
Fig. 4 is a kind of block diagram of the pattern recognition device according to an exemplary embodiment, as shown in figure 4, image is known
Other device includes：
Input module 410, is configured to the first convolutional neural networks for having trained images to be recognized input, the first convolution
Neutral net includes P group convolutional layers, and each group of convolutional layer includes two convolutional layers, and the convolution kernel of two convolutional layers is respectively size
For 1*N the first convolution kernel and size for N*1 the second convolution kernel；
Characteristic extracting module 420, is configured to each group of convolutional layer of the first convolutional neural networks to figure to be identified
As carrying out feature extraction, the feature to be extracted of each group of convolutional layer is obtained；
Determining module 430, is configured to determine the identification of images to be recognized according to the feature to be extracted of each group of convolutional layer
As a result.
Fig. 5 is the block diagram of another kind of pattern recognition device according to an exemplary embodiment, as shown in figure 5, upper
On the basis of stating embodiment illustrated in fig. 4, in one embodiment, device also includes：
Decomposing module 440, is configured to for each convolutional layer in the second convolutional neural networks to be decomposed into two convolution
Layer, obtains the first convolutional neural networks, and each convolutional layer in the second convolutional neural networks includes a square convolution kernel.
In one embodiment, the convolution kernel of each convolutional layer in the second convolutional neural networks is the side that size is N*N
Shape convolution kernel, the second convolutional neural networks include P convolutional layer；
Decomposing module 440 includes：
Decompose submodule 441, be configured to by the size of each convolutional layer in the second convolutional neural networks for N*N side
Shape convolution kernel is decomposed into the second convolution kernel that the first convolution kernel and size that size is 1*N is N*1, the first convolution kernel and volume Two
The product value of product core is square convolution kernel；
Determination sub-module 442, is configured with the first convolution kernel and the second convolution kernel replaces respectively the second convolutional Neural
The square convolution kernel of corresponding convolutional layer in network, obtains one group of convolutional layer of the first convolutional neural networks.
In one embodiment, characteristic extracting module 420 includes：
First extracting sub-module 421, be configured to, with each group of convolutional layer of the first convolutional neural networks
Convolutional layer carries out feature extraction to input information, obtains the intermediate value of the feature to be extracted of each group of convolutional layer；
Second extracting sub-module 422, is configured to the intermediate value of feature just to be extracted and is input in each group of convolutional layer
Another convolutional layer carries out feature extraction, obtains the feature to be extracted of each group of convolutional layer.
The function of unit and effect realizes that process specifically refers in said method correspondence step in said apparatus
Process is realized, be will not be described here.
For device embodiment, because it corresponds essentially to embodiment of the method, so related part is referring to method reality
Apply the part explanation of example.Device embodiment described above is only schematic, wherein illustrating as separating component
Unit can be or may not be physically separate, can be as the part that unit shows or may not be
Physical location, you can be located at a place, or can also be distributed on multiple NEs.Can be according to the actual needs
Select some or all of module therein to realize the purpose of disclosure scheme.Those of ordinary skill in the art are not paying wound
In the case that the property made is worked, you can to understand and implement.
Fig. 6 is a kind of block diagram suitable for pattern recognition device according to an exemplary embodiment.For example, device
600 can be electronic equipment, such as panel computer, smart mobile phone etc..
With reference to Fig. 6, device 600 can include following one or more assemblies：Process assembly 602, memorizer 604, power supply
Component 606, multimedia groupware 608, audio-frequency assembly 610, the interface 612 of input/output (I/O), sensor cluster 614, and
Communication component 616.
The integrated operation of the usual control device 600 of process assembly 602, such as with display, call, data communication, phase
Machine operates and records the associated operation of operation.Treatment element 602 can refer to including one or more processors 620 to perform
Order, to complete all or part of step of above-mentioned method.Additionally, process assembly 602 can include one or more modules, just
Interaction between process assembly 602 and other assemblies.For example, processing component 602 can include multi-media module, many to facilitate
Interaction between media component 608 and process assembly 602.
Memorizer 604 is configured to store various types of data to support the operation in equipment 600.These data are shown
Example includes the instruction of any application program for operating on device 600 or method, message, picture etc..Memorizer 604 can be with
Realized by any kind of volatibility or non-volatile memory device or combinations thereof, such as static RAM
(SRAM), Electrically Erasable Read Only Memory (EEPROM), Erasable Programmable Read Only Memory EPROM (EPROM) may be programmed
Read only memory (PROM), read only memory (ROM), magnetic memory, flash memory, disk or CD.
