Wide-area landslide identification method and system based on phase gradient stacking and deep learning

Abstract

The invention provides a wide-area landslide identification method and system based on phase gradient stacking and deep learning, and the method comprises the steps: obtaining an SAR image set covering a research region at different time periods, and obtaining a differential interferogram through a D-InSAR method; an IPGS method is adopted to obtain a gradient stacking graph of the differential interferogram, and the gradient stacking graph is cut to obtain a cut picture; performing landslide sample labeling on the cut picture through the auxiliary data and the optical image to obtain a labeled sample; training the improved YOLOv7 model by adopting a labeling sample to obtain a landslide target recognition model; and inputting the to-be-identified image of the research area into the landslide target identification model, and outputting a landslide identification result of the research area. According to the invention, the advantage of rapidly and accurately acquiring the deformation gradient signal by the improved phase gradient stacking method is combined with the deep learning model to overcome the defects in the prior art, so that the efficient and accurate wide-area potential landslide automatic identification method is provided.

Description

Technical Field

[0001] The present invention relates to the field of radar interferometry technology, and in particular to a wide-area landslide identification method and system based on phase gradient stacking and deep learning. Background Art

[0002] Landslides are a widespread and fatal type of geological disaster that can not only destroy infrastructure such as roads and bridges, but also cause significant casualties. With the gradual warming of the global climate, the continuous growth of the population, and the continuous deterioration of the environment, the frequency and intensity of landslide disasters have greatly increased. my country has a wide distribution of mountains and hills, accounting for about 65% of the country's total area. This unique geological environment has led to frequent geological disasters in my country. Therefore, conducting research on potential landslide hazards is of great significance for reducing casualties, environmental protection, and sustainable development.

[0003] Synthetic Aperture Radar (SAR) is a ground observation technology with the advantages of all-weather, all-day operation, wide coverage, and high spatial resolution. Interferometric Synthetic Aperture Radar (InSAR) technology can capture sub-centimeter ground deformation hundreds of kilometers away. After two or three decades of development, time-series InSAR technology has matured and is widely used in the field of landslide identification and monitoring. It provides important basic support for the monitoring, mechanism analysis, and physical inversion research of landslide disasters. However, there are certain limitations when using time-series InSAR technology for wide-area landslide identification, especially in complex mountainous areas. Its time-series results are often affected by severe atmospheric delays and potential phase unwrapping errors. In addition, since the time-series results need to be unwrapped and subsequently processed, the amount of calculation is large. Therefore, atmospheric delay effects, unwrapping errors, and massive calculations have seriously reduced the ability of InSAR technology to detect landslides in high mountainous areas. With the advent of the era of massive SAR data, the computational load of InSAR has increased exponentially, and a fast and accurate InSAR landslide identification technology is urgently needed. Therefore, how to effectively realize large-scale InSAR data processing and quickly and automatically interpret the results is an urgent problem that needs to be solved. Summary of the invention

[0004] The present invention provides a wide-area landslide identification method and system based on phase gradient stacking and deep learning, so as to solve the defects existing in the landslide disaster identification method in the prior art.

[0005] In a first aspect, the present invention provides a wide-area landslide identification method based on phase gradient stacking and deep learning, comprising: A set of SAR images covering the study area at different time periods was obtained, and the differential interferogram was obtained using the D-InSAR method; The IPGS method is used to obtain a gradient stacking image of the differential interference image, and the gradient stacking image is cropped to obtain a cropped image; Using auxiliary data and optical images, annotating the cropped image with landslide samples to obtain annotated samples; The labeled samples are used to train the improved YOLOv7 model to obtain a landslide target recognition model; The image to be identified in the study area is input into the landslide target identification model, and the landslide identification result of the study area is output.

[0006] According to a wide-area landslide identification method based on phase gradient stacking and deep learning provided by the present invention, an IPGS method is used to obtain a gradient stacking image of the differential interferogram, and the gradient stacking image is cropped to obtain a cropped image, including: Calculating the phase gradient in any direction of the differential interference pattern based on a preset step size; The phase gradients of each differential interferogram corresponding to any direction are stacked from the time dimension and median filtered; The phase stacking images in each direction are merged, and the merged results are gradient normalized; Converting the gradient normalization result to the geographic coordinate system to obtain the gradient stacking map; The gradient stacked image is cropped into an image of a preset size to obtain the cropped image.

