Automatic detection method and system for landslide settlement risk area

Abstract

The invention provides an automatic detection method and system for a landslide settlement risk area. The method comprises the following steps: acquiring satellite data of a target area; based on the satellite data, adopting a small baseline integrated aperture radar interferometry technology to obtain a surface deformation rate diagram of the target area; based on the earth surface deformation rate map, identifying an earth surface landslide settlement risk area in the target area through a pyramid scene analysis network; according to the method, interference measurement analysis is carried out on satellite data by adopting the small baseline integrated aperture radar interference measurement technology, so that interference signals can be overcome, and the surface deformation rate extracted by the small baseline integrated aperture radar interference measurement technology can be automatically identified through an improved pyramid scene analysis network; and the calculation amount can be reduced in the process of identifying the land surface settlement risk area, so that the cost of asking experts for analysis each time is saved, and the large-scale landslide monitoring efficiency and reliability are improved.

Description

Technical Field

[0001] The patent application of this invention belongs to the field of surface settlement detection technology, and specifically relates to a method and system for automatic detection of surface landslide settlement risk areas. Background Art

[0002] In high-voltage power transmission systems, transmission towers, as one of the core infrastructures, play an indispensable role in promoting regional economic prosperity and development. Transmission towers are widely distributed in mountainous areas with complex terrain and variable geological conditions. These areas are often located on the back edges of hillsides or on ridges, facing extremely complex geological structures and environmental challenges. Frequent geological disasters such as subsidence and landslides pose serious risks and potential threats to the long-term and effective operation of the transmission network. Therefore, systematic investigation and early identification of subsidence hazards in areas along the power grid are not only crucial for ensuring the safe and stable operation of the power grid, but also have far-reaching academic and practical significance for preventing natural disasters from damaging critical infrastructure and maintaining a stable regional power supply.

[0003] In the field of subsidence hazard identification, traditional field surveys, supplemented by classic measurement methods such as levels, Global Positioning Systems (GPS), and crack meters, have a long history. However, their applicability is limited to small areas and they have significant limitations in efficiency, cost, and environmental adaptability. This is particularly inadequate when dealing with high-lying, remote, or hidden landslide settlements. With the development of optical remote sensing technology and synthetic aperture radar (InSAR), these methods are becoming increasingly important tools for identifying landslide and subsidence hazards. Therefore, the field of subsidence and landslide hazard identification is undergoing a transformation from traditional methods to modern high-tech approaches.

[0004] Synthetic aperture radar interferometry (InSAR) may introduce interference signals due to natural environmental factors such as vegetation cover, atmospheric disturbances, and complex terrain conditions, which in turn affects the accuracy of detection results. The monitoring capability is limited in specific situations, and visual interpretation and manual editing methods are usually used to analyze the results of synthetic aperture radar interferometry, which is inefficient and costly. Summary of the Invention

[0005] In order to overcome the above-mentioned deficiencies of the prior art, the present invention patent application proposes a method for automatically detecting surface landslide settlement risk areas, comprising:

[0006] Collect satellite data of the target area;

[0007] Based on the satellite data, a small baseline synthetic aperture radar interferometry technique is used to obtain a surface deformation rate map of the target area;

[0008] Based on the surface deformation rate map, identifying the surface landslide settlement risk area in the target area through a pre-built improved pyramid scenario analysis network;

[0009] The improved pyramid scene analysis network includes an improved inverted residual layer, a multi-pyramid pooling structure and a dilated pyramid pooling structure. The improved inverted residual layer is a depth-separable convolution structure composed of point-by-point convolution combined with depth-wise convolution.

[0010] Preferably, the construction of the improved pyramid scene analysis network includes:

[0011] Based on the historical satellite data of the target area, a historical surface deformation rate map and deformation labels of the historical surface deformation rate map are obtained;

[0012] The historical surface deformation rate map and the deformation labels of the historical surface deformation rate map are used as the sample set;

[0013] Dividing the sample set into a training sample set and a validation sample set;

[0014] Based on the training sample set, iteratively training the improved inverted residual layer, multi-pyramid pooling structure and dilated pyramid pooling structure in the pyramid scene analysis network to obtain a trained pyramid scene analysis network;

[0015] Based on the verification sample set, the trained pyramid scene analysis network is corrected to complete the construction of the pyramid scene analysis network.

[0016] Preferably, obtaining a historical surface deformation rate map and a deformation label of the historical surface deformation rate map based on historical satellite data of the target area includes:

[0017] Based on historical satellite data of the target area, a small baseline synthetic aperture radar interferometry technique is used to obtain a historical surface deformation rate map;

[0018] Performing deformation identification on the historical surface deformation rate map to obtain a deformation label of the historical surface deformation rate map.

[0019] Preferably, the iterative training of the improved inverted residual layer, the multi-pyramid pooling structure, and the dilated pyramid pooling structure in the pyramid scene analysis network based on the training sample set to obtain a trained pyramid scene analysis network includes:

[0020] Taking the historical surface deformation rate map in the training sample set as input, the improved inverted residual layer is used to perform efficient feature compression to obtain a deformation rate feature vector;

[0021] Based on the deformation rate feature vector, multi-scale image feature information is obtained through a multi-pyramid pooling structure and a dilated pyramid pooling structure, and a deformation rate fusion feature map is generated according to the multi-scale image feature information;

[0022] Using a bilinear interpolation method, upsampling the deformation rate fusion feature map is performed to obtain a predicted deformation label of the historical surface deformation rate map;

[0023] The predicted deformation labels of the historical surface deformation rate map are compared with the deformation labels corresponding to the historical surface deformation rate map in the training sample set. Based on the comparison results, the weights of the improved inverted residual layer, the multi-pyramid pooling structure, and the dilated pyramid pooling structure are updated. The training is iterated until the number of training iterations reaches a preset number of training iterations, thereby completing the training of the pyramid scene analysis network.

[0024] Preferably, the point-by-point convolution includes a first point-by-point convolution and a second point-by-point convolution; the step of taking the historical surface deformation rate map in the training sample set as input and performing efficient feature compression through the improved inverted residual layer to obtain a deformation rate feature vector includes:

[0025] Based on the historical surface deformation rate map in the training sample set, performing deformation rate feature expansion through the first point-by-point convolution to obtain a first point-by-point convolution feature map;

[0026] Based on the first point-by-point convolution feature map, a convolution operation is performed through the depthwise convolution to obtain a depthwise convolution feature map; based on the depthwise convolution feature map, a deformation rate feature vector is obtained through the second point-by-point convolution.

[0027] Preferably, the obtaining of the surface deformation rate map of the target area based on the satellite data by using a small baseline synthetic aperture radar interferometry technique comprises:

[0028] selecting a plurality of synthetic aperture radar images from the satellite data;

[0029] Using pre-set temporal and spatial baseline limits, the plurality of synthetic aperture radar images are paired in pairs to generate a plurality of small baseline sets;

[0030] A pre-set baseline threshold is used to generate the disentangled interferogram corresponding to each small baseline set;

[0031] After performing interference phase unwrapping on each unwrapped interferogram, the obtained unwrapped interference phases are combined into an interference phase matrix, and the surface deformation rate is obtained based on the interference phase matrix using the least squares method;

[0032] Based on the surface deformation rate, a surface deformation rate map is generated.

[0033] Preferably, after performing interference phase unwrapping on each unwrapped interferogram, the obtained unwrapped interference phases are combined into an interference phase matrix, and based on the interference phase matrix, the surface deformation rate is obtained using the least squares method, including:

[0034] Calculating the unwrapping interference phase of each unwrapping interference pattern using an unwrapping interference phase calculation formula based on unwrapping interference phase related parameters;

[0035] generating an interference phase matrix according to all the unwrapped interference phases;

[0036] The least square method is used for inversion to extract a surface deformation time series from the interference phase matrix, and the surface deformation rate is obtained based on the surface deformation time series.

[0037] Preferably, the unwrapping interference phase calculation formula is as follows:

[0038]

[0039] Among them, Δφ j is the unwrapped interference phase of the jth unwrapped interference pattern, IU j is the starting index of the interference pair measurement time period of the jth unwrapped interferogram, IE j is the end index of the interference pair measurement time period of the jth unwrapped interferogram, k is the measurement time of the interference pair measurement time period, t k is the kth moment in the interferometer measurement period, t k-1 is the k-1 moment in the interferometer measurement time period, v k For t k The surface velocity deformation rate at .

