SAR image deep learning flood mapping method and system guided by geoscience knowledge

Abstract

The invention discloses an SAR image deep learning flood mapping method and system guided by geoscience knowledge. The method comprises the following steps: respectively acquiring SAR images before and after flood in an application area, and preprocessing the SAR images; multi-scale ground feature extraction is carried out on the SAR images before and after the flood; extracting initial multi-scale features, converting the extracted initial multi-scale features from a spatial domain to a frequency domain, performing global relevance modeling and filtering processing in the frequency domain, recovering the frequency domain to the spatial domain, quantifying differences between SAR images before and after flood, and generating an initial flood probability map; learning and extracting high-quality multi-scale difference features from the multi-scale ground feature features before and after the flood, and connecting the generated multi-scale difference features in series to obtain a comprehensive difference feature set; and taking the difference feature set as input to generate a final binarized flood map. The method is high in calculation efficiency, high in operability, high in flood drawing accuracy and automation degree, easy to implement and high in expandability.

Description

Technical Field

[0001] The present invention belongs to the field of remote sensing image processing, and in particular relates to a SAR image deep learning flood mapping method and system guided by geological knowledge. Background Art

[0002] Timely and accurate flood mapping is of great significance to the development of flood prevention and disaster relief work. Due to the advantages of remote sensing technology, such as timeliness, high efficiency, wide monitoring range and low cost, the current mapping method based on remote sensing images has gradually developed into the mainstream way of flood mapping. In particular, SAR images have the characteristics of all-day and all-weather, and are not affected by climate factors such as cloud cover and rainfall. Therefore, SAR images are a major data source for current remote sensing flood mapping.

[0003] However, due to the small amount of band information and serious noise contained in SAR images, the traditional flood mapping methods based on SAR images often have poor mapping accuracy, obvious flood leakage detection and false detection phenomena, and a large degree of manual intervention in the mapping process. With the continuous breakthroughs in various related fields of deep learning technology, the application of deep learning in the field of remote sensing image processing is becoming more and more extensive. In particular, convolutional neural network models such as ResNet have powerful feature extraction capabilities and can extract deep semantic information of objects in remote sensing images, thus significantly improving the accuracy and reliability of remote sensing image interpretation. Therefore, in this context, it is very necessary to explore how to effectively utilize existing geological knowledge, combine the advantages of deep learning models in image deep feature modeling and extraction, and propose more effective SAR image flood mapping methods to improve the accuracy and reliability of its mapping. Summary of the invention

[0004] The purpose of the present invention is to provide a SAR image deep learning flood mapping method and system guided by geological knowledge in view of the above-mentioned deficiencies in the prior art.

[0005] In order to solve the above problems, the present invention adopts the following technical solutions:

[0006] A SAR image deep learning flood mapping method guided by geoscience knowledge includes the following steps:

[0007] Step 1: Obtain SAR images of the application area before and after the flood, and preprocess them;

[0008] Step 2: Use a deep convolutional neural network model to extract multi-scale features of objects from SAR images before and after the flood;

[0009] Step 3: The multi-scale features of the SAR images before and after the flood extracted in step 2 are converted from the spatial domain to the frequency domain, and the global correlation of the low-frequency signals is modeled using the self-attention mechanism in the frequency domain, and the high-frequency signals are filtered. After the global correlation modeling and filtering, the features are restored from the frequency domain to the spatial domain to obtain enhanced multi-scale features of the features before and after the flood.

[0010] Step 4: Combined with existing geological knowledge, a dual-temporal image difference measurement model is used to quantify the differences between SAR images before and after the flood, and to generate an initial flood probability map P 0 ;

[0011] Step 5: construct a flood probability map guided difference feature extraction module, and use the enhanced multi-scale ground feature features before and after the flood obtained in step 3 as input to extract multi-scale high-quality difference features based on the constructed flood probability map guided difference feature extraction module;

[0012] Step 6: Integrate the multi-scale high-quality difference features generated in step 5 in series to obtain the difference feature set for the final flood mapping. And the difference feature set As input, a multi-layer perceptron (MLP) is used as a classifier to generate the final binary flood map.

