Flood disaster water body range detection method, device, equipment and storage medium

Abstract

The present invention provides a method and device, equipment and storage medium for detecting the range of a water body in a flood disaster, wherein the method comprises: obtaining a neighborhood logarithmic ratio difference map of two-phase SAR images before and after the flood disaster; performing fuzzy clustering segmentation processing on the neighborhood logarithmic ratio difference map to obtain an initial water body change detection map; using a preset network classifier to predict and process the intermediate pixels in the initial water body change detection map to obtain target change pixels and target non-change pixels; the preset network classifier includes a classifier obtained by training an initial convolutional neural network based on the change pixels and non-change pixels in the initial water body change detection map; based on the target change pixels and the target non-change pixels, the initial water body change detection map is updated to obtain a target water body change detection map, wherein the intermediate pixels include pixels between change and non-change. The present invention can improve the detection accuracy and change detection effect of the range of a water body in a flood disaster.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a method and device, equipment and storage medium for detecting a water body range of a flood disaster. Background Art

[0002] In recent years, the use of SAR images for change detection has been applied in flood disasters. For example, the use of SAR images for water body change detection has unique advantages in the quantitative assessment of the scope of flood disasters. However, when flood disasters are accompanied by extremely complex and severe weather conditions such as cloudy, rainy, and foggy, the SAR satellite cannot penetrate the clouds, making it difficult to use SAR impact assessment to assess the flood disaster situation.

[0003] In the existing methods, when combining neural network models for SAR image water body change detection, the high-dimensional features of the SAR images that are less affected by noise are first extracted, and then based on the learning ability of the neural network model for samples, a model distribution based on high-dimensional features is established to achieve learning and prediction of the input data, thereby achieving the purpose of identifying image information.

[0004] However, the existing methods require a large number of samples due to the need to obtain high-dimensional features of SAR images. Among the large number of samples obtained, there are few available water samples, many available samples of other ground objects, and few high-quality samples with high clarity, resulting in low accuracy and poor detection effect in water range detection. Summary of the invention

[0005] The present invention provides a method and device, equipment and storage medium for detecting the range of water bodies in flood disasters, so as to solve the defects of low accuracy and poor detection effect of water body detection in the prior art, and achieve the purpose of reliable and efficient high-precision detection of the range of water bodies in flood disasters.

[0006] The present invention provides a method for detecting a flood disaster water body range, comprising:

[0007] Obtain the neighborhood logarithmic ratio difference map of the two-phase SAR images before and after the flood disaster;

[0008] Performing fuzzy clustering segmentation processing on the neighborhood logarithmic ratio difference map to obtain an initial water body change detection map; wherein the initial water body change detection map includes intermediate pixels, first pixels and high-confidence non-changed pixels, the intermediate pixels include pixels between changed and non-changed, and the first pixels include changed pixels and non-changed pixels;

[0009] Using a preset network classifier to perform prediction processing on the intermediate pixels to obtain target changed pixels and target non-changed pixels; wherein the preset network classifier includes a classifier obtained by training an initial convolutional neural network initial classifier based on the first pixel;

[0010] Based on the target changed pixels and the target non-changed pixels, the initial water body change detection map is updated to obtain a target water body change detection map.

[0011] According to a flood disaster water body range detection method provided by the present invention, the fuzzy clustering segmentation process is performed on the neighborhood logarithmic ratio difference map to obtain an initial water body change detection map, including:

[0012] Performing fuzzy clustering segmentation processing on the neighborhood logarithmic ratio difference map to obtain Q clusters and the weight of the cluster center of each cluster;

[0013] Sorting the Q clusters according to the size relationship between the weights to obtain Q sorted clusters and a total number NUM of pixels included in the Q sorted clusters;

[0014] According to the preset category pixel determination conditions, Num is determined from the total number of pixels NUM. CP Number of changed pixels, Num IP The number of intermediate pixels and Num UP non-changing pixels;

[0015] Obtaining high confidence non-changing pixels in the neighborhood log ratio difference map;

[0016] Determined by the Num CP The number of changed pixels, the IP The Num intermediate pixels UP The initial water body change detection map is composed of non-changing pixels and the high-confidence non-changing pixels.

[0017] According to a flood disaster water body range detection method provided by the present invention, the training process of the preset network classifier includes:

[0018] The SAR images of the two time phases before and after the flood disaster are processed with sliding windows of preset step lengths to obtain M×N image patches of the SAR image of each time phase;

[0019] Based on the initial water body change detection map, channel fusion processing is performed on the M×N image spots to obtain M×N dual-channel image spots;

[0020] Performing sample labeling processing on the M×N dual-channel spots respectively to obtain change samples and non-change samples in the M×N training samples; wherein the sample labeling processing is used to characterize each of the dual-channel spots as a change sample, a non-change sample, an intermediate sample or a high-confidence sample;

[0021] The inverse of the sum of the changed samples and the non-changed samples is used as the loss function weight of the initial convolutional neural network, and the changed samples and the non-changed samples are used to train the initial classifier of the initial convolutional neural network to obtain a preset network classifier.

[0022] According to a flood disaster water body range detection method provided by the present invention, according to the preset category pixel determination condition, Num is determined from the total number of pixels NUM. CP Number of changed pixels, Num IP The number of intermediate pixels and Num UP non-changing pixels, including:

[0023] Set the number of pixels contained in each of the Q sorted clusters to be Num1, Num2, ..., Num Q ; Among them, Num1>Num2>…>Num Q , Num1+Num2+…+Num Q =NUM;

[0024] When Num1+…+Num a-1 <Num CP-max And Num1+…+Num a-1 +Num a >Num CP-max When Num is determined, the number of pixels changed CP =Num1+…+Num a-1 And the category t included in the changed pixel c =a-1; where a is an integer greater than 1 and less than Q, Num CP-max Num is the preset upper limit of the number of pixels changed. CP-max =Num×T low , T low is the preset threshold value of the number of pixels of the change category, Num is the number of pixels of the change in the difference enhancement map of the SAR images before and after the flood disaster, t c ∈[1,Q];

[0025] when and When And the category t contained in the intermediate pixel i =t c +b-1; where b is an integer greater than 2 and less than Q, t i ∈[1,Qt c ], T high is the preset threshold value of the number of pixels of unknown category;

[0026] when And m=min{2,…,Qt c -t i}, determine the number of non-changing pixels And the category t included in the non-changing pixel u =t i +t c +m; where t u ∈[1,Qt i -t c ].