Power supply module 606 provides electric power for the various assemblies of device 600.Electric power assembly 606 can include power management system
System, one or more power supplys, and other generate, manage and distribute the component that electric power is associated with for device 600.
Multimedia groupware 608 is included in the screen of one output interface of offer between device 600 and user.In some realities
In applying example, screen can include liquid crystal display (LCD) and touch panel (TP).If screen includes touch panel, screen can
To be implemented as touch screen, to receive the input signal from user.Touch panel include one or more touch sensors with
Gesture on sensing touch, slip and touch panel.Touch sensor can not only sensing touch or sliding action border, and
And also detection and touch or slide related persistent period and pressure.In certain embodiments, multimedia groupware 608 includes
One front-facing camera and/or post-positioned pick-up head.When equipment 600 is in operator scheme, such as screening-mode or during video mode is front
Putting photographic head and/or post-positioned pick-up head can receive the multi-medium data of outside.Each front-facing camera and post-positioned pick-up head can
Being a fixed optical lens system or with focusing and optical zoom capabilities.
Audio-frequency assembly 610 is configured to output and/or input audio signal.For example, audio-frequency assembly 610 includes a Mike
Wind (MIC), when device 600 is in operator scheme, such as call model, logging mode and speech recognition mode, mike is matched somebody with somebody
It is set to reception external audio signal.The audio signal for being received can be further stored in memorizer 604 or via communication set
Part 616 sends.In certain embodiments, audio-frequency assembly 610 also includes a speaker, for exports audio signal.
, to provide interface between process assembly 602 and peripheral interface module, above-mentioned peripheral interface module can for I/O interfaces 612
To be keyboard, click wheel, button etc..These buttons may include but be not limited to：Home button, volume button, start button and lock
Determine button.
Sensor cluster 614 includes one or more sensors, and the state for providing various aspects for device 600 is commented
Estimate.For example, sensor cluster 614 can detect the opening/closed mode of equipment 600, such as relative localization of component, component
For the display and keypad of device 600, sensor cluster 614 can be with the position of 600 1 components of detection means 600 or device
Change is put, user is presence or absence of with what device 600 was contacted, the temperature of the orientation of device 600 or acceleration/deceleration and device 600
Change.Sensor cluster 614 can include proximity transducer, be configured to when without any physical contact detect near
The presence of object.Sensor cluster 614 can also include optical sensor, such as CMOS or ccd image sensor, for answering in imaging
With used in.In certain embodiments, the sensor cluster 614 can also include acceleration transducer, gyro sensor, magnetic
Sensor, distance-sensor, pressure transducer or temperature sensor.
Communication component 616 is configured to facilitate the communication of wired or wireless way between device 600 and other equipment.Device
600 can access based on the wireless network of communication standard, such as WIFI, 2G or 3G, or combinations thereof.In an exemplary enforcement
In example, communication component 616 receives the broadcast singal or broadcast related information from external broadcasting management system via broadcast channel.
In one exemplary embodiment, communication component 616 also includes near-field communication (NFC) module, to promote junction service.For example,
RF identification (RFID) technology, Infrared Data Association (IrDA) technology, ultra broadband (UWB) technology, bluetooth can be based in NFC module
(BT) technology and other technologies are realizing.
In the exemplary embodiment, device 600 can be by one or more application specific integrated circuits (ASIC), numeral letter
Number processor (DSP), digital signal processing appts (DSPD), PLD (PLD), field programmable gate array
(FPGA), controller, microcontroller, microprocessor or other electronic components realizations, for performing said method, including：
The first convolutional neural networks that images to be recognized input has been trained, the first convolutional neural networks include P group convolution
Layer, each group of convolutional layer includes two convolutional layers, the convolution kernel of two convolutional layers be respectively the first convolution kernel that size is 1*N and
Size is second convolution kernel of N*1；
Feature extraction is carried out to images to be recognized by each group of convolutional layer of the first convolutional neural networks, each group is obtained
The feature to be extracted of convolutional layer；
The recognition result of images to be recognized is determined according to the feature to be extracted of each group of convolutional layer.
In the exemplary embodiment, a kind of non-transitorycomputer readable storage medium including instruction, example are additionally provided
Such as include the memorizer 604 of instruction, above-mentioned instruction can be performed to complete said method by the processor 620 of device 600.For example,
Non-transitorycomputer readable storage medium can be ROM, random access memory (RAM), CD-ROM, tape, floppy disk and light
Data storage device etc..
Those skilled in the art will readily occur to its of the disclosure after considering description and putting into practice disclosure disclosed herein
Its embodiment.The application is intended to any modification, purposes or the adaptations of the disclosure, these modifications, purposes or
Person's adaptations follow the general principle of the disclosure and including the undocumented common knowledge in the art of the disclosure
Or conventional techniques.Description and embodiments are considered only as exemplary, and the true scope of the disclosure and spirit are by following
Claim is pointed out.