[0013] Furthermore, the fast Fourier transform used in step 3 converts the multi-scale ground feature of the SAR images before and after the flood from the spatial domain to the frequency domain to obtain processed features, and uses the inverse fast Fourier transform to restore the processed features from the frequency domain to the spatial domain.

[0014] Furthermore, the process of using the self-attention mechanism in the frequency domain to model the global correlation of low-frequency signals in step 3 is as follows:

[0015] Use three different linear transformations to transform the low frequency signal into three different matrices X Q , X K , X V ;

[0016] Calculate the global attention weight Attention(X Q ,X K ,X V )

[0017] The original low-frequency signal is combined with the global attention weight Attention(X Q ,X K ,X V ) are multiplied to obtain a new low-frequency signal containing global correlation information, thus completing the modeling of global correlation.

[0018] Furthermore, the global attention weight Attention(X Q ,X K ,X V ) is calculated as follows:

[0019]

[0020] Where d represents X Q , X K , X V , and softmax(·) is a normalized exponential function.

[0021] Furthermore, in step 4, a neighborhood ratio operator is used to measure the pixel-by-pixel difference between the SAR images before and after the flood, thereby generating a difference map M; and an initial flood probability map P is generated from the difference map M. 0 :

[0022]

[0023] Among them, M max and M min Represent the maximum and minimum values ​​of all pixels in the difference map M respectively.

[0024] Furthermore, the workflow of constructing the difference feature extraction module in step 5 is as follows:

[0025] The initial difference feature ΔF is generated by direct matrix subtraction of the features before and after the flood at scale s extracted in step 3;

[0026] Use the flood probability map P generated at the previous scale s-1 s-1 As a guide map to guide the extraction of difference features, that is, the flood probability map P s-1 Resample to the same spatial resolution as the initial difference feature ΔF, and integrate it with ΔF in series to form a new feature set ΔF′;

[0027] Based on the new feature set ΔF′, the self-attention mechanism from the visual transformer is used to extract the global context-dependent features of the flood, and the extracted dependent features are integrated with the feature set ΔF′ in a series manner to obtain the feature set ΔF″;

[0028] The hybrid attention model is used to refine the feature set ΔF″ to generate the final difference features at scale s

[0029] Furthermore, the difference features at each scale are generated under the guidance of the flood probability map of the previous scale. For the first scale, the initial flood probability map P 0Extract differential features under the guidance of

[0030] Flood probability map P at the previous scale s-1 s-1 It can be generated by the following formula:

[0031]

[0032] in, represents the final difference feature set generated by the difference feature extraction module at scale s-1; N represents The number of differential features contained in ; express The nth difference feature in , n = 1, 2, ..., N.

[0033] On the other hand, the present invention provides a SAR image deep learning flood mapping system guided by geological knowledge, comprising:

[0034] SAR image acquisition module: it is used to obtain SAR images of the application area before and after the flood, and pre-process them;

[0035] Multi-scale ground feature extraction module: It is used to extract multi-scale ground feature features from SAR images before and after the flood using a deep convolutional neural network model;

[0036] Multi-scale surface feature enhancement module: It is used to convert the multi-scale surface feature of the extracted SAR images before and after the flood from the spatial domain to the frequency domain, use the self-attention mechanism in the frequency domain to model the global correlation of low-frequency signals, and filter the high-frequency signals; after global correlation modeling and filtering, restore the features from the frequency domain to the spatial domain to obtain enhanced multi-scale surface feature features before and after the flood;

[0037] Initial flood probability map generation module: It is used to combine existing geological knowledge, use a dual-temporal image difference measurement model to quantify the difference between SAR images before and after the flood, and generate an initial flood probability map P 0 ;