[0027] According to a flood disaster water body range detection method provided by the present invention, the T low is the preset threshold value of the number of pixels of the change class and the T high is a preset threshold value of the number of pixels of unknown category, and its determination process includes:

[0028] Obtaining a difference enhancement map of the two-phase SAR images before and after the flood disaster and a histogram of the difference enhancement map;

[0029] Performing triangular threshold processing on the histogram to obtain a first sub-two-dimensional triangular graph and a second sub-two-dimensional triangular graph representing the relationship between the gray level and the frequency of the histogram;

[0030] Obtain a first line segment between a first frequency of a first gray level and its perpendicular foot in the first sub-two-dimensional triangular graph, and a second line segment between a second frequency of a second gray level and its perpendicular foot in the second sub-two-dimensional triangular graph;

[0031] respectively determining a first target gray level corresponding to when the first line segment is a maximum value and a second target gray level corresponding to when the second line segment is a maximum value;

[0032] Determine the number of pixels of the changed category and the number of pixels of the unknown category in the difference enhancement map based on the first target grayscale level and the second target grayscale level;

[0033] Based on the number of pixels of the change category and the number of pixels of the to-be-determined category, the threshold for limiting the number of pixels of the change category is set to T low And the threshold limiting the number of pixels of the unknown category is set to T high .

[0034] According to a method for detecting the range of a water body in a flood disaster provided by the present invention, the step of obtaining a difference enhancement map of two-phase SAR images before and after the flood disaster comprises:

[0035] Respectively obtaining a logarithmic ratio difference map and a logarithmic likelihood ratio difference map of the two-phase SAR images before and after the flood disaster;

[0036] Based on the logarithmic ratio difference map and the logarithmic likelihood ratio difference map, a difference enhancement map of the SAR images in two phases before and after the flood disaster is obtained.

[0037] According to a flood disaster water body range detection method provided by the present invention, the step of obtaining high-confidence non-changing pixels in the neighborhood logarithmic ratio difference map comprises:

[0038] Remove Num from the neighborhood logarithmic ratio difference map CP The number of changed pixels, the IP The Num intermediate pixels UP The remaining pixels other than the non-changed pixels are determined as high-confidence non-changed pixels in the neighborhood logarithmic ratio difference map.

[0039] The present invention also provides a flood disaster water body range detection device, comprising:

[0040] The acquisition module is used to obtain the neighborhood logarithmic ratio difference map of the two-phase SAR images before and after the flood disaster;

[0041] A segmentation module is used to perform fuzzy clustering segmentation processing on the neighborhood logarithmic ratio difference map to obtain an initial water body change detection map; wherein the initial water body change detection map includes intermediate pixels, first pixels and high-confidence non-changed pixels, the intermediate pixels include pixels between changed and non-changed, and the first pixels include changed pixels and non-changed pixels;

[0042] A prediction module, used to perform prediction processing on the intermediate pixels using a preset network classifier to obtain target changed pixels and target non-changed pixels; wherein the preset network classifier includes a classifier obtained by training an initial convolutional neural network based on the first pixel.

[0043] An updating module is used to update the initial water body change detection map based on the target changed pixels and the target non-changed pixels to obtain a target water body change detection map.

[0044] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of any of the above-mentioned methods for detecting the range of a flood disaster water body are implemented.

[0045] The present invention also provides a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of any of the above-mentioned methods for detecting the range of a flood disaster water body are implemented.

[0046] The present invention also provides a computer program product, comprising a computer program, which, when executed by a processor, implements the steps of any of the above-mentioned methods for detecting the range of a flood disaster water body.

[0047] The present invention provides a method and device, equipment and storage medium for detecting the range of a water body in a flood disaster, wherein the method for detecting the range of a water body in a flood disaster first performs fuzzy clustering segmentation processing on a neighborhood logarithmic ratio difference map of two-phase SAR images before and after the flood disaster to obtain an initial water body change detection map; then uses a preset network classifier to predict and process the intermediate pixels in the initial water body change detection map to obtain target change pixels and target non-change pixels, and finally updates the initial water body change detection map based on the target change pixels and target non-change pixels to obtain a target water body change detection map. Since the preset network classifier is a classifier obtained by training an initial classifier in an initial convolutional neural network with change pixels and non-change pixels in the initial water body change detection map, the preset network classifier can accurately and reliably parse the target change pixels and target non-change pixels that constitute the intermediate pixels, and further uses the target change pixels and target non-change pixels to update the initial water body change detection map to achieve the purpose of efficient and accurate detection of the range of the water body in the flood disaster, thereby improving the detection accuracy and change detection effect of the range of the water body in the flood disaster. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0049] Figure 1 It is a schematic diagram of the flow of the flood disaster water body range detection method provided by the present invention;

[0050] Figure 2 It is a schematic diagram of a simulation experiment image of two-phase SAR images before and after a flood disaster provided by the present invention;

[0051] Figure 3 It is a schematic diagram of selecting a unilateral triangle threshold provided by the present invention;

[0052] Figure 4 It is a schematic diagram of the process of constructing a convolutional neural network input sample provided by the present invention;

[0053] Figure 5 It is a schematic diagram of transformation detection results of simulation experiments using the method of the present invention and various existing methods provided by the present invention;

[0054] Figure 6 It is a structural schematic diagram of a flood disaster water body range detection device provided by the present invention;

[0055] Figure 7 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0057] Since the United States successfully achieved its first airborne synthetic aperture radar experiment in 1960, the development of SAR sensors that actively emit electromagnetic waves and receive backscattered echo signals from the ground has been going on for nearly sixty years. The emergence and development of SAR sensors have effectively made up for the shortcomings of optical imaging being affected by cloudiness and light, and have the advantages of imaging not being restricted by insufficient light and climatic conditions, making it the only available means of acquiring remote sensing information in certain special scenarios.

[0058] Change detection using SAR images has been applied in many fields such as crop monitoring, building detection, urbanization research and geological disaster monitoring. For example, in forest fire detection, change detection using SAR images can effectively survey the fire range; based on the special sensitivity of SAR images to water areas, it has unique advantages in performing sea surface oil spill change detection, flood wave change detection, flood wave range change detection and coastline extraction; but when serious geological disasters are accompanied by extremely complex and severe weather and environmental conditions, optical satellites cannot penetrate the clouds to obtain images, making it difficult to use optical images to timely assess the disaster situation at the epicenter.

[0059] In recent years, the development of deep learning related methods has provided a different research idea for SAR image change detection from traditional methods. In particular, the development of neural network models has made it possible to extract high-dimensional features of images that are less affected by noise. By utilizing the powerful learning ability of the neural network model itself, it can effectively establish a model distribution based on high-dimensional features of data through learning samples, realize learning and prediction of input data, and thus achieve the purpose of accurately identifying image information. However, the existing methods not only have a high demand for samples, but also it is usually difficult to obtain high-quality samples, and the samples are prone to imbalance problems, resulting in a low degree of automation of the detection method and poor detection effect.

