CN117274821A - Multi-polarization SAR farmland flood detection method and system considering rainfall influence - Google Patents

Abstract

The invention relates to the technical field of radar image processing, specifically to a multi-polarization SAR farmland flood detection method and system that takes into account the impact of rainfall, including preprocessing dual-temporal multi-polarization SAR image data to generate a polarization covariance matrix; calculating the same The intensity ratio of polarization and cross-polarization channels and the HLT statistical measure are used to construct a SAR image polarization difference measure, and the polarization difference map is calculated and generated pixel by pixel; the Markov random field model is used to detect changes in multi-polarization SAR images; according to For the change detection results, the digital elevation model is used to reduce the interference of mountain shadows on the detection results, and the farmland damage results are extracted; the farmland damage results under the SAR coordinate system are geocoded, and the final farmland flood detection results under the geographical coordinate system are output. The present invention can effectively avoid false detections caused by rainfall, improve the applicability of multi-polarization SAR images for farmland flood detection in rainy weather, and improve the accuracy of polarization SAR flood disaster detection.

Description

Multi-polarization SAR farmland flood detection method and system taking into account the impact of rainfall

Technical field

The invention relates to the technical field of radar image processing, and in particular to a multi-polarization SAR farmland flood detection method and system that takes into account the impact of rainfall.

Background technique

Synthetic Aperture Radar (SAR) remote sensing is an active microwave imaging remote sensing method that can penetrate clouds and fog and has the advantage of working around the clock and all-weather. Since the occurrence of farmland floods is often accompanied by rainfall, traditional optical remote sensing technology faces difficulties in obtaining surface images. Therefore, the use of SAR remote sensing technology to monitor flood disasters shows great potential for development. In recent years, with the extensive application of multi-polarization SAR satellites and other observation platforms, the cost of obtaining polarization SAR data has been continuously reduced, and the application of multi-polarization SAR data has become more and more widespread. Compared with single-polarization SAR images, multi-polarization SAR images have multiple polarization channels including co-polarization and cross-polarization, which allows them to extract richer information.

In the process of flood disaster detection, change detection is a commonly used method. Change detection is a process in the field of remote sensing that analyzes changes in attributes of a specific area between two or more time phases. At present, the main change detection methods based on multipolarization SAR are as follows:

(1) Polarimetric SAR image change detection method based on similarity measure

The detection method based on similarity measure is an unsupervised change detection strategy. Its basic process is to use multi-temporal images to construct a difference map, and then analyze the difference map to identify the changed area. Since the covariance matrix of multi-temporal polarization SAR images obeys the complex Wishart distribution, metrics such as SRW distance, geodesic distance, and Kullback-Leibler divergence can be derived. The similarity measure can effectively evaluate the similarity of dual/multi-temporal polarization SAR images, construct difference maps between different phases, and further achieve change detection in the target area.

(2), PCC-based change detection method

Post Classification Comparison (PCC) refers to a method that first classifies the images to be detected and then performs change detection by comparing the dual/multi-temporal classification results. Compared with unsupervised methods based on similarity measures, this method adopts a supervised method and can directly obtain the change category. However, its accuracy relies on the classification results of a single phase, and it is prone to a decrease in change detection accuracy due to single-phase classification errors. situation; at the same time, due to its supervised nature, it is greatly affected by subjective factors.

(3). Change detection method based on deep learning

In recent years, with the development of neural network technologies such as convolutional neural networks and recurrent neural networks, change detection methods based on deep learning have also made greater progress. The deep learning model can independently extract image features and has stronger ability to express spatiotemporal information. Since deep learning requires a large number of well-labeled samples, today's remote sensing data sets are difficult to satisfy. Therefore, weakly supervised deep learning methods and reinforcement learning methods constitute the main research direction of change detection methods based on deep learning.

The above-mentioned change detection methods have achieved good performance in specific fields. However, since these methods do not consider the impact of rainfall that may be encountered in flood disaster change detection, their change detection accuracy still needs to be improved.

Contents of the invention

The purpose of the present invention is to provide a multi-polarization SAR farmland flood detection method that takes into account the impact of rainfall, so as to solve the problems raised in the background technology.