It should be appreciated that the disclosure is not limited to the precision architecture for being described above and being shown in the drawings, and
And can without departing from the scope carry out various modifications and changes.The scope of the present disclosure is only limited by appended claim.
Claims (9)
1. a kind of image-recognizing method, it is characterised in that methods described includes：
The first convolutional neural networks that images to be recognized input has been trained, first convolutional neural networks include P group convolution
Layer, each group of convolutional layer includes two convolutional layers, and the convolution kernel of described two convolutional layers is respectively the first convolution that size is 1*N
Core and the second convolution kernel that size is N*1；
Feature extraction is carried out to the images to be recognized by each group of convolutional layer of first convolutional neural networks, institute is obtained
State the feature to be extracted of each group of convolutional layer；
The recognition result of the images to be recognized is determined according to the feature to be extracted of each group of convolutional layer.
2. method according to claim 1, it is characterised in that methods described also includes：
Each convolutional layer in second convolutional neural networks is decomposed into into described two convolutional layers, the first convolution god is obtained
Jing networks, each convolutional layer in second convolutional neural networks includes a square convolution kernel.
3. method according to claim 2, it is characterised in that each convolutional layer in second convolutional neural networks
Convolution kernel be size for N*N square convolution kernel, second convolutional neural networks include P convolutional layer；
Described each convolutional layer by the second convolutional neural networks is decomposed into described two convolutional layers, including：
The size of each convolutional layer in second convolutional neural networks is decomposed into into size for 1* for the square convolution kernel of N*N
First convolution kernel of N and size for N*1 the second convolution kernel, first convolution kernel is with the product value of second convolution kernel
The square convolution kernel；
Corresponding volume in second convolutional neural networks is replaced respectively using first convolution kernel and second convolution kernel
The square convolution kernel of lamination, obtains one group of convolutional layer of first convolutional neural networks.
4. method according to claim 1, it is characterised in that each group by first convolutional neural networks
Convolutional layer carries out feature extraction to the images to be recognized, including：
Feature is carried out using a convolutional layer in each group of convolutional layer of first convolutional neural networks to input information to carry
Take, obtain the intermediate value of the feature to be extracted of each group of convolutional layer；
Another convolutional layer that the intermediate value of the feature to be extracted is input in each group of convolutional layer is carried out into feature extraction,
Obtain the feature to be extracted of each group of convolutional layer.
5. a kind of pattern recognition device, it is characterised in that described device includes：
Input module, is configured to the first convolutional neural networks for having trained images to be recognized input, the first convolution god
Jing networks include P group convolutional layers, and each group of convolutional layer includes two convolutional layers, and the convolution kernel of described two convolutional layers is respectively greatly
Little the first convolution kernel and size for 1*N is second convolution kernel of N*1；
Characteristic extracting module, is configured to each group of convolutional layer of first convolutional neural networks to the figure to be identified
As carrying out feature extraction, the feature to be extracted of each group of convolutional layer is obtained；
Determining module, is configured to determine the identification of the images to be recognized according to the feature to be extracted of each group of convolutional layer
As a result.
6. device according to claim 5, it is characterised in that described device also includes：
Decomposing module, is configured to for each convolutional layer in the second convolutional neural networks to be decomposed into described two convolutional layers,
First convolutional neural networks are obtained, each convolutional layer in second convolutional neural networks includes a square convolution
Core.
7. device according to claim 6, it is characterised in that each convolutional layer in second convolutional neural networks
Convolution kernel be size for N*N square convolution kernel, second convolutional neural networks include P convolutional layer；
The decomposing module includes：
Decompose submodule, it is the square of N*N to be configured to the size of each convolutional layer in second convolutional neural networks
Convolution kernel be decomposed into size be 1*N the first convolution kernel and size for N*1 the second convolution kernel, first convolution kernel with it is described
The product value of the second convolution kernel is the square convolution kernel；
Determination sub-module, is configured with first convolution kernel and second convolution kernel replaces respectively second convolution
The square convolution kernel of corresponding convolutional layer in neutral net, obtains one group of convolutional layer of first convolutional neural networks.
8. device according to claim 5, it is characterised in that the characteristic extracting module includes：
First extracting sub-module, a convolution being configured to, with each group of convolutional layer of first convolutional neural networks
Layer carries out feature extraction to input information, obtains the intermediate value of the feature to be extracted of each group of convolutional layer；
Second extracting sub-module, is configured to the intermediate value of just described feature to be extracted and is input in each group of convolutional layer
Another convolutional layer carries out feature extraction, obtains the feature to be extracted of each group of convolutional layer.