[0040] Based on the same inventive concept, the present patent application also provides a surface landslide settlement risk area automatic detection system, including: a satellite data acquisition module, a surface deformation rate map extraction module and a surface landslide settlement risk area identification module;

[0041] The satellite data acquisition module is used to collect satellite data of the target area;

[0042] The surface deformation rate map extraction module is used to obtain the surface deformation rate map of the target area based on the satellite data using a small baseline synthetic aperture radar interferometry technique;

[0043] The surface landslide settlement risk area identification module is used to identify the surface landslide settlement risk area in the target area based on the surface deformation rate map through a pre-built improved pyramid scenario analysis network;

[0044] The improved pyramid scene analysis network includes an improved inverted residual layer, a multi-pyramid pooling structure and a dilated pyramid pooling structure. The improved inverted residual layer is a depth-separable convolution structure composed of point-by-point convolution combined with depth-wise convolution.

[0045] Preferably, the system further comprises: a construction module of an improved pyramid scene analysis network, wherein the construction module of the improved pyramid scene analysis network comprises:

[0046] A historical satellite data processing submodule is used to obtain a historical surface deformation rate map and deformation labels of the historical surface deformation rate map based on historical satellite data of the target area;

[0047] A sample set generation submodule is used to use the historical surface deformation rate map and the deformation labels of the historical surface deformation rate map as a sample set;

[0048] A sample set division submodule, configured to divide the sample set into a training sample set and a validation sample set;

[0049] An iterative training submodule is used to iteratively train the improved inverted residual layer, multi-pyramid pooling structure and dilated pyramid pooling structure in the pyramid scene analysis network based on the training sample set to obtain a trained pyramid scene analysis network;

[0050] The model correction submodule is used to correct the trained pyramid scene analysis network based on the verification sample set to complete the construction of the pyramid scene analysis network.

[0051] Preferably, the historical satellite data processing submodule is specifically used to:

[0052] Based on historical satellite data of the target area, a small baseline synthetic aperture radar interferometry technique is used to obtain a historical surface deformation rate map;

[0053] Performing deformation identification on the historical surface deformation rate map to obtain a deformation label of the historical surface deformation rate map.

[0054] Preferably, the iterative training submodule includes:

[0055] an inverted residual layer training unit, configured to take the historical surface deformation rate map in the training sample set as input, perform efficient feature compression through the improved inverted residual layer, and obtain a deformation rate feature vector;

[0056] a deformation rate fusion feature map generating unit, configured to obtain multi-scale image feature information based on the deformation rate feature vector by using a multi-pyramid pooling structure and a dilated pyramid pooling structure, and generate a deformation rate fusion feature map according to the multi-scale image feature information;

[0057] an upsampling unit, configured to perform upsampling processing on the deformation rate fusion feature map by using a bilinear interpolation method to obtain a predicted deformation label of the historical surface deformation rate map;

[0058] The structural parameter updating unit is used to compare the predicted deformation label of the historical surface deformation rate map with the deformation label corresponding to the historical surface deformation rate map in the training sample set, update the weights of the improved inverted residual layer, the multi-pyramid pooling structure and the hollow pyramid pooling structure according to the comparison result, and iterate the training number until the number of training iterations reaches a preset number of training iterations, thereby completing the training of the pyramid scene analysis network.

[0059] Preferably, the point-by-point convolution includes a first point-by-point convolution and a second point-by-point convolution; the inverted residual layer training unit is specifically used to:

[0060] Based on the historical surface deformation rate map in the training sample set, performing deformation rate feature expansion through the first point-by-point convolution to obtain a first point-by-point convolution feature map;

[0061] Based on the first point-by-point convolution feature map, performing a convolution operation through the depthwise convolution to obtain a depthwise convolution feature map;

[0062] Based on the depth convolution feature map, a deformation rate feature vector is obtained through the second point-by-point convolution.

[0063] Preferably, the satellite data acquisition module includes:

[0064] A data screening submodule, configured to screen out a plurality of synthetic aperture radar images from the satellite data;

[0065] A small baseline set generation submodule is configured to pair the plurality of synthetic aperture radar images in pairs using preset time and space baseline limits to generate a plurality of small baseline sets;

[0066] A disentangled interferogram generation submodule is used to generate a disentangled interferogram corresponding to each small baseline set using a preset baseline threshold;

[0067] The surface deformation rate extraction submodule is used to perform interference phase unwrapping on each unwrapped interferogram, form an interference phase matrix with the obtained unwrapped interference phases, and obtain the surface deformation rate based on the interference phase matrix using the least squares method;

[0068] The surface deformation rate map generating submodule is used to generate a surface deformation rate map based on the surface deformation rate.

[0069] Preferably, the surface deformation rate extraction submodule is specifically used to:

[0070] Calculating the unwrapping interference phase of each unwrapping interference pattern using an unwrapping interference phase calculation formula based on unwrapping interference phase related parameters;

[0071] generating an interference phase matrix according to all the unwrapped interference phases;

[0072] The least square method is used for inversion to extract a surface deformation time series from the interference phase matrix, and the surface deformation rate is obtained based on the surface deformation time series.

[0073] Preferably, the unwrapping interference phase calculation formula is as follows:

[0074]

[0075] Among them, Δφ j is the unwrapped interference phase of the jth unwrapped interference pattern, IU j is the starting index of the interference pair measurement time period of the jth unwrapped interferogram, IE j is the end index of the interference pair measurement time period of the jth unwrapped interferogram, k is the measurement time of the interference pair measurement time period, t k is the kth moment in the interferometer measurement period, t k-1 is the k-1 moment in the interferometer measurement time period, v k For t k The surface velocity deformation rate at .

[0076] Based on the same inventive concept, the present invention patent application also provides an electronic device, comprising: at least one processor and a memory; the memory and the processor are connected via a bus;

[0077] The memory is used to store one or more programs;

[0078] When the one or more programs are executed by the at least one processor, the aforementioned method for automatically detecting surface landslide settlement risk areas is implemented.

[0079] Based on the same inventive concept, the patent application of the present invention also provides a readable storage medium on which a computer program and an execution program are stored. When the execution program is executed, an automatic detection method for surface landslide settlement risk areas as described above is implemented.

[0080] Compared with the closest prior art, the patent application of this invention has the following beneficial effects:

[0081] The patent application of the present invention provides a detection method and system for the Sichuan-Tibet line area, comprising: collecting satellite data of a target area; based on the satellite data, using a small baseline set synthetic aperture radar interferometry measurement technology to obtain a surface deformation rate map of the target area; based on the surface deformation rate map, identifying the surface landslide settlement risk area in the target area through a pre-constructed improved pyramid scene analysis network; wherein the improved pyramid scene analysis network includes an improved inverted residual layer, a multi-pyramid pooling structure and a void pyramid pooling structure, and the improved inverted residual layer is a depth-separable convolution structure composed of point-by-point convolution combined with depth convolution; the present invention adopts Small baseline aggregated aperture radar interferometry technology performs interferometric measurement and analysis on satellite data, thereby overcoming interference signals that may be introduced by natural environmental factors on the ground, such as vegetation cover, atmospheric disturbances, and complex terrain conditions. The baseline aggregated aperture radar interferometry technology of the present invention is used to monitor surface settlement and landslide deformation with high precision. The improved pyramid scene analysis network not only can automatically identify the surface deformation rate extracted by the small baseline aggregated aperture radar interferometry technology, but also can reduce the amount of calculation in the process of identifying surface settlement risk areas, thereby saving the cost of expert analysis each time and helping to improve the efficiency and reliability of large-scale landslide monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] Figure 1 A schematic flow chart of a method for automatically detecting surface landslide settlement risk areas provided in the patent application of this invention;

[0083] Figure 2 A schematic diagram of the training loss of the improved PSPNet model provided in the patent application of this invention;

[0084] Figure 3 Sample schematic diagram of historical surface deformation rate map and deformation label provided for the patent application of this invention;

[0085] Figure 4 A schematic diagram of the MobilenetV2 structure provided for the patent application of this invention;

[0086] Figure 5 A schematic diagram of the overall technical route of the present invention provided for the present patent application;

[0087] Figure 6 Schematic diagram of the study area and Sentinel-1 data coverage provided for the patent application of this invention;

[0088] Figure 7 A deformation rate map of the study area from January 2020 to March 2023 obtained using SBAS-InSAR technology, provided for the patent application of this invention;

[0089] Figure 8 A schematic diagram of an automatic detection system for surface landslide and settlement risk areas provided in the patent application of this invention;

[0090] Figure 9 A schematic structural diagram of an electronic device provided for the patent application of this invention. DETAILED DESCRIPTION

[0091] The specific implementation methods of the patent application of the present invention are further described in detail below with reference to the accompanying drawings.