[0038] Difference feature extraction module: It is used to construct a difference feature extraction module guided by a flood probability map, and takes the obtained enhanced multi-scale ground feature features before and after the flood as input, and extracts multi-scale high-quality difference features based on the difference feature extraction module guided by the constructed flood probability map;

[0039] Flood map generation module: It is used to integrate the multi-scale high-quality difference features generated in series to obtain the difference feature set for the final flood mapping. And the difference feature set As input, a multi-layer perceptron (MLP) is used as a classifier to generate the final binary flood map.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] (1) The method proposed in the present invention effectively utilizes the advantages of the convolutional neural network model in extracting deep features of images, and uses existing geological knowledge to guide the learning and extraction of multi-scale difference features. Therefore, high-quality multi-scale difference features between SAR images before and after floods can be obtained, thereby effectively improving the accuracy and reliability of flood mapping.

[0042] (2) The SAR image flood mapping method based on deep learning proposed in the present invention not only has high computational efficiency and strong operability, but also has high accuracy and automation of flood mapping, is easy to implement and has strong scalability. Therefore, the present invention has high practical value. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] 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.

[0044] Figure 1 is a flow chart of an embodiment of the present invention.

[0045] Figure 2 Schematic diagram of some initial features extracted from post-flood images according to an embodiment of the present invention.

[0046] Figure 3 Schematic diagram of the enhanced features obtained in step 3 of the embodiment of the present invention.

[0047] Figure 4 Schematic diagram of flood probability generated at different scales according to an embodiment of the present invention.

[0048] Figure 5 The figure is a result diagram of the embodiment of the present invention and a comparison diagram with the existing commonly used method. DETAILED DESCRIPTION

[0049] 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.

[0050] It should be noted that, in the description of the embodiments of the present invention, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "include one..." do not exclude the existence of other identical elements in the process, method, article or device including the elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to the specific circumstances.

[0051] The terms "first", "second", etc. in this application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present application can be implemented in an order other than those illustrated or described here, and the objects distinguished by "first", "second", etc. are generally of one type, and the number of objects is not limited. For example, the first object can be one or more. In addition, "and / or" represents at least one of the connected objects, and the character " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0052] Example 1

[0053] like Figure 1 , 2 As shown, the present invention takes flood mapping of a certain area of ​​Poyang Lake in China as an example to implement the method proposed by the present invention. The process of the embodiment of the present invention includes the following steps:

[0054] Step 1: Taking a certain area of ​​Poyang Lake in China as an example, obtain SAR images of the area before and after the flood, and use the image preprocessing tools of ENVI software to perform necessary preprocessing such as denoising, calibration, and registration.

[0055] Step 2: Use the current mature convolutional neural network model ResNet to extract the initial multi-scale ground object details and semantic features of the SAR images before and after the flood. These features include the feature maps generated by the ResNet network at different stages. Figure 2 Some of the initial features extracted from the post-flood imagery are shown (dark colors represent water areas, and gray-white represents non-water areas).

[0056] Step 3: First, the multi-scale ground object details and semantic features of the SAR images before and after the flood extracted in step 2 are converted from the spatial domain to the frequency domain using the fast Fourier transform algorithm. Then, the global correlation of low-frequency signals is modeled using the self-attention mechanism in the frequency domain. The specific process is as follows:

[0057] First, use three different linear transformations to transform the low-frequency signal into three different matrices X Q , X K , X V Then, the global attention weight Attention(X Q ,X K ,X V ), where d represents X Q , X K , X V , and softmax(·) is a normalized exponential function.

[0058]

[0059] Finally, by combining the original low-frequency signal with the global attention weight Attention(X Q ,X K ,X V ) are multiplied to obtain a new low-frequency signal containing global correlation information, thus completing the modeling of global correlation.