In order to achieve the above objectives, the present invention provides a multi-polarization SAR farmland flood detection method that takes into account the impact of rainfall, including the following steps:

(1) Dual-phase multi-polarization SAR image data preprocessing: perform registration, multi-viewing and filtering processing on the two-phase multi-polarization SAR data to generate the polarization covariance matrix C ;

(2) Generation of polarization difference map: Based on the polarization covariance matrix C generated in step (1), calculate the intensity ratio and the Hotelling-Lawley trace statistical measure of the co-polarization and cross-polarization channels respectively, and construct The polarization difference measure D of multi-polarization SAR images is taken into account the influence of rainfall, and then the polarization difference map is calculated and generated pixel by pixel;

(3) Multi-polarization SAR image change detection based on Markov random fields: Use the Otsu method to pre-segment the polarization difference map obtained in step (2), and input the segmented binary matrix as the label field Markov random field model, using Markov random field model to detect changes in multi-polarization SAR images;

(4) Extraction of farmland flood disaster areas: Based on the change detection results in step (3), use the digital elevation model to weaken the interference of mountain shadows on the detection results, and extract the farmland disaster results;

(5) Output the farmland flood detection results: After extracting the farmland damage results, geocode the farmland damage results in the SAR coordinate system and output the final farmland flood detection results in the geographical coordinate system.

Further, in the step (1), the data preprocessing of the dual-phase multi-polarization SAR image is specifically: register the image of the second phase to the first phase, and perform multi-view and refined Lee filtering. Process and generate polarization covariance matrix C :

1);

In Formula 1), H represents the conjugate transpose operation, and k represents the multi-polarization scattering vector.

Further, in step (2), a polarization intensity ratio is constructed based on the intensities of co-polarization and cross-polarization channels. :

2);

In formula 2), Represents the intensity value of the co-polarized channel,/> Represents the intensity value of the cross-polarization channel; based on the difference in the performance of water pixels on two different polarization channels under the influence of rainfall,/> Ability to identify areas of water bodies disturbed by rainfall.

Further, based on the polarization covariance matrix of the two phases and/> , the Hotelling-Lawley trace statistical measure can be calculated/> :

3);

In formula 3), Represents trace operation,/> Represents the maximum value operation,/> Indicates the polarization dimension, in the case of full polarization/> , in the case of dual polarization/> ;

Taking the Hotelling-Lawley trace statistical measure and polarization intensity ratio/> The difference of is used to construct a polarization difference measure D that takes into account the influence of rainfall:

4);

Calculate the difference measure pixel by pixel according to Equation 4) , you can get the polarization difference diagram/> .

Further, in step (3), the polarization difference diagram is calculated based on the Otsu method The specific steps for segmentation are: convert the polarization difference map/> Pre-segment into unchanged areas and changed areas, and use the Otsu method to calculate the polarization difference map/> Inter-class variance corresponding to gray value/> , the corresponding threshold T when the inter-class variance takes the maximum value will polarize the difference map/> Segmentation into changed or unchanged results, the image segmentation results can be used as the initial label field of the Markov random field model.

Further, the Otsu method is used to calculate the polarization difference map. When the gray value is a gray value, when the inter-class variance takes any threshold T, the inter-class variance/> The calculation formula is:

5);

In equation 5), Indicates the proportion of sample pixels of the two types of sample pixels, the changed class and the unchanged class, in the total pixels when the threshold T is taken,/> Represents the average gray value of two types of sample pixels,/> Represents the overall average gray value of the image.

Further, the polarization difference map As the observation field data, enter the Markov random field model, fuse the eight-neighborhood information, and construct the energy function/> , and generate the prior probability of the Markov random field marker field/> :

6);

In formula 6), is the normalization factor,/> is the temperature parameter,/> is an exponential function;

Afterwards, assuming that the polarization difference measure Follows a Gaussian distribution, based on changed and unchanged classes

mean of class pixels and variance/> Calculate the likelihood functions of the two categories/> :

7);

In equation 7), is an exponential function; the posterior probability can then be calculated/> :

8);

based on , determine the change type pixel by pixel according to the principle of maximum probability, and update the label field. Iterate the process from formula 6) to formula 8) until the maximum number of iterations is reached, and output the change detection result.