[0060] Based on the above problems existing in the existing methods, the present invention provides a method for detecting the range of water bodies in flood disasters. The execution subject of the method can be a flood disaster water body range detection device, and the flood disaster water body range detection device can be implemented as part or all of the terminal device through software, hardware or a combination of software and hardware. Optionally, the terminal device can be a personal computer (PC), a portable device, a laptop computer, a smart phone, a tablet computer, a portable wearable device and other electronic devices, such as a tablet computer, a mobile phone, etc. The present invention does not limit the specific form of the terminal device.

[0061] It should be noted that the execution subject of the following method embodiment may be part or all of the above terminal device. The following method embodiment is described by taking the execution subject as the terminal device as an example.

[0062] Figure 1 The flowchart of the flood disaster water body range detection method provided by the present invention is as follows: Figure 1 As shown, the flood disaster water body range detection method includes the following steps:

[0063] Step S110, obtaining a neighborhood logarithmic ratio difference map of two-phase SAR images before and after the flood disaster.

[0064] Specifically, the terminal device first obtains two-phase SAR images before and after the flood disaster, that is, obtains a first SAR image I1 before the flood disaster and a second SAR image I2 after the flood disaster. The size and number of pixels of the first SAR image I1 and the second SAR image I2 are the same. The simulation experiment image diagrams of the first SAR image I1 and the second SAR image I2 can be shown as follows: Figure 2 As shown, in Figure 2 In the figure, (a) is the first SAR image taken in July 1997, and (b) is the second SAR image taken in August 1997, both of which are 350*290 pixels in size.

[0065] Then, based on formula (1), the neighborhood logarithmic ratio difference map D of the two-phase SAR images before and after the flood disaster is determined. LMR , the neighborhood logarithmic ratio difference map D generated by formula (1) LMR It can have a better differential expression effect, which is more conducive to the accurate extraction of changed pixels; the neighborhood logarithmic ratio difference map D LMR Each neighborhood logarithmic ratio difference pixel D LMR The determination process of (i,j) includes:

[0066]

[0067] In formula (1), I1(m,n) is the pixel at position (m,n) in the P neighborhood of the pixel at position (i,j) in the first SAR image I1, and I2(m,n) is the pixel at position (m,n) in the P neighborhood of the pixel at position (i,j) in the second SAR image I2. is the P neighborhood of the pixel at position (m,n) in the first SAR image I1, is the P neighborhood of the pixel at position (m, n) in the second SAR image I2, μ(·) is the mean operation, max(·) is the maximum operation, log(·) is the logarithm operation, 1≤i≤M, 1≤j≤N, M×N is the size of the first SAR image I1 or the second SAR image I2. Preferably, P can be 3.

[0068] Step S120: Perform fuzzy clustering segmentation processing on the neighborhood logarithmic ratio difference map to obtain an initial water body change detection map.

[0069] The initial water body change detection map includes intermediate pixels, first pixels and high-confidence non-changed pixels, the intermediate pixels include pixels between changed and non-changed pixels, and the first pixels include changed pixels and non-changed pixels.

[0070] Specifically, the terminal device can use the multi-level FCM clustering algorithm to cluster the neighborhood logarithmic ratio difference map D LMR An initial segmentation process is performed to obtain an initial water body change detection map including high-confidence non-changing pixels, changing pixels, non-changing pixels, and pixels between changing and non-changing pixels. Since non-changing pixels can be low-confidence non-changing pixels and non-changing pixels and changing pixels have been determined, only non-changing pixels and changing pixels can be used as training samples. Since intermediate pixels are pixels between changing and non-changing pixels, intermediate pixels can be used as pixels to be identified. In this way, by adopting a multi-hierarchical FCM clustering algorithm, the purpose of reducing the impact of sample imbalance on the final result while accurately determining the number of samples is achieved, thereby improving the accuracy of change detection.

[0071] Step S130: Use a preset network classifier to perform prediction processing on the intermediate pixels to obtain target changed pixels and target non-changed pixels.

[0072] Among them, the preset network classifier includes a classifier obtained after training an initial classifier of an initial convolutional neural network based on the first pixel.

[0073] Specifically, since the preset network classifier is a classifier obtained after training the initial convolutional neural network based on the changed pixels and the non-changed pixels, the intermediate pixels are input into the preset classifier, and the target changed pixels and the target non-changed pixels that form the intermediate pixels can be quickly and efficiently extracted, thereby achieving the purpose of improving the detection accuracy of the method of the present invention by adopting an adaptive dual-channel convolutional neural network.

[0074] Step S140: Based on the target changed pixels and the target non-changed pixels, the initial water body change detection map is updated to obtain a target water body change detection map.

[0075] Specifically, when the terminal device extracts the target change pixels and target non-change pixels that form the intermediate pixels, it can further replace the intermediate pixels in the initial water body change detection map based on the target change pixels and the target non-change pixels, thereby realizing the update processing of the initial water body change detection map, thereby obtaining the target water body change detection map.

[0076] The method for detecting the range of a water body in a flood disaster provided by the present invention first performs fuzzy clustering segmentation processing on the neighborhood logarithmic ratio difference map of the two-phase SAR images before and after the flood disaster to obtain an initial water body change detection map; then uses a preset network classifier to predict and process the intermediate pixels in the initial water body change detection map to obtain target change pixels and target non-change pixels; finally, based on the target change pixels and target non-change pixels, the initial water body change detection map is updated and processed to obtain the target water body change detection map. Since the preset network classifier is a classifier obtained by training the initial classifier in the initial convolutional neural network with the change pixels and non-change pixels in the initial water body change detection map, the preset network classifier can accurately and reliably parse the target change pixels and target non-change pixels that constitute the intermediate pixels, and further by using the target change pixels and target non-change pixels to update the initial water body change detection map, the purpose of efficient and accurate detection of the range of the water body in the flood disaster is achieved, thereby improving the detection accuracy and change detection effect of the range of the water body in the flood disaster.

[0077] Optionally, the specific implementation process of step S120 includes:

[0078] First, the neighborhood logarithmic ratio difference map is subjected to fuzzy clustering segmentation processing to obtain Q clusters and the weight of the cluster center of each cluster; then, the Q clusters are sorted according to the size relationship between the weights to obtain Q sorted clusters and the total number of pixels NUM contained in the Q sorted clusters; then, according to the preset category pixel determination condition, Num are determined from the total number of pixels NUM. CP Number of changed pixels, Num IP The number of intermediate pixels and NumUP non-changed pixels; further obtaining high-confidence non-changed pixels in the neighborhood logarithmic ratio difference map; and finally determining the Num CP The number of changed pixels, the IP The Num intermediate pixels UP The initial water body change detection map is composed of non-changing pixels and the high-confidence non-changing pixels.