[0060] At the same time, an adaptive Gaussian filter is used in the frequency domain to filter the high-frequency signal. After global correlation modeling and filtering, the inverse fast Fourier transform algorithm is used to transform the features from the frequency domain back to the spatial domain to obtain enhanced multi-scale features of land objects before and after the flood. Figure 3 Some of the enhanced features obtained in this step are shown ( Figure 3 The five features in Figure 2 Enhanced features of the 5 initial features in ). Figure 3 and Figure 2 It can be seen that compared with the initial features, in the enhanced feature map, the difference between water bodies and non-water bodies is more significant, and the noise phenomenon is also weakened, which will help improve the accuracy of flood detection and mapping.

[0061] Step 4: Combined with existing geological knowledge, the existing dual-temporal image difference measurement model is used to quantify the difference between SAR images before and after the flood, and generate the initial flood probability map P 0 ,In the flood probability map, the larger the pixel value, the greater the ,probability that the pixel is a flood pixel.

[0062] Specifically, the neighborhood-based ratio operator (NRoperator) is used here to measure the pixel-by-pixel difference between the SAR images before and after the flood, and then generate the difference map M. The initial flood probability map P is generated from the difference map M using the following formula (2): 0 .

[0063]

[0064] In formula (2), M max and M min Represent the maximum and minimum values ​​of all pixels in the difference map M respectively.

[0065] Step 5: Construct a flood probability map-guided difference feature extraction module, and use the enhanced multi-scale features before and after the flood obtained in step 3 as input to extract high-quality multi-scale difference features using the constructed module.

[0066] Specifically, the workflow of the difference feature extraction module constructed here is as follows: Taking scale s as an example, in the difference feature extraction module, the initial difference feature ΔF is first generated by direct matrix subtraction of the features before and after the flood at scale s extracted in step 3. Then, the flood probability map P generated at the previous scale s-1 is used s-1 As a guide map to guide the extraction of difference features, that is, the flood probability map P s-1 Resample to the same spatial resolution as the initial difference feature ΔF, and integrate it with ΔF in series to form a new feature set ΔF′. Subsequently, based on the new feature set ΔF′, the self-attention mechanism from the visual transformer is used to extract the global context-dependent features of the flood, and the extracted dependent features are also integrated with the feature set ΔF′ in series to obtain the feature set ΔF″. Finally, the hybrid attention model CBAM (Convolutional Block Attention Module) is used to refine the feature set ΔF″ to generate the final difference feature at scale s.

[0067] The flood features before and after the flood at different scales extracted in step 3 are processed according to the above process to obtain high-quality difference features at different scales. It should be noted that the difference features at each scale are generated under the guidance of the flood probability map of the previous scale, and for the first scale, the initial flood probability map P 0 (generated in step 4) to extract the difference features. Taking scale s as an example, the flood probability map P at the previous scale s-1 s-1 It can be generated by formula (3).

[0068]

[0069] In the above formula, represents the final difference feature set generated by the difference feature extraction module at scale s-1; N represents The number of differential features contained in ; express The nth difference feature in , n = 1, 2, ..., N. Figure 4 The flood probability map and the initial flood probability map generated by step-by-step iteration when the scale parameter s is 1, 2, 3, and 4 respectively are shown. Figure 4 As can be seen from the figure, as the number of iterations increases (s increases), the generated flood probability map tends to be smoother, the noise is significantly reduced, and the white area (indicating a higher probability of flood occurrence) is Figure 5 The spatial distribution of flood areas in the ground reference data is more consistent. Therefore, under the guidance of the flood probability map, the proposed module can more easily focus on the areas where floods actually occur, and learn high-quality difference features with higher discrimination between flooded areas and non-flooded areas, thereby enhancing the accuracy of flood detection and mapping.

[0070] Step 6: Integrate the multi-scale high-quality difference features generated in step 5 in series to obtain a difference feature set that can be used for final flood mapping. And the difference feature set As input, a multi-layer perceptron (MLP) is used as a classifier to generate the final binary flood map.