Further, in step (4), the specific steps of extracting the farmland disaster area are:

First, extract the terrain slope based on the digital elevation model and set the slope threshold and/> ;Among them, the defined slope is less than the slope threshold/> The pixel is a water body, and the slope is defined to be between the slope threshold/> and/> The pixels between are weak water body candidate points;

Then, search within the neighborhood window of each weak candidate point whether there is a strong candidate point, and if so, update the pixel to a strong candidate point;

After that, all weak candidate point pixels are traversed, and pixels belonging to water bodies are screened out based on neighborhood information;

Finally, the strong candidate point results extracted twice are combined to generate farmland damage results.

Further, in step (5), the obtained farmland damage results are converted from the SAR coordinate system to the geographical coordinate system as follows:

First, based on the radar echo time length and echo Doppler characteristics, the initial position of the image point is calculated, and the initial lookup table of slant range pixels and geographical pixels is generated;

Then, the simulated SAR intensity image is calculated based on the digital elevation model, matched with the real SAR intensity image, the matching polynomial is obtained, a refined lookup table is generated, and the image positioning is completed based on the lookup table, and each SAR disaster detection result image pixel is assigned a geographical location. coordinates, and finally realize the mapping and release of farmland flood disaster detection results.

The present invention also provides a multi-polarization SAR farmland flood detection system that takes into account the impact of rainfall, including:

The polarization SAR data preprocessing module is used to perform registration, multi-view and filtering preprocessing operations on dual-temporal multi-polarization SAR data;

The polarization difference map generation module is used to calculate the polarization difference measure of the preprocessed image to generate a polarization difference map;

Polarimetric SAR change detection module, used to classify images into changed and unchanged classes;

The farmland flood disaster area extraction module uses digital elevation model data based on the Markov random field model to reduce the impact of mountain shadows and extract the disaster-stricken areas;

The geocoding module assigns geographical coordinates to each SAR disaster detection result image pixel and outputs the final farmland flood detection results in the geographical coordinate system.

Compared with the prior art, the present invention has the following beneficial effects:

The present invention is a multi-polarization SAR farmland flood detection method that takes into account the impact of rainfall. It uses the intensity ratio of co-polarization and cross-polarization channels to describe the impact of rainfall, and combines it with Hotelling-Lawley trace statistics to measure the detection distance of structural changes. The Markov random field model is introduced to realize farmland flood disaster detection. Compared with traditional methods, the method of the present invention can effectively avoid false detections caused by rainfall, improve the applicability of multi-polarization SAR images for farmland flood detection under rainfall weather, and improve The accuracy of polarization SAR flood disaster detection provides an effective solution for the detection of farmland flood disasters. In addition, the present invention also has the advantages of low computational cost and high reliability of results, and has broad application prospects. It can not only be used in the field of farmland flood disaster detection, but also has the potential to have an impact on other fields that rely on remote sensing image change detection. The method of the invention not only has practical value, but also has important commercial value and social benefit, and has made certain contributions to the advancement of flood disaster detection technology.

In addition to the objects, features and advantages described above, the present invention has other objects, features and advantages. The present invention will be described in further detail below with reference to the accompanying drawings.

Description of the drawings

The drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the description. Together with the following specific implementation modes, they are used to explain the embodiments of the present invention, but do not constitute a limitation to the embodiments of the present invention. In the attached picture:

Figure 1 is a flow chart of a specific embodiment of the present invention;

Figure 2 is a schematic diagram of the detection results of farmland flood disaster changes according to a specific embodiment of the present invention; (a) is a pre-disaster multi-polarization SAR pseudo-color composite image, (b) is a post-disaster multi-polarization SAR pseudo-color composite image , (c) is the polarization difference map, (d) is the flood disaster monitoring result;

Figure 3 is a schematic structural diagram of a detection system in a specific embodiment of the present invention.

Detailed ways

The present invention will be described in detail below with reference to the various embodiments shown in the accompanying drawings. However, it should be noted that these embodiments do not limit the invention. Those of ordinary skill in the art may make functional, method or structural modifications based on these embodiments. Equivalent transformations or substitutions are within the scope of the present invention.