[0092] Example 1:

[0093] The present invention patent application provides a method for automatically detecting landslide settlement risk areas on the surface. Figure 1 Shown, including:

[0094] Step 1: Collect satellite data of the target area;

[0095] Step 2: Based on the satellite data, a small baseline synthetic aperture radar interferometry technique is used to obtain a surface deformation rate map of the target area;

[0096] Step 3: Based on the surface deformation rate map, identify the surface landslide settlement risk areas in the target area through a pre-built improved pyramid scenario analysis network;

[0097] The improved pyramid scene analysis network includes an improved inverted residual layer, a multi-pyramid pooling structure and a dilated pyramid pooling structure. The improved inverted residual layer is a depth-separable convolution structure composed of point-by-point convolution combined with depth-wise convolution.

[0098] InSAR technology is prone to decoherence, especially when the target area undergoes rapid deformation, which limits the scope of effective monitoring. The inherent constraints of InSAR technology lead to uneven distribution of generated interference points, making it difficult to fully cover the study area, affecting data integrity and analysis depth. Therefore, accurate interpretation of landslide deformation usually requires combining expert knowledge and considering deformation characteristics. In previous studies, the lack of efficient automated interpretation methods still hinders the application of InSAR technology in large-scale landslide detection. Simple detection methods such as threshold methods are easily affected by noise in InSAR technology measurement results and are prone to false detection. Therefore, the development of automated methods to identify settlement landslide risks is of great significance and will help improve the efficiency and reliability of large-scale landslide monitoring.

[0099] With the rapid development of computer vision, deep learning technologies such as convolutional neural networks (CNNs) have made outstanding achievements in the automatic identification of landslides. In recent years, many studies have combined CNNs with optical images to detect landslides and have achieved encouraging results. These methods can be divided into object detection-based methods and semantic segmentation-based methods. Object detection-based methods use bounding boxes to locate landslides. For example, Cheng et al. (2021) combined the YOLO-V4 (You Only Look Once-Version 4, an object detection algorithm) model and the attention mechanism to detect landslides using optical remote sensing images from Google Earth Terrain Mapping (virtual globe software). Tanatipuknon et al. (2021) found that combining two region-based RCNN (Regions with Convolutional Neural Networks) models with a simple decision tree can achieve better results in detecting landslides. Semantic segmentation-based methods, such as U-Net and Fully Convolutional Network (FCN), can classify pixels of landslides and non-landslides to delineate their boundaries. For example, Ghorbanzadeh et al. (2022) proposed a new strategy that combines rule-based object-based image analysis (OBIA) with a fully convolutional neural network (FCN) to detect landslides from multi-temporal Sentinel-2 imagery. Lu et al. (2023) proposed a dual-encoder U-Net for landslide detection using Sentinel-2 and digital elevation model (DEM) data. Inspired by deep learning techniques in optical remote sensing, researchers have begun to study the application of deep learning in the field of InSAR. Several studies have used CNN and InSAR to monitor and detect geological hazards such as earthquakes, volcanoes, ground subsidence, and mining subsidence. These studies have demonstrated the great potential of CNN for automatically interpreting InSAR results and effectively detecting geological hazards. A recent study proposed a new method for automatically identifying active landslides using CNN and phase gradient stacking. Without phase unwrapping, phase gradient stacking can quickly obtain landslide displacement locations. However, this method is susceptible to phase noise (caused by shadows, overlaps, or DEM errors), which can lead to false landslide detection. Overall, there is limited research on automated detection of subsidence landslides using InSAR and CNN.

[0100] Therefore, in the present invention, an improved pyramid scene analysis network is first constructed, and the construction process is as follows:

[0101] In one implementation, the construction of the improved pyramid scene analysis network includes:

[0102] Based on the historical satellite data of the target area, a historical surface deformation rate map and deformation labels of the historical surface deformation rate map are obtained;

[0103] The historical surface deformation rate map and the deformation labels of the historical surface deformation rate map are used as the sample set;

[0104] For example, through visual interpretation of InSAR results in the study area, a sample set of landslide image data was created to train the improved PSPNe (Pyramid Scene Parsing Network).

[0105] Dividing the sample set into a training sample set and a validation sample set;

[0106] For example, in order to train the improved PSPNet, 70% of the images in the sample set are randomly selected as the training sample set, 20% of the images are set as the verification sample set, and 10% of the images are selected as the test sample set.

[0107] Based on the training sample set, iteratively training the improved inverted residual layer, multi-pyramid pooling structure and dilated pyramid pooling structure in the pyramid scene analysis network to obtain a trained pyramid scene analysis network;

[0108] Based on the verification sample set, the trained pyramid scene analysis network is corrected to complete the construction of the pyramid scene analysis network.

[0109] For example, in training the improved PSPNet, the initialization of weights, biases, and batch normalization scaling factors is crucial. To ensure effective network learning, these parameters are initialized using a normal distribution. Specifically, weights are sampled from a normal distribution with mean 0 and variance 1, biases are typically initialized to 0, and batch normalization scaling factors are set to 1. This ensures that the network's performance in the initial stages is not significantly affected by improper parameter initialization. Next, the network's initial learning rate is set to 0.01, a common choice when using the Stochastic Gradient Descent (SGD) optimizer. The learning rate controls the step size of each parameter update. A high learning rate may lead to unstable convergence, while a low learning rate may result in slow training. Regarding the batch size, a batch size of 8 was initially selected, meaning that each time the improved PSPNet model parameters were updated, 8 samples were used for backpropagation and weight updates. All initialized parameter data was then fed into the network to begin training. In terms of the optimizer, the stochastic gradient descent (SGD) algorithm is used to update the parameters. The update expression is as follows:

[0110]

[0111] Among them, W t+1 is the parameter at time t+1, W t is the parameter at time t, η is the learning rate, is the loss function with respect to W t The gradient of the parameter, λ is the weight decay factor (Weight Decay).

[0112] The optimizer can accelerate convergence through momentum, with the momentum parameter set to 0.9. Furthermore, to further improve training efficiency, a learning rate descent strategy is employed. Here, the cosine annealing learning rate descent method is chosen, meaning that the learning rate gradually decreases as the training progresses. Its variation can be expressed as follows:

[0113]

[0114] Among them, η t is the learning rate at time t, η min is the minimum learning rate, η max is the maximum learning rate, T max is the maximum learning cycle.

[0115] This method can effectively prevent the improved PSPNet from falling into local optimality in the later stage of training and maintain the stability and performance of the improved PSPNet; Figure 2 The figure shows the training loss of the improved PSPNet. The loss function graph shows that the improved PSPNet exhibits good convergence during training. The loss decreases significantly with the number of training epochs. Early in training, the loss decreases rapidly from approximately 1.8, especially within the first 20 epochs, indicating that the improved PSPNet quickly learns effective features. As training continues, the loss stabilizes, remaining around 0.2 near the 50th epoch, indicating that the training process of the improved PSPNet model has converged and no longer shows a significant loss decrease. This loss trend demonstrates that the improved PSPNet model is able to effectively update parameters and gradually approach the optimal solution under a given optimizer (such as stochastic gradient descent or other optimization algorithms) and learning rate adjustment strategy. Finally, we calculated that the improved PSPNet model achieved an accuracy of 90.72%.

[0116] In recent years, transmission network engineering projects have frequently encountered subsidence geological disasters. These events often result in the destruction of transmission towers and power outages, posing a serious threat to regional power security. Spaceborne synthetic aperture interferometry (InSAR) technology has the ability to monitor surface deformation around the clock and in all weather conditions. In recent years, its ability to monitor slow deformation has enabled the effective identification and early warning of potential geological hazards such as subsidence and landslides, thus providing important support for the safe operation of power facilities. However, research on the automated detection of subsidence risk using InSAR using deep learning techniques is relatively limited. Automated detection is crucial for interpreting subsidence risk by non-InSAR professionals. To overcome the challenges of identifying subsidence risk zones at different scales, a method for automatically detecting subsidence risk zones in power grid corridors based on PSPNet and InSAR technology is proposed to improve the efficiency and accuracy of subsidence geological hazard detection. This method utilizes a pyramid pooling module to pool InSAR deformation rate maps at different scales to capture the multi-scale contextual information of the deformation image, thereby improving the accuracy of subsidence risk identification. Through visual interpretation of InSAR results in the study area, the present invention creates a sample set to train an improved PSPNet, enabling automated monitoring of subsidence and landslide risk areas. By performing in-depth analysis and feature extraction on historical surface deformation rate maps fed into the improved PSPNet, the model is able to identify and classify various risk areas. After training, the improved PSPNet is deployed to monitor and update risk assessments in real time, effectively providing early warning of potential geological hazards and improving the efficiency of emergency response.