[0079] Specifically, a multi-level FCM clustering algorithm is performed on the neighborhood logarithmic ratio difference map to obtain Q clusters, and then the weight of the cluster center of each cluster is further determined, wherein the weight of the cluster center of each cluster can be the midpoint of multiple gray levels in the corresponding cluster, for example, when a cluster includes three gray levels of 11, 12, and 13, the weight of the cluster center of the cluster is 12. In this way, Q weights of Q cluster centers are obtained.

[0080] Then sort the Q weights from large to small, and sort the Q clusters according to the sorted Q weights. The resulting Q sorted clusters are C1, C2, ..., C Q , the cluster centers of each sorted cluster are Center1, Center2, ..., Center Q ; Among them, the weight of the cluster center Center1 of C1 is the largest, Q The cluster center Q The weight of the cluster center is the smallest; the smaller the weight of the cluster center, the smaller the probability of the corresponding category of pixels changing. Finally, Num is determined in turn from the total number of pixels NUM contained in the Q sorted clusters. CP Number of changed pixels, Num IP The number of intermediate pixels and Num UP non-changing pixels.

[0081] It should be noted that the step of obtaining the high-confidence non-changing pixels in the neighborhood logarithmic ratio difference map includes: removing Num CP The number of changed pixels, the IP The Num intermediate pixels UP The remaining pixels other than the non-changed pixels are determined as high-confidence non-changed pixels in the neighborhood logarithmic ratio difference map.

[0082] The flood disaster water body range detection method provided by the present invention first sorts Q clusters according to the size relationship between the weights of the cluster centers of Q clusters obtained after fuzzy cluster segmentation processing based on the neighborhood logarithmic ratio difference map, and then sequentially determines Num from the total number of pixels NUM contained in the sorted Q clusters. CP Number of changed pixels, Num IPThe number of intermediate pixels and Num UP The method of using non-changing pixels can achieve the purpose of efficiently and accurately obtaining the initial water body change detection map, and can also greatly improve the change detection accuracy.

[0083] Optionally, according to the preset category pixel determination condition, Num CP Number of changed pixels, Num IP The number of intermediate pixels and Num UP non-changing pixels, including:

[0084] Set the number of pixels contained in each of the Q sorted clusters to be Num1, Num2, ..., Num Q ; Among them, Num1>Num2>…>Num Q , Num1+Num2+…+Num Q =NUM;

[0085] When Num1+…+Num a-1 <Num CP-max And Num1+…+Num a-1 +Num a >Num CP-max When Num is determined, the number of pixels changed CP =Num1+…+Num a-1 And the category t included in the changed pixel c =a-1; where a is an integer greater than 1 and less than Q, Num CP-max Num is the preset upper limit of the number of pixels changed. CP-max =Num×T low , T low is the preset threshold value of the number of pixels of the change category, Num is the number of pixels of the change in the difference enhancement map of the SAR images before and after the flood disaster, t c ∈[1,Q];

[0086] when and When And the category t contained in the intermediate pixel i =t c +b-1; where b is an integer greater than 2 and less than Q, t i ∈[1,Qt c ], T high is the preset threshold value of the number of pixels of unknown category;

[0087] when And m=min{2,…,Qt c -t i}, determine the number of non-changing pixels And the category t included in the non-changing pixel u =t i +t c +m; where t u ∈[1,Qt i -t c ].

[0088] Specifically, when the value of Q is 6, that is, after the multi-level FCM clustering algorithm is performed on the neighborhood logarithmic ratio difference map, 6 clusters are obtained. At this time, it can be determined that Num1<Num CP-max , but it is not sure whether Num1+Num2 is greater than Num CP-max , when Num1+Num2>Num CP-max Determine Num CP =Num1 and t c =1; when Num1+Num2<Num CP-max And Num1+Num2+Num3>Num CP-max When CP =Num1+Num2 and t c =2; preferably, T low The value of is set to 0.8;

[0089] When t c =1, determine whether Num2+Num3 is greater than Num×T high , if Num2+Num3>Num×T high , determine the number of intermediate pixels Num IP =Num2 and the category t contained in the middle pixel i =1; if Num CP-max <Num2+Num3<Num×T high And Num2+Num3+Num4>Num×T high , determine the number of intermediate pixels Num IP =Num2+Num3 and the category t contained in the middle pixel i =2; preferably, T high The value of is set to 1.2;

[0090] When t c =1, t i =2, determine whether Num4+Num5 is greater than Num CP , when Num4+Num5>Num CP When Num is determined, the number of non-changing pixels is determined. UP =Num4 and the category t included in the non-changing pixels u=1; when Num4+Num5<Num CP And Num4+Num5+Num6>Num CP When Num is determined, the number of non-changing pixels is determined. UP =Num4+Num5 and the category t included in the non-changing pixels u =2.

[0091] The method for detecting the range of water bodies for flood disasters provided by the present invention achieves the purpose of successively determining the changed pixels, intermediate pixels and non-changed pixels constituting the initial water body change detection map by successively judging the size relationship between the sum of the number of pixels included in more than two sorted clusters and different pixel thresholds, thereby improving the reliability and accuracy of the initial water body change detection map.

[0092] Optionally, the T low is the preset threshold value of the number of pixels of the change class and the T high is a preset threshold value of the number of pixels of unknown category, and its determination process includes:

[0093] Obtain a difference enhancement map of the two-phase SAR images before and after the flood disaster and a histogram of the difference enhancement map; perform triangular threshold processing on the histogram to obtain a first sub-two-dimensional triangular map and a second sub-two-dimensional triangular map representing the relationship between the grayscale and the frequency of the histogram; obtain a first line segment between the first frequency of the first grayscale and its perpendicular foot in the first sub-two-dimensional triangular map, and a second line segment between the second frequency of the second grayscale and its perpendicular foot in the second sub-two-dimensional triangular map; respectively determine a first target grayscale corresponding to a maximum value when the first line segment is a maximum value and a second target grayscale corresponding to a maximum value when the second line segment is a maximum value; determine the number of pixels of the change class and the number of pixels of the unknown class in the difference enhancement map based on the first target grayscale and the second target grayscale; and set the threshold for limiting the number of pixels of the change class to T based on the number of pixels of the change class and the number of pixels of the class to be determined. low And the threshold limiting the number of pixels of the unknown category is set to T high .