[0007] According to a wide-area landslide identification method based on phase gradient stacking and deep learning provided by the present invention, landslide samples are annotated on the cropped image through auxiliary data and optical images to obtain annotated samples, including: The auxiliary data is cropped into a picture of a preset size to obtain cropped auxiliary data, and a slope map is obtained; The cropped image is annotated by using a preset annotation tool in combination with the optical image, the slope map and the cropped auxiliary data to obtain the annotated sample.

[0008] According to a wide-area landslide identification method based on phase gradient stacking and deep learning provided by the present invention, the improved YOLOv7 model is obtained by the following steps: Determining that the improved YOLOv7 model includes an input end, a backbone network, a head module, and a prediction layer; The backbone network includes a CBS module, an ELAN module and an MP module; The head module includes an SPPCSPC module, a CBS module, an ELAN module and an MP module, forming an FPN network structure; The prediction module includes three detection layers, so that the network detects objects of different sizes from three different receptive fields respectively; The CBAM attention mechanism is introduced into the backbone network to capture the correlation between features through adaptive learning channels and spatial attention weights.

[0009] According to a wide-area landslide identification method based on phase gradient stacking and deep learning provided by the present invention, the improved YOLOv7 model is trained using the labeled samples to obtain a landslide target identification model, including: Use the weights trained by the preset image dataset as the pre-trained weights of the network to determine the Batch Size and Epoch; The improved YOLOv7 model is trained using the labeled samples, and the model performance is evaluated using target detection performance indicators, and the landslide target recognition model is obtained after training convergence.

[0010] According to a wide-area landslide identification method based on phase gradient stacking and deep learning provided by the present invention, after the improved YOLOv7 model is trained using the labeled samples to obtain a landslide target identification model, the method further includes: If the study area is determined to be a snow mountain area, the snow cover normalized difference index is used to obtain the snow cover area of ​​the study area, and the snow mountain area is masked out from the gradient stacking map.

[0011] In a second aspect, the present invention further provides a wide-area landslide identification system based on phase gradient stacking and deep learning, comprising: The acquisition module is used to obtain a set of SAR images covering different periods of the study area and obtain differential interferograms using the D-InSAR method; A stacking module, used for obtaining a gradient stacking image of the differential interference image by using an IPGS method, and cropping the gradient stacking image to obtain a cropped image; A labeling module, used for labeling the landslide samples on the cropped images by using auxiliary data and optical images to obtain labeled samples; A training module, used to train the improved YOLOv7 model using the labeled samples to obtain a landslide target recognition model; The identification module is used to input the image to be identified in the study area into the landslide target identification model and output the landslide identification result in the study area.

[0012] In a third aspect, 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 wide-area landslide identification method based on phase gradient stacking and deep learning as described above is implemented.

[0013] In a fourth aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the wide-area landslide identification methods based on phase gradient stacking and deep learning as described above.

[0014] In a fifth aspect, the present invention further provides a computer program product, comprising a computer program, which, when executed by a processor, implements any of the wide-area landslide identification methods based on phase gradient stacking and deep learning as described above.

[0015] The wide-area landslide identification method and system based on phase gradient stacking and deep learning provided by the present invention obtains the gradient stacking results of the study area by adopting the IPGS method, which greatly reduces the amount of calculation compared with the traditional InSAR method. The results combined with deep learning can automatically identify potential landslides in a timely manner. The obtained gradient stacking results and corresponding auxiliary data (slope) are used as input data; at the same time, the present invention adds an input data channel in the YOLOv7 model to adapt to multiple data inputs; in addition, the CBAM attention mechanism is added to the network model to improve the performance of the network.

[0016] The present invention uses phase gradient stacking results and auxiliary data as inputs of the YOLOv7 model, which reduces the workload on the one hand and improves the accuracy of potential landslide identification on the other hand. The introduction of the attention mechanism enhances the recognition ability of the model, thereby improving the performance of the network model. The IPGS method not only effectively suppresses the influence of atmospheric delay noise, but also can quickly obtain gradient stacking results in a wide range of research areas, so that a census of geological disasters can be carried out in a timely manner. The obtained gradient stacking results and auxiliary data (slope) are used as input data. Adding input data channels in the YOLOv7 model to adapt to multiple data inputs; adding the CBAM attention mechanism in the network model improves the performance of the network. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. 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.