Referring to Figures 1 and 2, embodiments of the present invention provide a multi-polarization SAR farmland flood detection method that takes into account the impact of rainfall, in which the Sentinel-1 dual-polarization spaceborne SAR system is used to detect changes in farmland flood disasters in the Poyang Lake area. Take detection as an example. There are many features in the area, including farmland, cities, and lakes, and the situation is complex; post-disaster images are greatly affected by rainfall. The multipolarization SAR farmland flood detection method includes the following steps:

Step 1. Dual-phase multi-polarization SAR image data preprocessing: Perform preprocessing operations such as registration, multi-viewing and filtering on the multi-polarization SAR data of the first and second phases to obtain the polarization covariance matrix C. Among them, the registration operation is based on the first phase, and the image of the second phase is registered to the first phase, and the filtering uses a refined Lee filter. The obtained polarization covariance matrix C is shown in Equation 1):

1);

In Formula 1), H represents the conjugate transpose operation, and k represents the multi-polarization scattering vector.

Step 2. Generation of polarization difference map: Based on the polarization covariance matrix C generated in step 1, calculate the intensity of the co-polarization and cross-polarization channels respectively, and construct the polarization intensity ratio :

2);

In formula 2), Represents the intensity value of the co-polarized channel,/> Represents the intensity value of the cross-polarization channel; based on the difference in the performance of water pixels on two different polarization channels under the influence of rainfall,/> Ability to identify areas of water bodies disturbed by rainfall. Further, based on the polarization covariance matrix of the two phases/> and/> , the Hotelling-Lawley trace (HLT) statistical measure can be calculated/> :

3);

In formula 3), Represents trace operation,/> Represents the maximum value operation,/> Indicates the polarization dimension, in the case of full polarization/> , in the case of dual polarization/> . After that, take/> Statistical Measures/> and polarization intensity ratio/> The difference of is used to construct a polarization difference measure D that takes into account the influence of rainfall:

4);

According to Equation 4), the difference measure D is calculated pixel by pixel, and the polarization difference map can be obtained .

Step 3. Multi-polarization SAR image change detection based on Markov Random Field (MRF): Use the Otsu method to pre-segment the polarization difference map constructed in step (2), and first calculate the polarization difference map Inter-class variance corresponding to gray value/> , the specific formula is as follows:

5);

In equation 5), Indicates the proportion of sample pixels of the two types of sample pixels, the changed class and the unchanged class, in the total pixels when the threshold T is taken,/> Represents the average gray value of sample pixels in the changed and unchanged classes,/> Represents the overall average gray value of the image. Taking the between-class variance/> The corresponding threshold T when taking the maximum value is used as the threshold, and the binarized matrix obtained by dividing it is input into the Markov random field model as the label field, and the eight-neighbor information is fused to construct the energy function/> , and generate the prior probability of the Markov random field marker field/> :

6);

In formula 6), is the normalization factor,/> is the temperature parameter,/> is an exponential function. Afterwards, assuming a polarization difference measure/> Obey Gaussian distribution, based on the mean value of the changed and unchanged class cells/> and variance/> Calculate the likelihood functions of the two categories/> :

7);

In equation 7), is an exponential function; the posterior probability can then be calculated/> :

8);

based on , determine the change type pixel by pixel according to the principle of maximum probability, and update the label field. Iterate the above process until the maximum number of iterations is reached, and output the change detection result.

Step 4. Extract the farmland flood disaster area: Based on the change detection results in Step 3, use the Digital Elevation Model (DEM) to extract the terrain slope and set the slope threshold. and/> . Among them, the defined slope is less than the slope threshold/> pixels belong to water bodies and the slope is between the slope threshold/> and/> The pixels between them may belong to water bodies or mountain shadows. The slope of this part of pixels is in the overlapping area of the histogram distribution of the two types of samples and is called a weak water body candidate point; then, in the neighborhood of each weak candidate point Search whether there is a strong candidate point in the domain window, and if so, update the pixel to a strong candidate point; then, traverse all weak candidate point pixels, and filter out pixels belonging to the water body based on neighborhood information; finally, merge the two The extracted strong candidate point results are used to generate farmland damage results.