[0094] Specifically, first obtain the difference enhancement map of the two-phase SAR images before and after the flood disaster, and then further obtain the two-dimensional triangular map that represents the grayscale-frequency relationship of the histogram. The horizontal axis direction of the two-dimensional triangular map includes multiple different grayscales L1-Lmax, each grayscale corresponds to a frequency and each frequency is represented by a vertical line. The higher the vertical line, the greater the frequency of the corresponding grayscale. The frequency of each vertical line is H1-Hmax. The frequency of L1 is the smallest and coincides with its frequency H1, and the frequency of Lmax is the largest and its frequency Hmax is the highest. In addition, the two-dimensional triangular map is divided into a first sub-two-dimensional triangular map and a second sub-two-dimensional triangular map with the vertical line segment LmaxHmax as the symmetry axis.

[0095] like Figure 3 The figure shows a schematic diagram of the selection of the single-sided triangle threshold. The gray level Hn of the gray level Ln in the first sub-two-dimensional triangle is selected, and the height of Hn is extended until it intersects with the line segment H1Hmax. The intersection point is recorded as hn. Then a perpendicular line is drawn from Hn to H1Hmax. The foot of the perpendicular line is recorded as Dn. The first target gray level L corresponding to the maximum length of the first line segment DnHn is n Set as the segmentation threshold of the maximum frequency value in the first sub-2D triangular graph.

[0096] Similarly, select the gray level Hm of the gray level Lm in the second sub-two-dimensional triangle graph, extend the height of Hm until it intersects with the line segment H1Hmax, record the intersection as hm, and then draw a perpendicular line from Hm to H1Hmax, record the foot of the perpendicular line as Dm, and write the second target gray level L corresponding to the maximum length of the second line segment DmHm m Set as the segmentation threshold of the maximum frequency value in the second sub-2D triangular graph.

[0097] Then, the gray level between L n and L m The pixels between are determined as unchanged pixels, and the grayscale values ​​greater than L in the difference enhancement map are m Or the gray level is less than L n The pixels are determined as changed pixels, and the number of statistically changed pixels is Num.

[0098] Finally, based on the value of Num, the threshold value of the number of pixels in the change category is determined as T low And the threshold value of the number of pixels of unknown category is T high .

[0099] The method for detecting the range of water bodies in flood disasters provided by the present invention realizes the purpose of extracting potential changed pixels in the difference enhancement map by combining the histogram information through triangular threshold segmentation of the histogram of the difference enhancement map of the two-phase SAR images before and after the flood disaster, which is beneficial to accurately control the number of changed samples in the sample, thereby improving the detection accuracy.

[0100] Optionally, obtaining a difference enhancement map of two-phase SAR images before and after the flood disaster includes:

[0101] First, a logarithmic ratio difference map and a logarithmic likelihood ratio difference map of the two-phase SAR images before and after the flood disaster are obtained respectively; then, based on the logarithmic ratio difference map and the logarithmic likelihood ratio difference map, a difference enhancement map of the two-phase SAR images before and after the flood disaster is obtained.

[0102] Specifically, we first calculate the logarithmic ratio difference map D of the SAR images before and after the flood disaster using the logarithmic ratio operator. LR , log-ratio difference plot D LR Each logarithmic ratio difference pixel D LR The calculation formula for (i,j) is as follows:

[0103]

[0104] Secondly, the likelihood ratio difference map D is calculated by the likelihood ratio operator for the SAR images before and after the flood disaster. LLI , likelihood ratio difference map D LLI Each likelihood ratio difference pixel D LLI The calculation formula for (i,j) is as follows:

[0105]

[0106] Furthermore, for the likelihood ratio difference map D LLI Perform logarithmic operations to obtain the log-likelihood ratio difference map D LLR , log-likelihood ratio difference map D LLR Each log-likelihood ratio difference pixel D LLR The calculation formula for (i,j) is as follows:

[0107]

[0108] Finally, the difference enhancement map D of the two-phase SAR images before and after the flood disaster was determined. DI , Difference Enhanced Graph D DI Each difference enhancement pixel D DI The calculation formula for (i,j) is as follows:

[0109] D DI (i,j)=D LR (i,j)×D LLR (i,j) (5)

[0110] Where I1(i, j) is the pixel at position (i, j) in the first SAR image I1, I2(i, j) is the pixel at position (i, j) in the second SAR image I2, Ωi,j is the P neighborhood of the pixel at position (i, j) in the first SAR image I1 or the second SAR image I2, I1(m, n) is the pixel at position (m, n) in the P neighborhood of the pixel at position (i, j) in the first SAR image I1, I2(m, n) is the pixel at position (m, n) in the P neighborhood of the pixel at position (i, j) in the second SAR image I2, L is the number of multi-views in the first SAR image I1 or the second SAR image I2, ∑ is the summation symbol, 1≤i≤M, 1≤j≤N, and M×N is the size of the first SAR image I1 or the second SAR image I2.

[0111] The flood disaster water body range detection method provided by the present invention realizes the purpose of quickly obtaining the difference enhancement map of the two-phase SAR images before and after the flood disaster through the logarithmic ratio operator, the likelihood ratio operator and the logarithmic operation, so as to provide a strong basis for the subsequent accurate extraction of changed pixels in combination with the better difference expression effect of the difference enhancement map.

[0112] Optionally, the training process of the preset network classifier includes:

[0113] The two-phase SAR images before and after the flood disaster are respectively subjected to sliding window processing with a preset step length to obtain M×N spots of each phase SAR image; based on the initial water body change detection map, the M×N spots are subjected to channel fusion processing to obtain M×N dual-channel spots; the M×N dual-channel spots are respectively subjected to sample annotation processing to obtain changed samples and non-changed samples in M×N training samples; wherein the sample annotation processing is used to characterize each of the dual-channel spots as a changed sample, a non-changed sample, an intermediate sample or a high-confidence sample; the inverse of the sum of the changed samples and the non-changed samples is used as the loss function weight of the initial convolutional neural network, and the changed samples and the non-changed samples are used to train the initial classifier of the initial convolutional neural network to obtain a preset network classifier.

[0114] Specifically, Figure 4As shown in FIG. 1 , firstly, sliding window processing with a preset step length is performed on the first SAR image I1 and the second SAR image I2, such as sliding window processing with a step length of 1, so as to obtain M×N spots of the first SAR image I1 and M×N spots of the second SAR image I2, respectively; then, channel fusion processing is performed on the M×N spots of each phase SAR image based on the initial water body change detection map, so as to obtain M×N dual-channel spots; further, sample labeling processing is performed on the M×N dual-channel spots, so as to determine whether each dual-channel spot is a change category, a non-change category, an intermediate category or a high confidence category, so as to obtain change samples and non-change samples in the M×N training samples; the change samples and non-change samples in the M×N training samples are the same as the number of change pixels and non-change pixels in the initial water body change detection map, respectively. Figure 4 The initial segmentation result in is the initial water body change detection map.