[0018] Figure 1 It is one of the flow diagrams of the wide-area landslide identification method based on phase gradient stacking and deep learning provided by the present invention; Figure 2 This is the second flow chart of the wide-area landslide identification method based on phase gradient stacking and deep learning provided by the present invention; Figure 3It is a YOLOv7 network model structure diagram provided by the present invention; Figure 4 It is a schematic diagram of the structure of the CBAM attention mechanism provided by the present invention; Figure 5 It is a potential landslide identification result map of a certain province provided by the present invention; Figure 6 It is a structural schematic diagram of a wide-area landslide identification system based on phase gradient stacking and deep learning provided by the present invention; Figure 7 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0019] 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.

[0020] Figure 1 is one of the flow charts of the wide-area landslide identification method based on phase gradient stacking and deep learning provided by an embodiment of the present invention, such as Figure 1 As shown, including: Step 100: Obtain a set of SAR images covering the study area at different time periods, and obtain a differential interferogram using the D-InSAR method; Step 200: using the IPGS method to obtain a gradient stacking image of the differential interference image, and cropping the gradient stacking image to obtain a cropped image; Step 300: annotating the landslide samples on the cropped image using auxiliary data and optical images to obtain annotated samples; Step 400: using the labeled samples to train the improved YOLOv7 model to obtain a landslide target recognition model; Step 500: input the image to be identified in the study area into the landslide target identification model, and output the landslide identification result in the study area.

[0021] Specifically, the embodiment of the present invention adopts a wide-area potential landslide automatic identification method based on phase gradient stacking and deep learning, which supports wide-area landslide automatic identification, such as Figure 2 As shown, the implementation process includes: SAR images covering the study area at different time periods were obtained, and differential interferograms were obtained by conventional Differential Interferometric Synthetic Aperture Radar (D-InSAR) method. The improved phase gradient stacking (IPGS) method was used to obtain the gradient stacking map of the study area, and the map was cropped. With the assistance of auxiliary data and optical images, the cropped images were annotated with landslide samples. For the target detection model YOLOv7, a multi-channel input window was added and an attention mechanism was introduced into the backbone network to improve the performance of the network model in landslide recognition. The annotated images and the corresponding auxiliary images were input into the target recognition model for training and testing. The trained model was used to carry out wide-area potential landslide recognition.

[0022] Based on the above embodiment, step 200 includes: Calculating the phase gradient in any direction of the differential interference pattern based on a preset step size; The phase gradients of each differential interferogram corresponding to any direction are stacked from the time dimension and median filtered; The phase stacking images in each direction are merged, and the merged results are gradient normalized; Converting the gradient normalization result to the geographic coordinate system to obtain the gradient stacking map; The gradient stacked image is cropped into an image of a preset size to obtain the cropped image.

[0023] Specifically, an embodiment of the present invention calculates the phase gradient of an interference pattern in a certain direction with a preset step size s; stacks the phase gradient of each interference pattern corresponding to a certain direction in the time dimension and performs median filtering; merges and normalizes the phase stacking results in each direction; and converts the phase gradient stacking results to a geographic coordinate system to obtain a gradient stacking map.

[0024] In this embodiment, a certain province is taken as the study area. The province is located in the Yunnan-Guizhou Plateau, between 103°36′-109°35′ east longitude and 24°37′-29°13′ north latitude. It is about 595 kilometers long from east to west and about 509 kilometers from north to south, with a total area of ​​176,167 square kilometers. Due to the fragile geological environment and increasing human activities, according to the national geological disaster prevention and classification regulations, the province belongs to a geological disaster-prone area, of which 77% of the area belongs to a medium- and high-risk area for geological disasters. It is one of the provinces most severely affected by geological disasters in my country. The data source is the European Space Agency Sentinel-1 Sentinel 1 data covering the province, and a total of about 116 SAR satellite image data from January 2018 to February 2022 were obtained.

[0025] Furthermore, the IPGS method was used to obtain the gradient stacking map of the study area and the map was cropped. The phase gradient of a certain direction of the interference map was calculated with a preset step size s; the phase gradient of each interference map corresponding to a certain direction was stacked in the time dimension and median filtered; the phase stacking results of each direction were merged and the merged results were gradient normalized; the results were converted to the geographic coordinate system. Accordingly, the slope results corresponding to the gradient results were calculated using DEM data. Finally, the gradient stacking results were cropped into a 640×640 image.

[0026] Based on the above embodiment, step 300 includes: The auxiliary data is cropped into a picture of a preset size to obtain cropped auxiliary data, and a slope map is obtained; The cropped image is annotated by using a preset annotation tool in combination with the optical image, the slope map and the cropped auxiliary data to obtain the annotated sample.