Step 5. Output the farmland flood disaster detection results: After extracting the disaster area, geocode the disaster results in the SAR coordinate system. Based on the radar echo time length and echo Doppler characteristics, calculate the initial position of the image point and generate the slant range. Initial lookup table for pixels and geographical pixels; then, calculate the simulated SAR intensity image based on the DEM, match it with the real SAR intensity image, obtain the matching polynomial, generate a refined lookup table, and complete image positioning based on the lookup table, for each The image pixels of the SAR disaster detection results are assigned geographical coordinates, and the final farmland flood detection results under the geographical coordinate system are output.

As shown in Figure 3, the present invention also provides a multi-polarization SAR farmland flood detection system that takes into account the impact of rainfall, including:

Polarimetric SAR data preprocessing module 1 is used to perform registration, multi-view and filtering preprocessing operations on dual-temporal multi-polarization SAR data;

Polarization difference map generation module 2 is used to calculate the polarization difference measure of the preprocessed image to generate a polarization difference map;

Polarimetric SAR change detection module 3, used to divide images into changed and unchanged classes;

Farmland flood disaster area extraction module 4 uses digital elevation model data based on the Markov random field model to reduce the impact of mountain shadows and extract the disaster-stricken area;

The geocoding module 5 assigns geographical coordinates to each SAR disaster detection result image pixel, and outputs the final farmland flood detection results in the geographical coordinate system.

Each module included in the multipolarization SAR farmland flood detection system provided by this embodiment belongs to the functional implementation module of the multipolarization SAR farmland flood detection method in the above embodiment. The specific working principle and implementation method are the same as those described in the above method embodiment. are the same and will not be repeated here.

The above descriptions are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and changes. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection scope of the present invention.

Claims (10)