[0115] In addition, the initial classifier of the initial convolutional neural network is iteratively trained using changing samples and non-changing samples. When the value of the loss function of the intermediate convolutional neural network obtained after this round of iterative training reaches the preset loss accuracy, the iterative training is stopped, and the classifier of the intermediate convolutional neural network corresponding to the time when the iterative training stops is used as the preset network classifier; when the value of the loss function of the intermediate convolutional neural network obtained after this round of iterative training does not reach the preset loss accuracy, the intermediate convolutional neural network is iteratively trained again until the preset network classifier is obtained.

[0116] The flood disaster water body range detection method provided by the present invention not only achieves the purpose of quickly and efficiently obtaining a preset network classifier by determining the loss function weight based on the number of changed samples and non-changed samples and training the initial convolutional neural network based on changed samples and non-changed samples, but also reduces the impact of the imbalance between changed and non-changed samples in the samples by discarding the patches corresponding to high-confidence non-changed pixels, thereby greatly improving the prediction accuracy and prediction reliability of the preset network classifier.

[0117] In the actual processing process, in order to verify the effectiveness of the method of the present invention, the method of the present invention, the K-means clustering algorithm based on principal component analysis, the Markov random field and FCM algorithm, the Gabor feature-based double-layer clustering method, the Garbor feature-based principal component analysis network method and the method of the present invention can be used for simulation experiments, so as to obtain the following Figure 5The various simulation result diagrams shown in Figure (5) are: (a) is a schematic diagram of the change detection results of the K-means clustering algorithm based on principal component analysis, (b) is a schematic diagram of the change detection results based on Markov random field and FCM, (c) is a schematic diagram of the change detection results based on the Gabor feature double-layer clustering method, (d) is a schematic diagram of the change detection results based on the Garbor feature principal component analysis network method, (e) is a schematic diagram of the change detection results obtained based on the method of the present invention, and (f) is a reference schematic diagram of change detection.

[0118] Simulation experiments show that compared with other SAR image change detection methods, the method of the present invention has a higher degree of automation and better change detection effect.

[0119] The flood disaster water body range detection device provided by the present invention is described below. The flood disaster water body range detection device described below and the flood disaster water body range detection method described above can correspond to each other.

[0120] Figure 6 An example of a flood disaster water body range detection device is provided. Figure 6 As shown, the flood disaster water body range detection device 600 includes:

[0121] The acquisition module 610 is used to obtain a neighborhood logarithmic ratio difference map of two-phase SAR images before and after a flood disaster; the segmentation module 620 is used to perform fuzzy clustering segmentation processing on the neighborhood logarithmic ratio difference map to obtain an initial water body change detection map; wherein the initial water body change detection map includes intermediate pixels, first pixels and high-confidence non-change pixels, the intermediate pixels include pixels between changes and non-changes, and the first pixels include changes and non-change pixels; the prediction module 630 is used to use a preset network classifier to perform prediction processing on the intermediate pixels to obtain target changes pixels and target non-change pixels; wherein the preset network classifier includes a classifier obtained by training an initial convolutional neural network based on the first pixels. The update module 640 is used to update the initial water body change detection map based on the target changes pixels and the target non-change pixels to obtain a target water body change detection map.

[0122] Optionally, the segmentation module 620 can be specifically used to perform fuzzy clustering segmentation processing on the neighborhood logarithmic ratio difference map to obtain Q clusters and the weight of the cluster center of each cluster; sort the Q clusters according to the size relationship between the weights to obtain Q sorted clusters and the total number of pixels NUM contained in the Q sorted clusters; determine Num from the total number of pixels NUM according to the preset category pixel determination condition CP Number of changed pixels, Num IP The number of intermediate pixels and NumUP non-changed pixels; obtaining high-confidence non-changed pixels in the neighborhood logarithmic ratio difference map; determining the non-changed pixels by the Num CP The number of changed pixels, the IP The Num intermediate pixels UP The initial water body change detection map is composed of non-changing pixels and the high-confidence non-changing pixels.

[0123] Optionally, the prediction module 630 can be specifically used to perform sliding window processing with a preset step size on the two-phase SAR images before and after the flood disaster to obtain M×N spots of each phase SAR image; based on the initial water body change detection map, the M×N spots are subjected to channel fusion processing to obtain M×N dual-channel spots; the M×N dual-channel spots are subjected to sample annotation processing to obtain changed samples and non-changed samples in the M×N training samples; wherein the sample annotation processing is used to characterize each of the dual-channel spots as a changed sample, a non-changed sample, an intermediate sample or a high-confidence sample; the inverse of the sum of the changed samples and the non-changed samples is used as the loss function weight of the initial convolutional neural network, and the changed samples and the non-changed samples are used to train the initial classifier of the initial convolutional neural network to obtain a preset network classifier.

[0124] Optionally, the segmentation module 620 may be specifically configured to set the number of pixels contained in each of the Q sorted clusters to be Num1, Num2, ..., Num Q ; Among them, Num1>Num2>…>Num Q , Num1+Num2+…+Num Q =NUM; when Num1+…+Num a-1 <Num CP-max And Num1+…+Num a-1 +Num a >Num CP-max When Num is determined, the number of pixels changed CP =Num1+…+Num a-1 And the category t included in the changed pixel c =a-1; where a is an integer greater than 1 and less than Q, Num CP-max Num is the preset upper limit of the number of pixels changed. CP-max =Num×T low , T low is the preset threshold value of the number of pixels of the change category, Num is the number of pixels of the change in the difference enhancement map of the SAR images before and after the flood disaster, t c ∈[1,Q]; when and When And the category t contained in the intermediate pixel i =t c +b-1; where b is an integer greater than 2 and less than Q, t i ∈[1,Qt c ], T high is the preset unknown category pixel number threshold; when And m=min{2,…,Qt c -t i}, determine the number of non-changing pixels And the category t included in the non-changing pixel u =t i +t c +m; where t u ∈[1,Qt i -t c ].