[0117] In one implementation, obtaining a historical surface deformation rate map and a deformation label of the historical surface deformation rate map based on historical satellite data of the target area includes:

[0118] Based on historical satellite data of the target area, a small baseline synthetic aperture radar interferometry technique is used to obtain a historical surface deformation rate map;

[0119] Performing deformation identification on the historical surface deformation rate map to obtain a deformation label of the historical surface deformation rate map.

[0120] For example, Figure 3 This is a sample diagram of a historical surface deformation rate map and deformation labels. The size of the historical surface deformation rate map is 224*224. The historical surface deformation rate map is manually deformed to obtain the deformation labels of the historical surface deformation rate map.

[0121] For example, synthetic aperture interferometry (InSAR), part of the small baseline subset interferometric synthetic aperture radar (SBAS-InSAR) technique, not only significantly improves identification efficiency and achieves wider monitoring coverage, but also demonstrates significant economic advantages by reducing operating costs. Compared to traditional methods, it exhibits superior overall performance and development potential. Optical remote sensing technology, with its high resolution, wide coverage, and rich image information, can rapidly identify ancient landslides undergoing overall deformation and landslide bodies with significant deformation characteristics. However, its ability to identify landslides that are slowly deforming and lack obvious signs of deformation is limited. Furthermore, cloud and fog obstruction significantly impacts the quality of optical remote sensing images, making all-day, all-weather monitoring impossible. Synthetic aperture radar interferometry utilizes satellite-acquired radar imagery to monitor surface subsidence and landslide deformation with high precision through interferometry. By analyzing phase differences between radar images acquired at different time points, InSAR generates high-resolution surface deformation maps, revealing subtle deformation and dynamic changes in subsidence and landslide areas. Currently, InSAR technology has reached a relatively mature stage. Leveraging its ability to deeply analyze radar image phase information, it enables precise detection of millimeter-level surface deformation. Its unique advantages lie in its significantly improved measurement accuracy, immunity to cloud cover, all-weather operation, and wide monitoring coverage, all of which have been extensively explored in the literature. InSAR technology not only overcomes the limitations of optical applications in cloudy and rainy conditions, but also demonstrates significant value and potential in the acquisition of high-precision surface deformation data for both scientific research and practical applications, including a deeper understanding of the dynamic evolution of landslide deformation, analysis of its formation mechanisms, and early identification and early warning.

[0122] In one implementation, based on the training sample set, iteratively training the improved inverted residual layer, multi-pyramid pooling structure, and dilated pyramid pooling structure in the pyramid scene analysis network to obtain a trained pyramid scene analysis network includes:

[0123] Taking the historical surface deformation rate map in the training sample set as input, the improved inverted residual layer is used to perform efficient feature compression to obtain a deformation rate feature vector;

[0124] For example, the historical surface deformation rate map in the training sample set is received. In the improved pyramid scene analysis network, the historical surface deformation rate map is uniformly adjusted to 224×224 pixel size, and then efficient feature compression is performed through the improved Bottleneck (inverted residual) layer. This process maps the deformation historical surface deformation rate map features into a deformation rate feature vector with a 28×28 spatial resolution and 192 channels.

[0125] Based on the deformation rate feature vector, multi-scale image feature information is obtained through a multi-pyramid pooling structure and a dilated pyramid pooling structure, and a deformation rate fusion feature map is generated according to the multi-scale image feature information;

[0126] For example, the deformation rate feature vector is fed in parallel into two core components of PSPNet: the multi-pyramid pooling module and the ASPP (Atrous Spatial Pyramid Pooling) dilated pyramid pooling structure. These two structures work together to capture rich feature information from different scales in the image, effectively improving the PSPNet model's ability to recognize complex scenes and multi-scale objects. Finally, after the above feature extraction and fusion, the resulting deformation rate fused feature map is obtained.

[0127] Using a bilinear interpolation method, upsampling the deformation rate fusion feature map is performed to obtain a predicted deformation label of the historical surface deformation rate map;

[0128] For example, bilinear interpolation is used to perform upsampling to restore and map the historical surface deformation rate map to its original size, thereby achieving fine semantic segmentation of the historical surface deformation rate map. During this process, each pixel in the historical surface deformation rate map is accurately divided into two categories: risk-free area and subsidence risk area, and marked with masks 0 and 1, respectively, to intuitively display the segmentation results.

[0129] The predicted deformation labels of the historical surface deformation rate map are compared with the deformation labels corresponding to the historical surface deformation rate map in the training sample set. Based on the comparison results, the weights of the improved inverted residual layer, the multi-pyramid pooling structure, and the dilated pyramid pooling structure are updated, and the training is iterated until the number of training iterations reaches a preset number of training iterations, thereby completing the training of the pyramid scene analysis network. The present invention trains the pyramid scene analysis network using deformation rate sample images and deformation labels of the deformation rate sample images. The depthwise separable convolution structure can significantly reduce the dimension of the feature map during the convolution process, thereby reducing the computational complexity of the pyramid scene analysis network. The core components of the pyramid scene analysis network, namely the multi-pyramid pooling structure and the dilated pyramid pooling structure, work together to capture rich feature information in the image at different scales, effectively improving the model's recognition ability for complex scenes and multi-scale targets. The feature map obtained after the above feature extraction and fusion is upsampled by a bilinear interpolation method to restore and map it to the original image size, thereby achieving fine semantic segmentation of the image content and obtaining the predicted deformation label of the deformation rate sample image.

[0130] In one implementation, the point-by-point convolution includes a first point-by-point convolution and a second point-by-point convolution; the step of taking the historical surface deformation rate map in the training sample set as input and performing efficient feature compression through the improved inverted residual layer to obtain a deformation rate feature vector includes:

[0131] Based on the historical surface deformation rate map in the training sample set, performing deformation rate feature expansion through the first point-by-point convolution to obtain a first point-by-point convolution feature map;

[0132] Based on the first point-by-point convolution feature map, a convolution operation is performed through the depthwise convolution to obtain a depthwise convolution feature map; based on the depthwise convolution feature map, a deformation rate feature vector is obtained through the second point-by-point convolution.

[0133] For example, the improved inverted residual layer of the present invention is a depth-separable convolution MobilenetV2 (a lightweight convolutional neural network model) structure, such as Figure 4The figure shows a schematic diagram of the MobilenetV2 structure. Pointwise convolution stands for point-by-point convolution. Pointwise convolution refers to a 1×1 convolution operation, which operates only on the channel dimension. It multiplies each input channel by the corresponding weight point by point to change the number of channels in the feature map. Depthwise convolution stands for depthwise convolution. Depthwise convolution is an operation that performs convolution independently on each channel, that is, each channel is multiplied by a convolution kernel separately. Unlike traditional convolution, it is mainly used to reduce the amount of model calculation and is used together with Pointwise convolution to construct depth-separable convolution. The innovation of MobileNetV2 compared to traditional convolutional neural networks (CNN) is that a significant difference is that it introduces the Depthwise Separable convolution mechanism, which replaces the traditional standard convolution operation. Depthwise Separable convolution first performs convolution operations on each input channel independently through a single-channel convolution layer, referred to as depthwise convolution, and then uses a 1×1 convolution kernel, referred to as pointwise convolution, to perform cross-channel feature fusion and dimensionality expansion. This convolution method significantly reduces the dimension of the feature map during the convolution process. Therefore, it is usually necessary to pre-process the feature with a 1×1 point-by-point convolution for feature expansion to prepare for subsequent convolution operations, and perform feature dimensionality reduction through a 1×1 point-by-point convolution when necessary. This design pattern constitutes an improved inverted residual (Bottleneck) structure; the point-by-point convolution of the present invention operates on the channel dimension, multiplying each input channel by the corresponding weight point by point to change the number of channels and feature dimensionality reduction of the feature map. The depth convolution can perform cross-channel feature fusion and dimensionality expansion, reducing the computational complexity of the pyramid scene analysis network.

[0134] In one implementation, the step 1 above collects satellite data of the target area;

[0135] For example, the C-band (wavelength 55.466 mm) Sentinel-1 satellite data publicly released by the European Space Agency is used as the main data source for InSAR processing. Sentinel-1 is an all-day, all-weather radar imaging system. It is the first satellite developed by the European Commission and the European Space Agency for the Copernicus global Earth observation project. It was launched in April 2014 and entered the application phase in October 2014. In 2016, the B satellite of the series, Sentinel-1B, was launched, together forming the Sentinel-1 series SAR (Synthetic Aperture Radar) satellite system of the Copernicus program. The original intention of this system is to target large-scale InSAR applications.