[0027] Specifically, the embodiment of the present invention annotates landslide samples on a cropped portion of the image with the assistance of auxiliary data and optical images.

[0028] Since the occurrence of landslides is significantly correlated with the slope, the corresponding optical images and slope maps are considered when annotating potential landslides on the cropped gradient stacked images. The auxiliary data are also cropped into 640×640 images, in which the labelimg annotation tool is used to annotate potential landslides.

[0029] Based on the above embodiments, the improved YOLOv7 model constructed in the embodiments of the present invention adds a multi-channel input window to the target detection model YOLOv7 and introduces an attention mechanism in the backbone network to improve the performance of landslide recognition of the network model.

[0030] like Figure 3 As shown in the figure, the YOLOv7 model mainly consists of four parts, including the input, the backbone network, the head module and the prediction layer. The backbone network mainly consists of the CBS module, the ELAN module and the MP module. The head module mainly consists of the SPPCSPC, CBS, ELAN and MP modules, forming the FPN network structure. The prediction layer contains three detection layers, allowing the network to detect targets of different sizes from three different receptive fields. A multi-channel input window is added to introduce auxiliary data (slope data) to improve the accuracy of landslide identification. In addition, the CBAM attention mechanism is introduced on the basis of the backbone network. CBAM captures the correlation between features by adaptively learning channel and spatial attention weights, thereby improving the performance of image recognition tasks.

[0031] Since the occurrence of landslides has a certain correlation with slope, the introduction of slope data during network model training makes the network training have a certain direction, which significantly improves the accuracy of network identification of landslides and also reduces the reasoning time of the model.

[0032] The ELAN module achieves deep learning and convergence of the network by controlling the shortest and longest gradient paths. The MP structure combines two downsampling techniques, pooling and convolution, so that the network can choose the best downsampling method between the two. The role of SPPCSPC is to use the maximum pooling of four different scales for processing to achieve the fusion of information of different feature scales. Its main function is to perform multi-scale feature fusion after the backbone network feature extraction and pass these features to the prediction layer. The end of the head module consists of REP and CBM modules. REP adjusts the number of channels of different scale features of the backbone network, extracts features, and then uses CBM to predict bounding boxes and target categories. The prediction layer includes three detection layers, which enable the network to detect targets of different sizes from three receptive fields respectively.

[0033] For the application of the single-stage target detection model YOLOv7 in potential landslide identification, the present invention introduces the CBAM attention mechanism based on the backbone network, which enhances the network's ability to effectively extract deep features and effectively improves the network's ability to identify small landslides.

[0034] CBAM is an attention mechanism that combines channel attention and spatial attention modules. By adaptively assigning weights to different channels and spatial positions of the feature map, the convolutional neural network's attention to important features is enhanced, thereby improving the performance of the model. The CBAM module has a simple but powerful structure and is divided into two sub-modules: the channel attention module and the spatial attention module. The channel attention module extracts features through global average pooling and maximum pooling, and processes them through a shared multi-layer perceptron to generate a channel attention map. The spatial attention module processes the channel-weighted feature map and generates a spatial attention map through convolution operations. The main feature of CBAM is that it can pay attention to channel and spatial information at the same time. This combination of dual attention mechanisms enables the network to extract and utilize key information more accurately when processing complex visual tasks. The module structure diagram is shown below. Figure 4 shown.

[0035] Based on the above embodiment, step 400 includes: Use the weights trained by the preset image dataset as the pre-trained weights of the network to determine the Batch Size and Epoch; The improved YOLOv7 model is trained using the labeled samples, and the model performance is evaluated using target detection performance indicators, and the landslide target recognition model is obtained after training convergence.

[0036] Specifically, in the embodiment of the present invention, the weights trained with COCO data are used as the pre-trained weights of the network; Batch_size is set to 8; Epoch is set to 300. This model uses the performance indicators commonly used in target detection algorithms, recall rate (Recall), precision (Precision) and average precision (Average Precision, AP) values ​​for evaluation.

[0037] In addition, the experimental environment of the present invention is: system: Windows 10; processor Intel (R) Core (TM) i7-10700 CPU @ 2.90 GHz; memory: 64G; graphics card: NVIDIA TITAN RTX; deep learning framework is Pytorch version 1.12; compilation language is Python version 3.8; CUDA version is 11.2; Cudnn version is 8.3.0.