[0125] Optionally, the segmentation module 620 may be specifically used to obtain a difference enhancement map of the SAR images of two time phases before and after the flood disaster and a histogram of the difference enhancement map;

[0126] The histogram is subjected to triangular threshold processing to obtain a first sub-two-dimensional triangular graph and a second sub-two-dimensional triangular graph representing the grayscale-frequency relationship of the histogram; a first line segment between a first frequency of a first grayscale and its perpendicular foot in the first sub-two-dimensional triangular graph and a second line segment between a second frequency of a second grayscale and its perpendicular foot in the second sub-two-dimensional triangular graph are obtained; a first target grayscale corresponding to a maximum value of the first line segment and a second target grayscale corresponding to a maximum value of the second line segment are determined respectively; the number of pixels of the changed class and the number of pixels of the unknown class in the difference enhancement image are determined based on the first target grayscale and the second target grayscale; a threshold limiting the number of pixels of the changed class is set to T based on the number of pixels of the changed class and the number of pixels of the class to be determined. low And the threshold limiting the number of pixels of the unknown category is set to T high .

[0127] Optionally, the acquisition module 610 can be specifically used to respectively obtain a logarithmic ratio difference map and a logarithmic likelihood ratio difference map of the two-phase SAR images before and after the flood disaster; based on the logarithmic ratio difference map and the logarithmic likelihood ratio difference map, a difference enhancement map of the two-phase SAR images before and after the flood disaster is obtained.

[0128] Optionally, the acquisition module 610 may be specifically configured to remove Num CP The number of changed pixels, the IPThe Num intermediate pixels UP The remaining pixels other than the non-changed pixels are determined as high-confidence non-changed pixels in the neighborhood logarithmic ratio difference map.

[0129] Figure 7 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 7 As shown, the electronic device 700 may include: a processor (processor) 710, a communication interface (Communications Interface) 720, a memory (memory) 730 and a communication bus 740, wherein the processor 710, the communication interface 720, and the memory 730 communicate with each other through the communication bus 740. The processor 710 can call the logic instructions in the memory 730 to execute a method for detecting the range of a water body in a flood disaster, the method comprising: obtaining a neighborhood logarithmic ratio difference map of two-phase SAR images before and after a flood disaster; performing fuzzy clustering segmentation processing on the neighborhood logarithmic ratio difference map to obtain an initial water body change detection map; wherein the initial water body change detection map comprises an intermediate pixel, a first pixel and a high-confidence non-change pixel, the intermediate pixel comprises a pixel between a change and a non-change, and the first pixel comprises a change pixel and a non-change pixel; using a preset network classifier to perform prediction processing on the intermediate pixel to obtain a target change pixel and a target non-change pixel; wherein the preset network classifier comprises a classifier obtained by training an initial convolutional neural network based on the first pixel; based on the target change pixel and the target non-change pixel, the initial water body change detection map is updated to obtain a target water body change detection map.

[0130] In addition, the logic instructions in the above-mentioned memory 730 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0131] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the flood disaster water body range detection method provided by the above-mentioned methods, and the method includes: obtaining a neighborhood logarithmic ratio difference map of two-phase SAR images before and after the flood disaster; performing fuzzy clustering segmentation processing on the neighborhood logarithmic ratio difference map to obtain an initial water body change detection map; wherein the initial water body change detection map includes intermediate pixels, first pixels and high-confidence non-change pixels, the intermediate pixels include pixels between changes and non-changes, and the first pixels include change pixels and non-change pixels; using a preset network classifier to predict the intermediate pixels to obtain target change pixels and target non-change pixels; wherein the preset network classifier includes a classifier obtained by training an initial convolutional neural network based on the first pixel. The initial water body change detection map is updated based on the target change pixels and the target non-change pixels to obtain a target water body change detection map.

[0132] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the flood disaster water body range detection method provided by the above-mentioned methods, the method comprising: obtaining a neighborhood logarithmic ratio difference map of two-phase SAR images before and after the flood disaster; performing fuzzy clustering segmentation processing on the neighborhood logarithmic ratio difference map to obtain an initial water body change detection map; wherein the initial water body change detection map includes intermediate pixels, first pixels and high-confidence non-change pixels, the intermediate pixels include pixels between changes and non-changes, and the first pixels include changes and non-change pixels; using a preset network classifier to perform prediction processing on the intermediate pixels to obtain target changes pixels and target non-change pixels; wherein the preset network classifier includes a classifier obtained by training an initial convolutional neural network based on the first pixels. The initial water body change detection map is updated based on the target changes pixels and the target non-change pixels to obtain a target water body change detection map.

[0133] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0134] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0135] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting the range of a flood disaster water body, characterized in that: include: Obtain the neighborhood logarithmic ratio difference map of the two-phase SAR images before and after the flood disaster; Performing fuzzy clustering segmentation processing on the neighborhood logarithmic ratio difference map to obtain an initial water body change detection map; wherein the initial water body change detection map includes intermediate pixels, first pixels and high-confidence non-changed pixels, the intermediate pixels include pixels between changed and non-changed, and the first pixels include changed pixels and non-changed pixels; Using a preset network classifier to perform prediction processing on the intermediate pixels to obtain target changed pixels and target non-changed pixels; wherein the preset network classifier includes a classifier obtained by training an initial convolutional neural network initial classifier based on the first pixel; Based on the target changed pixels and the target non-changed pixels, the initial water body change detection map is updated to obtain a target water body change detection map; The fuzzy clustering segmentation process is performed on the neighborhood logarithmic ratio difference map to obtain an initial water body change detection map, including: Performing fuzzy clustering segmentation processing on the neighborhood logarithmic ratio difference map to obtain Q clusters and the weight of the cluster center of each cluster; Sorting the Q clusters according to the size relationship between the weights to obtain Q sorted clusters and a total number NUM of pixels included in the Q sorted clusters; According to the preset category pixel determination conditions, Num is determined from the total number of pixels NUM. CP Number of changed pixels, Num IP The number of intermediate pixels and Num UP non-changing pixels; Obtaining high confidence non-changing pixels in the neighborhood log ratio difference map; Determined by the Num CP The number of changed pixels, the Num IP The Num intermediate pixels UP An initial water body change detection map composed of non-changing pixels and the high-confidence non-changing pixels; According to the preset category pixel determination condition, Num is determined from the total number of pixels NUM. CP Number of changed pixels, Num IP The number of intermediate pixels and Num UP non-changing pixels, including: Set the number of pixels contained in each of the Q sorted clusters to be Num1, Num2, ..., Num Q ; Among them, Num1>Num2>…>Num Q , Num1+Num2+…+Num Q =NUM; When Num1+…+Num a-1 <Num CP-max And Num1+…+Num a-1 +Num a >Num CP-max When Num is determined, the number of pixels that change CP =Num1+…+Num a-1 And the category t included in the changed pixel c =a-1; where a is an integer greater than 1 and less than Q, Num CP-max Num is the preset upper limit of the number of pixels changed. CP-max =Num×T low , T low is the preset threshold value of the number of pixels of the change category, Num is the number of pixels of the change in the difference enhancement map of the SAR images before and after the flood disaster, t c ∈[1,Q]; when and When And the category t contained in the intermediate pixel i =t c +b-1; where b is an integer greater than 2 and less than Q, t i ∈[1,Qt c ], T high is the preset threshold value of the number of pixels of unknown category; when And m=min{2,…,Qt c -t i }, determine the number of non-changing pixels And the category t included in the non-changing pixel u =t i +t c +m; where t u ∈[1,Qt i -t c ].