[0136] For example, the Sentinel-1A / B satellites utilize a short spatial baseline design and a short revisit period: 12 days in single-satellite mode and 6 days in dual-satellite mode. Data is re-covered approximately every 6 days across Europe, while other major regions are covered every 12 days. This makes Sentinel-1 satellite data the richest data source for InSAR applications. To date, the Sentinel-1 historical archive exceeds 6PB. In addition to its global coverage, short revisit period, and wide swath coverage, Sentinel-1 data is crucially open and freely distributed to the public worldwide by ESA (http: / / www.copernicus.eu). Therefore, the high-quality Sentinel-1 data provides a reliable foundation for this project.

[0137] In one implementation, the step 2 above, based on the satellite data, adopting a small baseline synthetic aperture radar interferometry technique to obtain a surface deformation rate map of the target area, includes:

[0138] selecting a plurality of synthetic aperture radar images from the satellite data;

[0139] For example, Synthetic Aperture Radar (SAR) images suitable for SBAS-InSAR processing are collected from the selected C-band Sentinel-1 satellite data. Because they cover the entire world and are revisited every 6 to 12 days, they provide data with high temporal and spatial resolution.

[0140] For example, first, select Sentinel-1 satellite data covering the target area. One satellite cycle can acquire multiple SAR images, and prioritize those with a short time interval (12 days) and a small spatial baseline distance (typically less than 200 meters). This effectively reduces data interference caused by temporal and spatial incoherence, improving the continuity and reliability of monitoring data.

[0141] Using pre-set temporal and spatial baseline limits, the plurality of synthetic aperture radar images are paired in pairs to generate a plurality of small baseline sets;

[0142] For example, the SBAS (Satellite-Based Augmentation System) algorithm is used to combine the SAR images obtained above by setting specific time and space baseline limits to form multiple pairwise combinations, and a total of P pairwise small baseline sets are generated.

[0143] A pre-set baseline threshold is used to generate the disentangled interferogram corresponding to each small baseline set;

[0144] Finally, for each small baseline set, the corresponding disentangled interferogram is generated by setting an appropriate baseline threshold.

[0145] After performing interference phase unwrapping on each unwrapped interferogram, the obtained unwrapped interference phases are combined into an interference phase matrix, and the surface deformation rate is obtained based on the interference phase matrix using the least squares method;

[0146] Based on the surface deformation rate, a surface deformation rate map is generated.

[0147] For example, after screening multiple synthetic aperture radar images, small baseline set synthetic aperture radar interferometry (SBAS-InSAR) technology is applied for data processing; by establishing a small baseline set, the impact caused by temporal and spatial incoherence is reduced. The goal of this step is to accurately extract surface deformation information from SAR images and provide a scientific basis for landslide monitoring; the present invention generates a small baseline set based on synthetic aperture radar images, which can reduce the impact caused by temporal and spatial incoherence. The interference points in the differential detangling interferogram generated based on the small baseline set are evenly distributed, which improves data integrity and analysis depth. The surface deformation rate reflects the surface deformation information of the target area.

[0148] In one implementation, after performing interference phase unwrapping on each unwrapped interferogram, the obtained unwrapped interference phases are combined into an interference phase matrix. Based on the interference phase matrix, a least squares method is used to obtain the surface deformation rate, including:

[0149] Calculating the unwrapping interference phase of each unwrapping interference pattern using an unwrapping interference phase calculation formula based on unwrapping interference phase related parameters;

[0150] For example, calculate the unwrapped interferogram composed of two SAR images, and obtain the unwrapped phase at the target pixel in the unwrapped interferogram. Define the components contained in the phase. Assume that the kth unwrapped interferogram is composed of t A and t B The SAR image is generated at time t A >t B , then the calculation formula for defining the unwrapped phase at pixel (x, r) is:

[0151]

[0152] Among them, φ j (x, r) is the unwrapping phase of the jth unwrapping interferogram at pixel (x, r), φ(t C ,x,r) is the CAt the moment, the unwrapped phase at the pixel (x, r), φ(t A ,x,r) is the A At time t, the unwrapped phase at pixel (x, r), λ is the radar wavelength, d(t C ,x,r) is the C At the moment, the distance from the satellite to the ground at pixel (x, r), d(t A ,x,r) is the A At the moment, the distance from the satellite to the ground at pixel (x, r), B ⊥j represents the vertical baseline distance of the jth observation, that is, the vertical spatial baseline of the radar imaging satellite at the jth observation, Δz represents the elevation error, r represents the slant distance, θ represents the incident angle, and φ atm (t C ,x,r) is the C At the moment, the atmospheric phase at the pixel (x, r), φ atm (t A ,x,r) is the A At the moment, the atmospheric phase at the pixel (x, r), Δn j is the jth random noise term, j is the number of disentangled interferograms or random noise terms, and tP is the total number of disentangled interferograms or random noise terms.

[0153] Generate the deformation time series of each target pixel, and then the deformation time series of each target pixel is defined as follows:

[0154] φ T =[φ(t1),…,φ(t N )

[0155] Among them, φ T is the deformation time series of the target pixel, T is the time of SAR image observation, φ(t1) is the deformation information of the target pixel at time t1, φ(t N ) is the target pixel at t N Deformation information at the moment.

[0156] For the first moment, that is, the deformation information φ(t0) of the target pixel at time t0=0, there are a total of N+1 SAR images.

[0157] For each pixel (x, r), the vector of P disentangled interference patterns obtained at its position is as follows:

[0158] δφ T =[δφ1,…,δφ P ]

[0159] Among them, δφ Tis the unwrapped interferometric phase, T is the time of SAR image observation, δφ1 is the first unwrapped interferometric phase, δφ P is the pth unwrapped interference phase;

[0160] For each unwrapped phase, two index vectors, the master image and the slave image, are defined. The index vector is the time index. The master image is defined as follows:

[0161] IM=[IM1,…,IM P ]

[0162] Among them, IM is the main image of the unwrapped interferogram, IM1 is the main image of the first unwrapped interferogram, and IM P is the main image of the pth unwrapped interferogram;

[0163] The slave image is defined as follows:

[0164] IS=[IS1,…,IS P ]

[0165] Among them, IS is the slave image of the unwrapped interferogram, IS1 is the slave image of the first unwrapped interferogram, and IS P is the slave image of the pth unwrapped interference pattern;

[0166] And the main image and the slave image are sorted in the order of acquisition time so that they satisfy the following formula:

[0167]

[0168] Among them, IM j is the main image of the jth unwrapped interferogram, IS j is the slave image of the jth unwrapped interferogram;

[0169] Therefore, based on the two index vectors of the master image and the slave image, the unwrapped phase of the unwrapped interferogram can be calculated as follows:

[0170]

[0171] Among them, δφ j is the unwrapping phase of the jth unwrapping interferogram, is the main image vector of the jth unwrapped interferogram, is the slave image phase of the jth unwrapped interferogram.

[0172] For t A With t B The unwrapped interferogram generated by two SAR images acquired at the same time can be expressed by the following formula for any point of the unwrapped phase:

[0173]

[0174] Among them, Δφ i is the unwrapped phase of any point in the i-th unwrapped interferogram, t C The unwrapped phase of the moment, t A The unwrapped phase at time , λ is the radar wavelength, For t C The radar line of sight is constantly accumulating surface deformation. For t A The radar line of sight is constantly accumulating surface deformation. is the terrain phase in the i-th unwrapped interferogram, is the atmospheric delay phase in the i-th unwrapped interferogram, is the noise phase in the i-th unwrapped interferogram.

[0175] Remove the terrain phase from the unwrapped interferogram Atmospheric delay phase and noise phase After that, the calculation formula for the unwrapped phase at any point only includes the surface deformation phase. At the same time, the deformation phase in the calculation formula for the unwrapped phase at the pixel (x, r) is rewritten as the product of the average deformation rate and the time interval between the two acquisition times, thus obtaining the calculation formula for the unwrapped interferometric phase.

[0176] generating an interference phase matrix according to all the unwrapped interference phases;

[0177] For example, the mathematical expression of the interference phase matrix is ​​as follows:

[0178] A·v=Δφ

[0179] Where Δφ is the interference phase matrix, A is a coefficient matrix with dimension M×N, and v is the surface velocity deformation rate.

[0180] The least square method is used for inversion to extract a surface deformation time series from the interference phase matrix, and the surface deformation rate is obtained based on the surface deformation time series.

[0181] For example, the least square method is used to solve the above interference phase matrix to obtain the surface deformation rate.