[0071] Figure 5 The results of the embodiment of the present invention (in a certain area of ​​Poyang Lake in China) are shown, including a comparison chart of flood mapping results of the proposed new method and existing commonly used methods (PCAKmeans, MRF_FCM, GarborPCANet, NR_ELM). Figure 5 It can be seen from the figure that compared with the existing commonly used methods, the method of the present invention can more accurately identify the flood area, especially can effectively reduce the noise phenomenon in the generated flood map (such as Figure 5 The area marked by the red ellipse in the figure). From the perspective of the quantitative accuracy index OA, the overall mapping accuracy OA of the method of the present invention is significantly higher than that of other methods (1.13% to 11.18% higher than that of other existing methods). The results of the above examples prove the superiority of the method of the present invention.

[0072] Example 2

[0073] This embodiment provides a SAR image deep learning flood mapping system guided by geological knowledge, including:

[0074] SAR image acquisition module: it is used to obtain SAR images of the application area before and after the flood, and pre-process them;

[0075] Multi-scale ground feature extraction module: It is used to extract multi-scale ground feature features from SAR images before and after the flood using a deep convolutional neural network model;

[0076] Multi-scale surface feature enhancement module: It is used to convert the multi-scale surface feature of the extracted SAR images before and after the flood from the spatial domain to the frequency domain, use the self-attention mechanism in the frequency domain to model the global correlation of low-frequency signals, and filter the high-frequency signals; after global correlation modeling and filtering, restore the features from the frequency domain to the spatial domain to obtain enhanced multi-scale surface feature features before and after the flood;

[0077] Initial flood probability map generation module: It is used to combine existing geological knowledge, use a dual-temporal image difference measurement model to quantify the difference between SAR images before and after the flood, and generate an initial flood probability map P 0 ;

[0078] Difference feature extraction module: It is used to construct a difference feature extraction module guided by a flood probability map, and takes the obtained enhanced multi-scale ground feature features before and after the flood as input, and extracts multi-scale high-quality difference features based on the difference feature extraction module guided by the constructed flood probability map;

[0079] Flood map generation module: It is used to integrate the generated multi-scale high-quality difference features in series to obtain the difference feature set for the final flood mapping And the difference feature set As input, a multi-layer perceptron (MLP) is used as a classifier to generate the final binary flood map.

[0080] 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.

[0081] 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 SAR image deep learning flood mapping method guided by geological knowledge, characterized in that: The following steps are involved: Step 1: Obtain SAR images of the application area before and after the flood, and preprocess them; Step 2: Use a deep convolutional neural network model to extract multi-scale features of objects from SAR images before and after the flood; Step 3: The multi-scale features of the SAR images before and after the flood extracted in step 2 are converted from the spatial domain to the frequency domain, and the global correlation of the low-frequency signals is modeled using the self-attention mechanism in the frequency domain, and the high-frequency signals are filtered. After the global correlation modeling and filtering, the features are restored from the frequency domain to the spatial domain to obtain enhanced multi-scale features of the features before and after the flood. Step 4: Combining existing geological knowledge, a dual-temporal image difference measurement model is used to quantify the differences between SAR images before and after the flood, and to generate an initial flood probability map P0; Step 5: construct a flood probability map guided difference feature extraction module, and use the enhanced multi-scale ground feature features before and after the flood obtained in step 3 as input to extract multi-scale high-quality difference features based on the constructed flood probability map guided difference feature extraction module; Step 6: Integrate the multi-scale high-quality difference features generated in step 5 in series to obtain the difference feature set for the final flood mapping. And the difference feature set As input, a multi-layer perceptron (MLP) is used as a classifier to generate the final binary flood map.

2. The method for flood mapping using SAR images guided by geological knowledge as claimed in claim 1, characterized in that: The fast Fourier transform used in step 3 converts the multi-scale ground feature features of the SAR images before and after the flood from the spatial domain to the frequency domain to obtain processed features, and uses the inverse fast Fourier transform to restore the processed features from the frequency domain to the spatial domain.