2. The flood disaster water body range detection method according to claim 1 is characterized in that: The training process of the preset network classifier includes: The SAR images of the two time phases before and after the flood disaster are processed with sliding windows of preset step lengths to obtain M×N image patches of the SAR image of each time phase; Based on the initial water body change detection map, channel fusion processing is performed on the M×N image spots to obtain M×N dual-channel image spots; Performing sample labeling processing on the M×N dual-channel spots respectively to obtain change samples and non-change samples in the M×N training samples; wherein the sample labeling processing is used to characterize each of the dual-channel spots as a change sample, a non-change sample, an intermediate sample or a high-confidence sample; The inverse of the sum of the changed samples and the non-changed samples is used as the loss function weight of the initial convolutional neural network, and the changed samples and the non-changed samples are used to train the initial classifier of the initial convolutional neural network to obtain a preset network classifier.

3. The flood disaster water body range detection method according to claim 1 is characterized in that: The T low is the preset threshold value of the number of pixels of the change class and the T high is a preset threshold value of the number of pixels of unknown category, and its determination process includes: Obtaining a difference enhancement map of the two-phase SAR images before and after the flood disaster and a histogram of the difference enhancement map; Performing triangular threshold processing on the histogram to obtain a first sub-two-dimensional triangular graph and a second sub-two-dimensional triangular graph representing the relationship between the gray level and the frequency of the histogram; Obtain a first line segment between a first frequency of a first gray level and its perpendicular foot in the first sub-two-dimensional triangular graph, and a second line segment between a second frequency of a second gray level and its perpendicular foot in the second sub-two-dimensional triangular graph; respectively determining a first target gray level corresponding to when the first line segment is a maximum value and a second target gray level corresponding to when the second line segment is a maximum value; Determine the number of pixels of the changed category and the number of pixels of the unknown category in the difference enhancement map based on the first target grayscale level and the second target grayscale level; Based on the number of pixels in the change category and the number of pixels in the category to be determined, the threshold for limiting the number of pixels in the change category is set to T low And the threshold limiting the number of pixels of the unknown category is set to T high .

4. The flood disaster water body range detection method according to claim 3 is characterized in that: The step of obtaining the difference enhancement map of the SAR images in two phases before and after the flood disaster comprises: Respectively obtaining a logarithmic ratio difference map and a logarithmic likelihood ratio difference map of the two-phase SAR images before and after the flood disaster; Based on the logarithmic ratio difference map and the logarithmic likelihood ratio difference map, a difference enhancement map of the SAR images in two phases before and after the flood disaster is obtained.

5. The flood disaster water body range detection method according to claim 1 is characterized in that: The obtaining of high confidence non-changing pixels in the neighborhood logarithmic ratio difference map comprises: Remove Num from the neighborhood logarithmic ratio difference map CP The number of changed pixels, the Num IP The Num intermediate pixels UP The remaining pixels other than the non-changed pixels are determined as high-confidence non-changed pixels in the neighborhood logarithmic ratio difference map.

6. A flood disaster water body range detection device, characterized in that: include: The acquisition module is used to obtain the neighborhood logarithmic ratio difference map of the two-phase SAR images before and after the flood disaster; A segmentation module is used to perform fuzzy clustering segmentation processing on the neighborhood logarithmic ratio difference map to obtain an initial water body change detection map; wherein the initial water body change detection map includes intermediate pixels, first pixels and high-confidence non-changed pixels, the intermediate pixels include pixels between changed and non-changed, and the first pixels include changed pixels and non-changed pixels; A prediction module, used to perform prediction processing on the intermediate pixels using a preset network classifier to obtain target changed pixels and target non-changed pixels; wherein the preset network classifier includes a classifier obtained by training an initial convolutional neural network based on the first pixel. An updating module, configured to update the initial water body change detection map based on the target changed pixels and the target non-changed pixels to obtain a target water body change detection map; Wherein, the segmentation module is specifically used for: Performing fuzzy clustering segmentation processing on the neighborhood logarithmic ratio difference map to obtain Q clusters and the weight of the cluster center of each cluster; Sorting the Q clusters according to the size relationship between the weights to obtain Q sorted clusters and a total number NUM of pixels included in the Q sorted clusters; According to the preset category pixel determination conditions, Num is determined from the total number of pixels NUM. CP Number of changed pixels, Num IP The number of intermediate pixels and Num UP non-changing pixels; Obtaining high confidence non-changing pixels in the neighborhood log ratio difference map; Determined by the Num CP The number of changed pixels, the Num IP The Num intermediate pixels UP An initial water body change detection map composed of non-changing pixels and the high-confidence non-changing pixels; The segmentation module is also specifically used for: Set the number of pixels contained in each of the Q sorted clusters to be Num1, Num2, ..., Num Q ; Among them, Num1>Num2>…>Num Q , Num1+Num2+…+Num Q =NUM; When Num1+…+Num a-1 <Num CP-max And Num1+…+Num a-1 +Num a >Num CP-max When Num is determined, the number of pixels that change CP =Num1+…+Num a-1 And the category t included in the changed pixel c =a-1; where a is an integer greater than 1 and less than Q, Num CP-max Num is the preset upper limit of the number of pixels changed. CP-max =Num×T low , T low is the preset threshold value of the number of pixels of the change category, Num is the number of pixels of the change in the difference enhancement map of the SAR images before and after the flood disaster, t c ∈[1,Q]; when and When And the category t contained in the intermediate pixel i =t c +b-1; where b is an integer greater than 2 and less than Q, t i ∈[1,Qt c ], T high is the preset threshold value of the number of pixels of unknown category; when And m=min{2,…,Qt c -t i }, determine the number of non-changing pixels And the category t included in the non-changing pixel u =t i +t c +m; where t u ∈[1,Qt i -t c ].

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the flood disaster water body range detection method as claimed in any one of claims 1 to 5 are implemented.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the flood disaster water body range detection method as claimed in any one of claims 1 to 5 are implemented.

Images