[0182] For example, the present invention utilizes small-baseline synthetic aperture radar interferometry to monitor landslide geological hazards along the Sichuan-Tibet Highway. In the context of InSAR (Interferometric Synthetic Aperture Radar) data processing, this method generally refers to extracting time series information about surface deformation from a series of interferograms. Specifically, this process involves several key computational steps: Unwrapping the interferogram phase: First, each interferogram must be phase-unwrapped. This is because the phase directly derived from the interferogram is modulo 2π²\pi²π, meaning that the phase values ​​cycle between -π and π. Unwrapping converts these "wrapped" phases into continuous, real-world phase variations. Constructing a matrix representation of the interferogram phase: The unwrapped interferogram phases at multiple time points are combined into a matrix, where each column represents the phase map at a specific time point. This matrix serves as the basis for subsequent analysis. Time series analysis: Least squares methods are used to extract the time series of surface deformation from the phase matrix. This process may involve comparing and analyzing the phases at different time points to estimate the deformation of each pixel over time.

[0183] In one implementation, the unwrapping interference phase calculation formula is as follows:

[0184]

[0185] Among them, Δφ j is the unwrapped interference phase of the jth unwrapped interference pattern, IU j is the starting index of the interference pair measurement time period of the jth unwrapped interferogram, IE j is the end index of the interference pair measurement time period of the jth unwrapped interferogram, k is the measurement time of the interference pair measurement time period, t k is the kth moment in the interferometer measurement period, t k-1 is the k-1 moment in the interferometer measurement time period, v k For t k The surface velocity deformation rate at .

[0186] In the present invention, Figure 5 Figure 1 is a schematic diagram of the overall technical route of the present invention. In the figure, Sentinel is the first radar imaging satellite in the Sentinel series of satellites of the European Space Agency, Sentinel-1 data is Sentinel-1 satellite data, DEM is Digital Elevation Model, SLC is Single Look Complex, and SAR is Synthetic Aperture Radar.

[0187] DEM data: DEM (Digital Elevation Model) data in InSAR is mainly used in two aspects:

[0188] (1) Eliminating terrain effects: InSAR technology monitors surface deformation by comparing radar images acquired over the same area at different times. Because radar signals are affected by terrain (such as mountains and hills), the measured phase changes include components of terrain height. DEM data can help remove this terrain effect from the InSAR phase difference, leaving only the actual deformation information of the surface. This step is often referred to as "terrain phase removal."

[0189] (2) Geocoding: DEM data is required to convert data from the radar coordinate system to the geographic coordinate system.

[0190] Precise orbital data: Sentinel-1's precise orbital data is used to improve the accuracy of InSAR processing. By providing the satellite's exact position and motion in Earth orbit, this data eliminates phase deviations caused by orbital errors, ensuring the accuracy of interferograms.

[0191] SLC single-view complex image: It is a raw data format of SAR image, which contains the amplitude and phase information of each pixel.

[0192] The present invention obtains an SLC single-view complex image based on Sentinel-1 data, DEM data and precise orbit data, and converts the DEM coordinates to the SAR coordinate system. When using InSAR technology, converting the DEM coordinate system to the SAR coordinate system is a common step. SAR images usually use the radar coordinate system, while DEM data may use the geographic coordinate system. Therefore, coordinate system conversion and registration are necessary processes to ensure that SAR data and DEM data are correctly registered. The SAR image is registered based on the SLC single-view complex image combined with the SAR coordinate system, and interference pairs are selected based on the baseline map to generate a twisted interferogram. The twisted interferogram is an important step in InSAR processing. It is used to represent the SAR images obtained by two satellite observations (acquired at different times). The twisted interferogram is usually obtained by calculating the interference phase between two SAR images. This process can reveal information such as surface deformation and terrain features, and an unwrapped interferogram is obtained based on the twisted interferogram; the unwrapped interferogram is sequentially subjected to least squares inversion, atmospheric phase removal, and geocoding to obtain a time series surface deformation inversion, and the time series deformation can be inverted based on the unwrapped phase based on the least squares; and based on the time series surface deformation inversion, the annual average deformation rate inversion is obtained, and the deformation rate map can be obtained based on the time series deformation using the least squares method; the deformation rate map is passed through the improved pyramid scene analysis network to obtain a feature map, and the feature map is subjected to pooling, upsampling, and convolution operations to automatically identify surface landslide settlement risk areas.

[0193] Example 2

[0194] The present invention provides a specific embodiment of a method for automatically detecting surface landslide and settlement risk areas, as follows:

[0195] The experimental areas for verifying the effectiveness of the method proposed in this special topic are Shijiazhuang and Baoding in Hebei Province, such as Figure 6 The following figure shows the study area and the Sentinel-1 data coverage. The area is characterized by typical plains, low hills, and mid-mountainous areas. The vertical differences in the mountains are significant, providing the basic conditions for the occurrence of geological disasters such as landslides. The area is also rich in power transmission facilities, and the transmission corridor runs through the mountainous area, providing an excellent test area for verifying the validity of this patent.

[0196] In this experiment, we collected a total of 188 Sentinel-1 single-view complex image data from January 2020 to March 2023 to obtain surface deformation in Shijiazhuang, Hebei Province and other areas. After determining the monitoring area, we selected a small baseline set with a short time interval of 12 days and a small spatial baseline of less than 200 meters according to the terrain and monitoring needs to ensure the coherence and accuracy of the data for subsequent processing; and based on the small baseline set, we determined the surface deformation rate map. Based on the surface deformation rate map, we used the improved PSPNet deep learning network to automatically monitor the subsidence risk area to implement the landslide deformation risk area detection task.

[0197] like Figure 7 The figure shows the deformation rate map of the study area from January 2020 to March 2023 obtained based on SBAS-InSAR technology. By statistically analyzing the deformation rate, we found that 95.4% of the monitoring targets have a deformation rate distribution of -25 to 25 mm / year. In the entire study area, we found that obvious local subsidence is mainly concentrated in areas A and B in the figure, and the regional power grid is dense. We found that these subsidences have obvious local subsidence characteristics, that is, there are characteristics of subsidence funnels. In addition, we found that there are more obvious trend deformation characteristics in area A, which is important for interpreting Figure 7 The localized sedimentation funnels shown here pose certain challenges. To automate the identification of these localized sedimentation features, we created samples based on these funnels and then used the PSPNet deep learning network for identification.

[0198] Finally, we used the trained improved PSPNet to obtain 40350km in the entire study area. 2 Automatically detected within the range of 109.94km 2Subsidence risk areas. A large number of obvious local subsidence areas were found around the power grid corridors, which will bring risks to power facilities. The final experimental results show that the method proposed in this paper shows good applicability and result accuracy for different test areas; in this invention, we propose a method for automatic detection of subsidence risks in power grid corridors based on the pyramid scene analysis network PSPNet and synthetic aperture radar interferometry InSAR technology. This method extracts multi-scale contextual information from the InSAR deformation rate map through the pyramid pooling module, and can identify subsidence risk areas at different scales. We conducted experiments in Shijiazhuang and Baoding, Hebei Province, and other areas to verify the effectiveness of the method proposed in this patent. A total of 188 scenes of Sentinel-1 single-view complex SAR image data from January 2020 to March 2023 were used, covering an area of ​​40,350 square kilometers. A total of 109.94 square kilometers of subsidence risk areas were successfully and automatically detected, and the model accuracy rate reached 90.72%. Experimental results demonstrate that this method not only effectively detects localized subsidence risks across a large area, but also overcomes the difficulty of traditional methods in detecting areas with trending deformation, demonstrating high accuracy and adaptability. The proposed technique has significant application value for the safety monitoring of power facilities and provides a simplified and efficient subsidence risk identification tool for non-InSAR professionals.

[0199] Example 3:

[0200] Based on the same inventive concept, the present invention patent application also provides a surface landslide settlement risk area automatic detection system such as Figure 8 As shown, it includes: satellite data acquisition module, surface deformation rate map extraction module and surface landslide settlement risk area identification module;

[0201] The satellite data acquisition module is used to collect satellite data of the target area;

[0202] The surface deformation rate map extraction module is used to obtain the surface deformation rate map of the target area based on the satellite data using a small baseline synthetic aperture radar interferometry technique;

[0203] The surface landslide settlement risk area identification module is used to identify the surface landslide settlement risk area in the target area based on the surface deformation rate map through a pre-built improved pyramid scenario analysis network;

[0204] The improved pyramid scene analysis network includes an improved inverted residual layer, a multi-pyramid pooling structure and a dilated pyramid pooling structure. The improved inverted residual layer is a depth-separable convolution structure composed of point-by-point convolution combined with depth-wise convolution.