3. The method for flood mapping using SAR images guided by geological knowledge as claimed in claim 1, characterized in that: The process of using the self-attention mechanism in the frequency domain to model the global correlation of low-frequency signals in step 3 is as follows: Use three different linear transformations to transform the low frequency signal into three different matrices X Q , X K , X V ; Calculate the global attention weight Attention(X Q ,X K ,X V ) The original low-frequency signal is combined with the global attention weight Attention(X Q ,X K ,X V ) are multiplied to obtain a new low-frequency signal containing global correlation information, thus completing the modeling of global correlation.

4. The method for flood mapping using SAR images guided by geological knowledge as claimed in claim 3, characterized in that: Global attention weight Attention(X Q ,X K ,X V ) is calculated as follows: Where d represents X Q , X K , X V , and softmax(·) is a normalized exponential function.

5. The method for flood mapping using SAR images guided by geological knowledge as claimed in claim 1, characterized in that: In step 4, a neighborhood ratio operator is used to measure the pixel-by-pixel difference between the SAR images before and after the flood, thereby generating a difference map M; and an initial flood probability map P0 is generated from the difference map M: Among them, M max and M min They represent the maximum and minimum values ​​of all pixels in the difference map M respectively.

6. The method for flood mapping using SAR images guided by geological knowledge as claimed in claim 1, characterized in that: The workflow of constructing the difference feature extraction module in step 5 is as follows: The initial difference feature ΔF is generated by direct matrix subtraction of the features before and after the flood at scale s extracted in step 3; Use the flood probability map P generated at the previous scale s-1 s-1 As a guide map to guide the extraction of difference features, that is, the flood probability map P s-1 Resample to the same spatial resolution as the initial difference feature ΔF, and integrate it with ΔF in series to form a new feature set ΔF′; Based on the new feature set ΔF′, the self-attention mechanism from the visual transformer is used to extract the global context-dependent features of the flood, and the extracted dependent features are integrated with the feature set ΔF′ in a series manner to obtain the feature set ΔF″; The hybrid attention model is used to refine the feature set ΔF″ to generate the final difference features at scale s 7. The method for flood mapping using SAR images guided by geological knowledge as claimed in claim 6, characterized in that: The difference features at each scale are generated under the guidance of the flood probability map of the previous scale. For the first scale, the difference features are extracted under the guidance of the initial flood probability map P0; Flood probability map P at the previous scale s-1 s-1 It can be generated by the following formula: in, represents the final difference feature set generated by the difference feature extraction module at scale s-1; N represents The number of differential features contained in ; express The nth difference feature in , n = 1, 2, ..., N.

8. A SAR image deep learning flood mapping system guided by geological knowledge, characterized by: include: SAR image acquisition module: it is used to obtain SAR images of the application area before and after the flood, and pre-process them; Multi-scale ground feature extraction module: It is used to extract multi-scale ground feature features from SAR images before and after the flood using a deep convolutional neural network model; Multi-scale surface feature enhancement module: It is used to convert the multi-scale surface feature of the extracted SAR images before and after the flood from the spatial domain to the frequency domain, use the self-attention mechanism in the frequency domain to model the global correlation of low-frequency signals, and filter the high-frequency signals; after global correlation modeling and filtering, restore the features from the frequency domain to the spatial domain to obtain enhanced multi-scale surface feature features before and after the flood; Initial flood probability map generation module: It is used to combine existing geological knowledge, use a dual-temporal image difference measurement model to quantify the difference between SAR images before and after the flood, and generate an initial flood probability map P0; Difference feature extraction module: It is used to construct a difference feature extraction module guided by a flood probability map, and takes the obtained enhanced multi-scale ground feature features before and after the flood as input, and extracts multi-scale high-quality difference features based on the difference feature extraction module guided by the constructed flood probability map; Flood map generation module: It is used to integrate the generated multi-scale high-quality difference features in series to obtain the difference feature set for the final flood mapping And the difference feature set As input, a multi-layer perceptron MLP is used as a classifier to generate the final binary flood map; The geological knowledge-guided SAR image deep learning flood mapping system is used to execute the steps in the geological knowledge-guided SAR image deep learning flood mapping method described in any one of claims 1-7.

Images