[0038] Based on the above embodiment, after step 400, the following steps are further included: If the study area is determined to be a snow mountain area, the snow cover normalized difference index is used to obtain the snow cover area of ​​the study area, and the snow mountain area is masked out from the gradient stacking map.

[0039] Specifically, after completing the training of the model, the embodiment of the present invention inputs the phase gradient stacking map and auxiliary data into the trained model to carry out automatic identification of wide-area potential landslides. Since the deformation caused by the melting of snow in the snow-capped mountain area is similar to the landslide deformation, if the study area is covered with snow-capped mountains, the optical image of the study area will be collected, and the Normalized Difference Snow Index (NDSI) will be used to obtain the snow-covered area of ​​the study area and mask out the snow-capped mountain area from the IPGS results before automatic landslide identification is carried out to finally obtain the landslide identification result of the study area. Figure 5 This is the spatial distribution of potential landslides identified in a certain province by the present invention.

[0040] Finally, using the landslide target recognition model obtained by the present invention, the system can automatically identify potential landslides according to the trained model as long as the user provides a gradient stacking map and corresponding auxiliary images.

[0041] The wide-area landslide identification system based on phase gradient stacking and deep learning provided by the present invention is described below. The wide-area landslide identification system based on phase gradient stacking and deep learning described below and the wide-area landslide identification method based on phase gradient stacking and deep learning described above can refer to each other.

[0042] Figure 6is a schematic diagram of the structure of a wide-area landslide identification system based on phase gradient stacking and deep learning provided by an embodiment of the present invention, such as Figure 6 As shown, it includes: an acquisition module 61, a stacking module 62, a labeling module 63, a training module 64 and a recognition module 65, wherein: The acquisition module 61 is used to obtain a set of SAR images covering different time periods in the study area, and obtain a differential interference map using the D-InSAR method; the stacking module 62 is used to obtain a gradient stacking map of the differential interference map using the IPGS method, and crop the gradient stacking map to obtain a cropped image; the labeling module 63 is used to label the cropped image with landslide samples using auxiliary data and optical images to obtain labeled samples; the training module 64 is used to train the improved YOLOv7 model using the labeled samples to obtain a landslide target recognition model; the recognition module 65 is used to input the image to be recognized in the study area into the landslide target recognition model, and output the landslide recognition result of the study area.

[0043] Figure 7 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 7 As shown, the electronic device may include: a processor 710, a communication interface 720, a 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 may call the logic instructions in the memory 730 to execute a wide-area landslide identification method based on phase gradient stacking and deep learning, the method comprising: obtaining a SAR image set covering different periods of the study area, and obtaining a differential interference map using the D-InSAR method; obtaining a gradient stacking map of the differential interference map using the IPGS method, and clipping the gradient stacking map to obtain a clipped image; annotating the clipped image with landslide samples using auxiliary data and optical images to obtain annotated samples; using the annotated samples to train the improved YOLOv7 model to obtain a landslide target recognition model; inputting the image to be identified in the study area into the landslide target recognition model, and outputting the landslide recognition result of the study area.

[0044] 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 this understanding, the technical solution of the present invention can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions to enable 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, etc. Various media that can store program codes.

[0045] On the other hand, the present invention also provides a computer program product, which includes a computer program, and 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 wide-area landslide identification method based on phase gradient stacking and deep learning provided by the above methods, and the method includes: obtaining a set of SAR images covering different time periods of the study area, and using the D-InSAR method to obtain a differential interference map; using the IPGS method to obtain a gradient stacking map of the differential interference map, and cropping the gradient stacking map to obtain a cropped image; using auxiliary data and optical images, annotating the cropped image with landslide samples to obtain annotated samples; using the annotated samples to train an improved YOLOv7 model to obtain a landslide target recognition model; inputting the image to be identified in the study area into the landslide target recognition model, and outputting the landslide recognition result of the study area.

[0046] 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 wide-area landslide identification method based on phase gradient stacking and deep learning provided by the above-mentioned methods, the method comprising: obtaining a set of SAR images covering different time periods of a study area, and obtaining a differential interference map using a D-InSAR method; obtaining a gradient stacking map of the differential interference map using an IPGS method, and cropping the gradient stacking map to obtain a cropped image; annotating the cropped image with landslide samples using auxiliary data and optical images to obtain annotated samples; training an improved YOLOv7 model using the annotated samples to obtain a landslide target recognition model; inputting the image to be identified in the study area into the landslide target recognition model, and outputting the landslide recognition result of the study area.