[0205] Preferably, the system further comprises: a construction module of an improved pyramid scene analysis network, wherein the construction module of the improved pyramid scene analysis network comprises:

[0206] A historical satellite data processing submodule is used to obtain a historical surface deformation rate map and deformation labels of the historical surface deformation rate map based on historical satellite data of the target area;

[0207] A sample set generation submodule is used to use the historical surface deformation rate map and the deformation labels of the historical surface deformation rate map as a sample set;

[0208] A sample set division submodule, configured to divide the sample set into a training sample set and a validation sample set;

[0209] An iterative training submodule is used to iteratively train the improved inverted residual layer, multi-pyramid pooling structure and dilated pyramid pooling structure in the pyramid scene analysis network based on the training sample set to obtain a trained pyramid scene analysis network;

[0210] The model correction submodule is used to correct the trained pyramid scene analysis network based on the verification sample set to complete the construction of the pyramid scene analysis network.

[0211] Preferably, the historical satellite data processing submodule is specifically used to:

[0212] Based on historical satellite data of the target area, a small baseline synthetic aperture radar interferometry technique is used to obtain a historical surface deformation rate map;

[0213] Performing deformation identification on the historical surface deformation rate map to obtain a deformation label of the historical surface deformation rate map.

[0214] Preferably, the iterative training submodule includes:

[0215] an inverted residual layer training unit, configured to take the historical surface deformation rate map in the training sample set as input, perform efficient feature compression through the improved inverted residual layer, and obtain a deformation rate feature vector;

[0216] a deformation rate fusion feature map generating unit, configured to obtain multi-scale image feature information based on the deformation rate feature vector by using a multi-pyramid pooling structure and a dilated pyramid pooling structure, and generate a deformation rate fusion feature map according to the multi-scale image feature information;

[0217] an upsampling unit, configured to perform upsampling processing on the deformation rate fusion feature map by using a bilinear interpolation method to obtain a predicted deformation label of the historical surface deformation rate map;

[0218] The structural parameter updating unit is used to compare the predicted deformation label of the historical surface deformation rate map with the deformation label corresponding to the historical surface deformation rate map in the training sample set, update the weights of the improved inverted residual layer, the multi-pyramid pooling structure and the hollow pyramid pooling structure according to the comparison result, and iterate the training number until the number of training iterations reaches a preset number of training iterations, thereby completing the training of the pyramid scene analysis network.

[0219] Preferably, the point-by-point convolution includes a first point-by-point convolution and a second point-by-point convolution; the inverted residual layer training unit is specifically used to:

[0220] Based on the historical surface deformation rate map in the training sample set, performing deformation rate feature expansion through the first point-by-point convolution to obtain a first point-by-point convolution feature map;

[0221] Based on the first point-by-point convolution feature map, performing a convolution operation through the depthwise convolution to obtain a depthwise convolution feature map;

[0222] Based on the depth convolution feature map, a deformation rate feature vector is obtained through the second point-by-point convolution.

[0223] Preferably, the satellite data acquisition module includes:

[0224] A data screening submodule, configured to screen out a plurality of synthetic aperture radar images from the satellite data;

[0225] A small baseline set generation submodule is configured to pair the plurality of synthetic aperture radar images in pairs using preset time and space baseline limits to generate a plurality of small baseline sets;

[0226] A disentangled interferogram generation submodule is used to generate a disentangled interferogram corresponding to each small baseline set using a preset baseline threshold;

[0227] The surface deformation rate extraction submodule is used to perform interference phase unwrapping on each unwrapped interferogram, form an interference phase matrix with the obtained unwrapped interference phases, and obtain the surface deformation rate based on the interference phase matrix using the least squares method;

[0228] The surface deformation rate map generating submodule is used to generate a surface deformation rate map based on the surface deformation rate.

[0229] Preferably, the surface deformation rate extraction submodule is specifically used to:

[0230] Calculating the unwrapping interference phase of each unwrapping interference pattern using an unwrapping interference phase calculation formula based on unwrapping interference phase related parameters;

[0231] generating an interference phase matrix according to all the unwrapped interference phases;

[0232] The least square method is used for inversion to extract a surface deformation time series from the interference phase matrix, and the surface deformation rate is obtained based on the surface deformation time series.

[0233] Preferably, the unwrapping interference phase calculation formula is as follows:

[0234]

[0235] Among them, Δφ j is the unwrapped interference phase of the jth unwrapped interference pattern, IU j is the starting index of the interference pair measurement time period of the jth unwrapped interferogram, IE j is the end index of the interference pair measurement time period of the jth unwrapped interferogram, k is the measurement time of the interference pair measurement time period, t k is the kth moment in the interferometer measurement period, t k-1 is the k-1 moment in the interferometer measurement time period, v k For t k The surface velocity deformation rate at .

[0236] Example 4

[0237] like Figure 9 As shown, the present invention also provides an electronic device, which may be a computer, a single-chip microcomputer, a smart mobile device, or the like. The electronic device in this embodiment may include a processor, a memory, a transceiver component, and the like. The memory, processor, and transceiver component are connected via a bus; the memory may be used to store an execution program, which may include instructions; and the processor may be used to execute the instructions stored in the memory. The memory may also be used to store data, which may be accessed and / or modified during the execution of the instructions.

[0238] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions in the storage medium to implement the corresponding method flow or corresponding function, so as to realize the steps of a method for automatically detecting a surface landslide settlement risk area in the above embodiment.

[0239] Example 5

[0240] Based on the same inventive concept, the present invention also provides a readable storage medium, specifically an electronic device readable storage medium (Memory), which is a memory device in the electronic device for storing programs and data. It can be understood that the storage medium here can include both the built-in storage medium in the electronic device and, of course, the extended storage medium supported by the electronic device. The storage medium provides a storage space, which stores the operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space. These instructions can be one or more execution programs (including program codes). It should be noted that the storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk memory. The processor loads and executes one or more instructions stored in the storage medium, which can implement the steps of a method for automatically detecting surface landslide settlement risk areas in the above embodiment.

[0241] It will be understood by those skilled in the art that the embodiments of the present invention patent application may be provided as a method, system, or computer program product. Therefore, the present invention patent application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention patent application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0242] The present invention patent application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present invention patent application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0243] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0244] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0245] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the patent application of the present invention and are not intended to limit its scope of protection. Although the patent application of the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that after reading the patent application of the present invention, those skilled in the art can still make various changes, modifications or equivalent substitutions to the specific implementation methods of the application, but these changes, modifications or equivalent substitutions are all within the scope of protection of the claims to be approved.

Claims

1. A method for automatically detecting surface landslide and settlement risk areas, characterized in that: include: Collect satellite data of the target area; Based on the satellite data, a small baseline synthetic aperture radar interferometry technique is used to obtain a surface deformation rate map of the target area; Based on the surface deformation rate map, identifying the surface landslide settlement risk area in the target area through a pre-built improved pyramid scenario analysis network; The improved pyramid scene analysis network includes an improved inverted residual layer, a multi-pyramid pooling structure and a dilated pyramid pooling structure. The improved inverted residual layer is a depth-separable convolution structure composed of point-by-point convolution combined with depth-wise convolution.

2. The method according to claim 1, wherein The construction of the improved pyramid scene analysis network includes: Based on the historical satellite data of the target area, a historical surface deformation rate map and deformation labels of the historical surface deformation rate map are obtained; The historical surface deformation rate map and the deformation labels of the historical surface deformation rate map are used as the sample set; Dividing the sample set into a training sample set and a validation sample set; Based on the training sample set, iteratively training the improved inverted residual layer, multi-pyramid pooling structure and dilated pyramid pooling structure in the pyramid scene analysis network to obtain a trained pyramid scene analysis network; Based on the verification sample set, the trained pyramid scene analysis network is corrected to complete the construction of the pyramid scene analysis network.

3. The method according to claim 2, wherein The method of obtaining a historical surface deformation rate map and a deformation label of the historical surface deformation rate map based on historical satellite data of the target area includes: Based on historical satellite data of the target area, a small baseline synthetic aperture radar interferometry technique is used to obtain a historical surface deformation rate map; Performing deformation identification on the historical surface deformation rate map to obtain a deformation label of the historical surface deformation rate map.

4. The method according to claim 2, wherein The improved inverted residual layer, multi-pyramid pooling structure and dilated pyramid pooling structure in the pyramid scene analysis network are iteratively trained based on the training sample set to obtain a trained pyramid scene analysis network, including: Taking the historical surface deformation rate map in the training sample set as input, the improved inverted residual layer is used to perform efficient feature compression to obtain a deformation rate feature vector; Based on the deformation rate feature vector, multi-scale image feature information is obtained through a multi-pyramid pooling structure and a dilated pyramid pooling structure, and a deformation rate fusion feature map is generated according to the multi-scale image feature information; Using a bilinear interpolation method, upsampling the deformation rate fusion feature map is performed to obtain a predicted deformation label of the historical surface deformation rate map; The predicted deformation labels of the historical surface deformation rate map are compared with the deformation labels corresponding to the historical surface deformation rate map in the training sample set. Based on the comparison results, the weights of the improved inverted residual layer, the multi-pyramid pooling structure, and the dilated pyramid pooling structure are updated. The training is iterated until the number of training iterations reaches a preset number of training iterations, thereby completing the training of the pyramid scene analysis network.