1. A multi-polarization SAR farmland flood detection method that takes into account the impact of rainfall, is characterized by including the following steps: (1) Dual-phase multi-polarization SAR image data preprocessing: perform registration, multi-viewing and filtering preprocessing on the two-phase multi-polarization SAR data to generate the polarization covariance matrix C ; (2) Generation of polarization difference map: Based on the polarization covariance matrix C generated in step (1), calculate the intensity ratio and Hotelling-Lawley trace statistical measure of the co-polarization and cross-polarization channels respectively, Construct a multi-polarization SAR image polarization difference measure D that takes into account the impact of rainfall, and then calculate and generate a polarization difference map pixel by pixel; (3) Multi-polarization SAR image change detection based on Markov random fields: Use the Otsu method to pre-segment the polarization difference map obtained in step (2), and input the segmented binary matrix as the label field Markov random field model, using Markov random field model to detect changes in multi-polarization SAR images; (4) Extraction of farmland flood disaster areas: Based on the change detection results in step (3), use the digital elevation model to weaken the interference of mountain shadows on the detection results, and extract the farmland disaster results; (5) Output the farmland flood detection results: After extracting the farmland damage results, geocode the farmland damage results in the SAR coordinate system and output the final farmland flood detection results in the geographical coordinate system. 2. The multi-polarization SAR farmland flood detection method according to claim 1, characterized in that in the step (1), data preprocessing of the dual-phase multi-polarization SAR image is specifically: The image of the phase is registered to the first phase, and multi-view and refined Lee filtering are performed to generate the polarization covariance matrix C : 1); In Formula 1), H represents the conjugate transpose operation, and k represents the multi-polarization scattering vector. 3. The multi-polarization SAR farmland flood detection method according to claim 2, characterized in that in the step (2), a polarization intensity ratio is constructed based on the intensities of co-polarization and cross-polarization channels. : 2); In formula 2), Represents the intensity value of the co-polarized channel,/> Represents the intensity value of the cross-polarization channel; based on the difference in the performance of water pixels on two different polarization channels under the influence of rainfall,/> Ability to identify areas of water bodies disturbed by rainfall. 4. The multi-polarization SAR farmland flood detection method according to claim 3, characterized in that, based on the polarization covariance matrix of two phases and/> , the Hotelling-Lawley trace statistical measure can be calculated/> : 3); In formula 3), Represents trace operation,/> Represents the maximum value operation,/> Represents the polarization dimension, in the case of full polarization , in the case of dual polarization/> ; Taking the Hotelling-Lawley trace statistical measure and polarization intensity ratio/> The difference between is used to construct a polarization difference measure D that takes into account the influence of rainfall: 4); Calculate the difference measure pixel by pixel according to Equation 4) , you can get the polarization difference diagram/> . 5. The multi-polarization SAR farmland flood detection method according to claim 4, characterized in that in the step (3), the polarization difference map is calculated based on the Otsu method. The specific steps for segmentation are: convert the polarization difference map/> Pre-segment into unchanged areas and changed areas, and use the Otsu method to calculate the polarization difference map/> Inter-class variance corresponding to gray value/> , the corresponding threshold T when the inter-class variance takes the maximum value will polarize the difference map/> Segmentation into changed or unchanged results, the image segmentation results can be used as the initial label field of the Markov random field model. 6. The multi-polarization SAR farmland flood detection method according to claim 5, characterized in that the inter-class variance When taking any threshold T, its calculation formula is: 5); In equation 5), Indicates the proportion of changed class and unchanged class sample pixels in the total pixels when the threshold T is taken,/> Represents the average gray value of sample pixels in the changed and unchanged classes,/> Represents the overall average gray value of the image. 7. The multi-polarization SAR farmland flood detection method according to claim 5, characterized in that the polarization difference map As the observation field data, enter the Markov random field model, fuse the eight-neighborhood information, and construct the energy function/> , and generate the prior probability of the Markov random field marker field/> : 6); In formula 6), is the normalization factor,/> is the temperature parameter,/> is an exponential function; Afterwards, assuming that the polarization difference measure Obey Gaussian distribution, based on the mean value of the changed and unchanged class cells/> and variance/> Calculate the likelihood functions of the two categories/> : 7); In equation 7), is an exponential function; the posterior probability can then be calculated/> : 8); based on , determine the change type pixel by pixel according to the principle of maximum probability, and update the label field. Iterate formula 6) to formula 8) until the maximum number of iterations is reached, and output the change detection result. 8. The multipolarization SAR farmland flood detection method according to claim 1, characterized in that in step (4), extracting the farmland disaster area is specifically: First, extract the terrain slope based on the digital elevation model and set the slope threshold and/> ;Among them, the defined slope is less than the slope threshold/> The pixel is a water body, and the slope is defined to be between the slope threshold/> and/> The pixels between are weak water body candidate points; Then, search within the neighborhood window of each weak candidate point whether there is a strong candidate point, and if so, update the pixel to a strong candidate point; After that, all weak candidate point pixels are traversed, and pixels belonging to water bodies are screened out based on neighborhood information; Finally, the strong candidate point results extracted twice are combined to generate farmland damage results. 9. The multi-polarization SAR farmland flood detection method according to claim 8, characterized in that in the step (5), converting the obtained farmland damage results from the SAR coordinate system to the geographical coordinate system is specifically: First, based on the radar echo time length and echo Doppler characteristics, the initial position of the image point is calculated, and the initial lookup table of slant range pixels and geographical pixels is generated; Then, the simulated SAR intensity image is calculated based on the digital elevation model, matched with the real SAR intensity image, the matching polynomial is obtained, a refined lookup table is generated, and the image positioning is completed based on the lookup table, and each SAR disaster detection result image pixel is assigned a geographical location. coordinates, and finally realize the mapping and release of farmland flood disaster detection results. 10. A system based on the multipolarization SAR farmland flood detection method according to any one of claims 1 to 9, characterized in that it includes: Polarimetric SAR data preprocessing module (1), used for registration, multi-view and filtering preprocessing operations on dual-temporal multi-polarization SAR data; The polarization difference map generation module (2) is used to calculate the polarization difference measure of the preprocessed image to generate a polarization difference map; Polarimetric SAR change detection module (3), used to classify images into changed and unchanged classes; The farmland flood disaster area extraction module (4) uses digital elevation model data based on the Markov random field model to reduce the influence of mountain shadows and extract the disaster-stricken areas; The geocoding module (5) assigns geographical coordinates to each SAR disaster detection result image pixel, and outputs the final farmland flood detection results in the geographical coordinate system.