[0047] 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.

[0048] 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, 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.

[0049] 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 wide-area landslide identification method based on phase gradient stacking and deep learning, characterized in that: include: A set of synthetic aperture radar (SAR) images covering the study area at different times was obtained, and the differential interferogram was obtained using the differential synthetic aperture radar interferometry (D-InSAR) method; An improved phase gradient stacking IPGS method is used to obtain a gradient stacking image of the differential interferogram, and the gradient stacking image is cropped to obtain a cropped image; Using auxiliary data and optical images, annotating the cropped image with landslide samples to obtain annotated samples; The labeled samples are used to train the improved YOLOv7 model to obtain a landslide target recognition model; The image to be identified in the study area is input into the landslide target identification model, and the landslide identification result of the study area is output.

2. The wide-area landslide identification method based on phase gradient stacking and deep learning according to claim 1 is characterized in that: The IPGS method is used to obtain a gradient stacking image of the differential interference image, and the gradient stacking image is cropped to obtain a cropped image, including: Calculating the phase gradient in any direction of the differential interference pattern based on a preset step size; The phase gradients of each differential interferogram corresponding to any direction are stacked from the time dimension and median filtered; The phase stacking images in each direction are merged, and the merged results are gradient normalized; Converting the gradient normalization result to the geographic coordinate system to obtain the gradient stacking map; The gradient stacked image is cropped into an image of a preset size to obtain the cropped image.

3. The wide-area landslide identification method based on phase gradient stacking and deep learning according to claim 1 is characterized in that: Using auxiliary data and optical images, the cropped image is annotated with landslide samples to obtain annotated samples, including: The auxiliary data is cropped into a picture of a preset size to obtain cropped auxiliary data, and a slope map is obtained; The cropped image is annotated by using a preset annotation tool in combination with the optical image, the slope map and the cropped auxiliary data to obtain the annotated sample.

4. The wide-area landslide identification method based on phase gradient stacking and deep learning according to claim 1 is characterized in that: The improved YOLOv7 model is obtained by the following steps: Determining that the improved YOLOv7 model includes an input end, a backbone network, a head module, and a prediction layer; The backbone network includes a CBS module, an ELAN module and an MP module; The head module includes an SPPCSPC module, a CBS module, an ELAN module and an MP module, forming an FPN network structure; The prediction module includes three detection layers, so that the network detects objects of different sizes from three different receptive fields respectively; The CBAM attention mechanism is introduced into the backbone network to capture the correlation between features through adaptive learning channels and spatial attention weights.

5. The wide-area landslide identification method based on phase gradient stacking and deep learning according to claim 1 is characterized in that: The improved YOLOv7 model is trained using the labeled samples to obtain a landslide target recognition model, including: Use the weights trained by the preset image dataset as the pre-trained weights of the network to determine the Batch Size and Epoch; The improved YOLOv7 model is trained using the labeled samples, and the model performance is evaluated using target detection performance indicators, and the landslide target recognition model is obtained after training convergence.

6. The wide-area landslide identification method based on phase gradient stacking and deep learning according to claim 1 is characterized in that: After the improved YOLOv7 model is trained using the labeled samples to obtain a landslide target recognition model, the method further includes: If the study area is determined to be a snow mountain area, the snow cover normalized difference index is used to obtain the snow cover area of ​​the study area, and the snow mountain area is masked out from the gradient stacking map.

7. A wide-area landslide identification system based on phase gradient stacking and deep learning, characterized in that: include: The acquisition module is used to obtain a set of SAR images covering different periods of the study area and obtain differential interferograms using the D-InSAR method; A stacking module, used for obtaining a gradient stacking image of the differential interference image by using an IPGS method, and cropping the gradient stacking image to obtain a cropped image; A labeling module, used for labeling the landslide samples on the cropped images by using auxiliary data and optical images to obtain labeled samples; A training module, used to train the improved YOLOv7 model using the labeled samples to obtain a landslide target recognition model; The identification module is used to input the image to be identified in the study area into the landslide target identification model and output the landslide identification result in the study area.

8. 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 wide-area landslide identification method based on phase gradient stacking and deep learning as described in any one of claims 1 to 6 is implemented.

9. 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 wide-area landslide identification method based on phase gradient stacking and deep learning as described in any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the wide-area landslide identification method based on phase gradient stacking and deep learning as described in any one of claims 1 to 6 is implemented.

Images