5. The method according to claim 4, wherein The point-by-point convolution includes a first point-by-point convolution and a second point-by-point convolution; the historical surface deformation rate map in the training sample set is used as input, and efficient feature compression is performed through the improved inverted residual layer to obtain a deformation rate feature vector, including: Based on the historical surface deformation rate map in the training sample set, performing deformation rate feature expansion through the first point-by-point convolution to obtain a first point-by-point convolution feature map; Based on the first point-by-point convolution feature map, performing a convolution operation through the depthwise convolution to obtain a depthwise convolution feature map; Based on the depth convolution feature map, a deformation rate feature vector is obtained through the second point-by-point convolution.

6. The method according to claim 1, wherein The method of obtaining a surface deformation rate map of the target area based on the satellite data by using a small baseline synthetic aperture radar interferometry technique includes: selecting a plurality of synthetic aperture radar images from the satellite data; Using pre-set temporal and spatial baseline limits, the plurality of synthetic aperture radar images are paired in pairs to generate a plurality of small baseline sets; A pre-set baseline threshold is used to generate the disentangled interferogram corresponding to each small baseline set; After performing interference phase unwrapping on each unwrapped interferogram, the obtained unwrapped interference phases are combined into an interference phase matrix, and the surface deformation rate is obtained based on the interference phase matrix using the least squares method; Based on the surface deformation rate, a surface deformation rate map is generated.

7. The method according to claim 6, wherein After performing interference phase unwrapping on each unwrapped interferogram, the obtained unwrapped interference phases are combined into an interference phase matrix. Based on the interference phase matrix, the surface deformation rate is obtained using the least squares method, including: Calculating the unwrapping interference phase of each unwrapping interference pattern using an unwrapping interference phase calculation formula based on unwrapping interference phase related parameters; generating an interference phase matrix according to all the unwrapped interference phases; The least square method is used for inversion to extract a surface deformation time series from the interference phase matrix, and the surface deformation rate is obtained based on the surface deformation time series.

8. The method according to claim 7, wherein The unwrapping interference phase calculation formula is as follows: Among them, Δφ j is the unwrapped interference phase of the jth unwrapped interference pattern, IU j is the starting index of the interference pair measurement time period of the jth unwrapped interferogram, IE j is the end index of the interference pair measurement time period of the jth unwrapped interferogram, k is the measurement time of the interference pair measurement time period, t k is the kth moment in the interferometer measurement period, t k-1 is the k-1 moment in the interferometer measurement time period, v k For t k The surface velocity deformation rate at .

9. An automatic detection system for surface landslide and settlement risk areas, characterized in that: include: Satellite data acquisition module, surface deformation rate map extraction module and surface landslide settlement risk area identification module; The satellite data acquisition module is used to collect satellite data of the target area; The surface deformation rate map extraction module is used to obtain the surface deformation rate map of the target area based on the satellite data using a small baseline synthetic aperture radar interferometry technique; The surface landslide settlement risk area identification module is used to identify the surface landslide settlement risk area in the target area based on the surface deformation rate map through a pre-built improved pyramid scenario analysis network; The improved pyramid scene analysis network includes an improved inverted residual layer, a multi-pyramid pooling structure and a dilated pyramid pooling structure. The improved inverted residual layer is a depth-separable convolution structure composed of point-by-point convolution combined with depth-wise convolution.

10. The system according to claim 9, wherein: The system further includes: an improved pyramid scene analysis network construction module, the improved pyramid scene analysis network construction module including: A historical satellite data processing submodule is used to obtain a historical surface deformation rate map and deformation labels of the historical surface deformation rate map based on historical satellite data of the target area; A sample set generation submodule is used to use the historical surface deformation rate map and the deformation labels of the historical surface deformation rate map as a sample set; A sample set division submodule, configured to divide the sample set into a training sample set and a validation sample set; An iterative training submodule is used to iteratively train the improved inverted residual layer, multi-pyramid pooling structure and dilated pyramid pooling structure in the pyramid scene analysis network based on the training sample set to obtain a trained pyramid scene analysis network; The model correction submodule is used to correct the trained pyramid scene analysis network based on the verification sample set to complete the construction of the pyramid scene analysis network.

11. The system according to claim 10, wherein: The historical satellite data processing submodule is specifically used to: Based on historical satellite data of the target area, a small baseline synthetic aperture radar interferometry technique is used to obtain a historical surface deformation rate map; Performing deformation identification on the historical surface deformation rate map to obtain a deformation label of the historical surface deformation rate map.

12. The system according to claim 10, wherein: The iterative training submodule includes: an inverted residual layer training unit, configured to take the historical surface deformation rate map in the training sample set as input, perform efficient feature compression through the improved inverted residual layer, and obtain a deformation rate feature vector; a deformation rate fusion feature map generating unit, configured to obtain multi-scale image feature information based on the deformation rate feature vector by using a multi-pyramid pooling structure and a dilated pyramid pooling structure, and generate a deformation rate fusion feature map according to the multi-scale image feature information; an upsampling unit, configured to perform upsampling processing on the deformation rate fusion feature map by using a bilinear interpolation method to obtain a predicted deformation label of the historical surface deformation rate map; The structural parameter updating unit is used to compare the predicted deformation label of the historical surface deformation rate map with the deformation label corresponding to the historical surface deformation rate map in the training sample set, update the weights of the improved inverted residual layer, the multi-pyramid pooling structure and the hollow pyramid pooling structure according to the comparison result, and iterate the training number until the number of training iterations reaches a preset number of training iterations, thereby completing the training of the pyramid scene analysis network.

13. The system according to claim 12, wherein: The point-by-point convolution includes a first point-by-point convolution and a second point-by-point convolution; the inverted residual layer training unit is specifically used to: Based on the historical surface deformation rate map in the training sample set, performing deformation rate feature expansion through the first point-by-point convolution to obtain a first point-by-point convolution feature map; Based on the first point-by-point convolution feature map, performing a convolution operation through the depthwise convolution to obtain a depthwise convolution feature map; Based on the depth convolution feature map, a deformation rate feature vector is obtained through the second point-by-point convolution.

14. The system according to claim 9, wherein: The satellite data acquisition module includes: A data screening submodule, configured to screen out a plurality of synthetic aperture radar images from the satellite data; A small baseline set generation submodule is configured to pair the plurality of synthetic aperture radar images in pairs using preset time and space baseline limits to generate a plurality of small baseline sets; A disentangled interferogram generation submodule is used to generate a disentangled interferogram corresponding to each small baseline set using a preset baseline threshold; The surface deformation rate extraction submodule is used to perform interference phase unwrapping on each unwrapped interferogram, form an interference phase matrix with the obtained unwrapped interference phases, and obtain the surface deformation rate based on the interference phase matrix using the least squares method; The surface deformation rate map generating submodule is used to generate a surface deformation rate map based on the surface deformation rate.

15. The system according to claim 14, wherein: The surface deformation rate extraction submodule is specifically used to: Calculating the unwrapping interference phase of each unwrapping interference pattern using an unwrapping interference phase calculation formula based on unwrapping interference phase related parameters; generating an interference phase matrix according to all the unwrapped interference phases; The least square method is used for inversion to extract a surface deformation time series from the interference phase matrix, and the surface deformation rate is obtained based on the surface deformation time series.

16. The system according to claim 15, wherein: The unwrapping interference phase calculation formula is as follows: Among them, Δφ j is the unwrapped interference phase of the jth unwrapped interference pattern, IU j is the starting index of the interference pair measurement time period of the jth unwrapped interferogram, IE j is the end index of the interference pair measurement time period of the jth unwrapped interferogram, k is the measurement time of the interference pair measurement time period, t k is the kth moment in the interferometer measurement period, t k-1 is the k-1 moment in the interferometer measurement time period, v k For t k The surface velocity deformation rate at .

17. An electronic device, characterized in that: include: at least one processor and memory; The memory and the processor are connected via a bus; The memory is used to store one or more programs; When the one or more programs are executed by the at least one processor, the method for automatically detecting surface landslide settlement risk areas according to any one of claims 1 to 8 is implemented.

18. A readable storage medium, characterized in that: An execution program is stored thereon, and when the execution program is executed, an automatic detection method for surface landslide settlement risk areas according to any one of claims 1 to 8 is implemented.

Images