9. a kind of pattern recognition device, it is characterised in that described device includes：
Processor；
For storing the memorizer of processor executable；
Wherein, the processor is configured to：
The first convolutional neural networks that images to be recognized input has been trained, first convolutional neural networks include P group convolution
Layer, each group of convolutional layer includes two convolutional layers, and the convolution kernel of described two convolutional layers is respectively the first convolution that size is 1*N
Core and the second convolution kernel that size is N*1；
Feature extraction is carried out to the images to be recognized by each group of convolutional layer of first convolutional neural networks, institute is obtained
State the feature to be extracted of each group of convolutional layer；
The recognition result of the images to be recognized is determined according to the feature to be extracted of each group of convolutional layer.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710034836.7A CN106682736A (en) | 2017-01-18 | 2017-01-18 | Image identification method and apparatus |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710034836.7A CN106682736A (en) | 2017-01-18 | 2017-01-18 | Image identification method and apparatus |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106682736A true CN106682736A (en) | 2017-05-17 |
Family
ID=58859233
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710034836.7A Pending CN106682736A (en) | 2017-01-18 | 2017-01-18 | Image identification method and apparatus |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106682736A (en) |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107944545A (en) * | 2017-11-10 | 2018-04-20 | 中国科学院计算技术研究所 | Computational methods and computing device applied to neutral net |
| CN107943750A (en) * | 2017-11-14 | 2018-04-20 | 华南理工大学 | A kind of decomposition convolution method based on WGAN models |
| CN107967459A (en) * | 2017-12-07 | 2018-04-27 | 北京小米移动软件有限公司 | convolution processing method, device and storage medium |
| CN109033940A (en) * | 2018-06-04 | 2018-12-18 | 上海依图网络科技有限公司 | A kind of image-recognizing method, calculates equipment and storage medium at device |
| CN109784186A (en) * | 2018-12-18 | 2019-05-21 | 深圳云天励飞技术有限公司 | A kind of pedestrian recognition methods, device, electronic equipment and computer readable storage medium again |
| WO2019105243A1 (en) * | 2017-11-28 | 2019-06-06 | 北京达佳互联信息技术有限公司 | Image processing method and device and terminal |
| CN109978137A (en) * | 2019-03-20 | 2019-07-05 | 厦门美图之家科技有限公司 | A kind of processing method of convolutional neural networks |
| CN110858323A (en) * | 2018-08-23 | 2020-03-03 | 北京京东金融科技控股有限公司 | Convolution-based image processing method, convolution-based image processing device, convolution-based image processing medium and electronic equipment |
| CN110880034A (en) * | 2018-09-06 | 2020-03-13 | 三星电子株式会社 | Computing device using convolutional neural network and method of operating the same |
| CN111247527A (en) * | 2017-12-20 | 2020-06-05 | 华为技术有限公司 | Method and device for determining characteristic image in convolutional neural network model |
| US10699160B2 (en) | 2017-08-23 | 2020-06-30 | Samsung Electronics Co., Ltd. | Neural network method and apparatus |
| CN111798520A (en) * | 2020-09-08 | 2020-10-20 | 平安国际智慧城市科技股份有限公司 | Image processing method, device, equipment and medium based on convolutional neural network |
| CN109840584B (en) * | 2017-11-24 | 2023-04-18 | 腾讯科技（深圳）有限公司 | Image data classification method and device based on convolutional neural network model |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104408435A (en) * | 2014-12-05 | 2015-03-11 | 浙江大学 | Face identification method based on random pooling convolutional neural network |
| CN106156807A (en) * | 2015-04-02 | 2016-11-23 | 华中科技大学 | The training method of convolutional neural networks model and device |
| CN106326985A (en) * | 2016-08-18 | 2017-01-11 | 北京旷视科技有限公司 | Neural network training method, neural network training device, data processing method and data processing device |
-
2017
- 2017-01-18 CN CN201710034836.7A patent/CN106682736A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104408435A (en) * | 2014-12-05 | 2015-03-11 | 浙江大学 | Face identification method based on random pooling convolutional neural network |
| CN106156807A (en) * | 2015-04-02 | 2016-11-23 | 华中科技大学 | The training method of convolutional neural networks model and device |
| CN106326985A (en) * | 2016-08-18 | 2017-01-11 | 北京旷视科技有限公司 | Neural network training method, neural network training device, data processing method and data processing device |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10909418B2 (en) | 2017-08-23 | 2021-02-02 | Samsung Electronics Co., Ltd. | Neural network method and apparatus |
| US10699160B2 (en) | 2017-08-23 | 2020-06-30 | Samsung Electronics Co., Ltd. | Neural network method and apparatus |
| CN107944545A (en) * | 2017-11-10 | 2018-04-20 | 中国科学院计算技术研究所 | Computational methods and computing device applied to neutral net |
| CN107944545B (en) * | 2017-11-10 | 2020-07-31 | 中国科学院计算技术研究所 | Computing method and computing device applied to neural network |
| CN107943750A (en) * | 2017-11-14 | 2018-04-20 | 华南理工大学 | A kind of decomposition convolution method based on WGAN models |
| CN109840584B (en) * | 2017-11-24 | 2023-04-18 | 腾讯科技（深圳）有限公司 | Image data classification method and device based on convolutional neural network model |
| WO2019105243A1 (en) * | 2017-11-28 | 2019-06-06 | 北京达佳互联信息技术有限公司 | Image processing method and device and terminal |
| CN107967459B (en) * | 2017-12-07 | 2021-08-24 | 北京小米移动软件有限公司 | Convolution processing method, convolution processing device and storage medium |
| CN107967459A (en) * | 2017-12-07 | 2018-04-27 | 北京小米移动软件有限公司 | convolution processing method, device and storage medium |
| CN111247527A (en) * | 2017-12-20 | 2020-06-05 | 华为技术有限公司 | Method and device for determining characteristic image in convolutional neural network model |
| CN111247527B (en) * | 2017-12-20 | 2023-08-22 | 华为技术有限公司 | Method and device for determining characteristic images in convolutional neural network model |
| CN109033940A (en) * | 2018-06-04 | 2018-12-18 | 上海依图网络科技有限公司 | A kind of image-recognizing method, calculates equipment and storage medium at device |
| CN110858323A (en) * | 2018-08-23 | 2020-03-03 | 北京京东金融科技控股有限公司 | Convolution-based image processing method, convolution-based image processing device, convolution-based image processing medium and electronic equipment |
| CN110880034A (en) * | 2018-09-06 | 2020-03-13 | 三星电子株式会社 | Computing device using convolutional neural network and method of operating the same |
| CN109784186A (en) * | 2018-12-18 | 2019-05-21 | 深圳云天励飞技术有限公司 | A kind of pedestrian recognition methods, device, electronic equipment and computer readable storage medium again |
| CN109978137B (en) * | 2019-03-20 | 2021-03-16 | 厦门美图之家科技有限公司 | Processing method of convolutional neural network |
| CN109978137A (en) * | 2019-03-20 | 2019-07-05 | 厦门美图之家科技有限公司 | A kind of processing method of convolutional neural networks |
| CN111798520A (en) * | 2020-09-08 | 2020-10-20 | 平安国际智慧城市科技股份有限公司 | Image processing method, device, equipment and medium based on convolutional neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106682736A (en) | Image identification method and apparatus | |
| JP6852150B2 (en) | Biological detection methods and devices, systems, electronic devices, storage media | |
| CN106651955B (en) | Method and device for positioning target object in picture | |
| CN105809704A (en) | Method and device for identifying image definition | |
| CN104700353B (en) | Image filters generation method and device | |
| CN106530227A (en) | Image restoration method and device | |
| CN106548145A (en) | Image-recognizing method and device | |
| CN106228556B (en) | image quality analysis method and device | |
| CN106528879A (en) | Picture processing method and device | |
| CN107527059A (en) | Character recognition method, device and terminal | |
| CN106980840A (en) | Shape of face matching process, device and storage medium | |
| CN107563994A (en) | The conspicuousness detection method and device of image | |
| CN108010060A (en) | Object detection method and device | |
| CN107480665A (en) | Character detecting method, device and computer-readable recording medium | |
| CN106225764A (en) | Based on the distance-finding method of binocular camera in terminal and terminal | |
| CN105354560A (en) | Fingerprint identification method and device | |
| CN107784279A (en) | Method for tracking target and device | |
| CN107766820A (en) | Image classification method and device | |
| CN108062547A (en) | Character detecting method and device | |
| CN110717399A (en) | Face recognition method and electronic terminal equipment | |
| CN107967459A (en) | convolution processing method, device and storage medium | |
| CN106648063A (en) | Gesture recognition method and device | |
| CN107832746A (en) | Expression recognition method and device | |
| CN107463903A (en) | Face key independent positioning method and device | |
| CN104867112B (en) | Photo processing method and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170517 |
|
| RJ01 | Rejection of invention patent application after publication |