Wide-area remote sensing detection method for highway facilities in alpine region

Abstract

The invention provides a highway facility wide-area remote sensing detection method for alpine regions, and the method achieves the automatic recognition of hidden troubles, such as subgrade settlement, cracks, slope slippage, bridge deformation, and the like, through the fusion of an optical remote sensing image and an SAR remote sensing image. The method comprises the following steps: firstly, carrying out ortho-rectification, filtering processing and mutual information joint registration on an optical image and an SAR image to ensure that multi-source data is aligned at a pixel level; then, training samples are marked by utilizing road vector data and historical hidden danger pattern spots, and the sample size is enhanced and expanded through single-sample and multi-sample data; in model construction, an optical branch and an SAR branch extract low-level and high-level features respectively, a cross attention fusion module fuses complementary information of the optical branch and the SAR branch, and a cavity space pyramid further improves the multi-scale perception capability. And finally, completing classification and positioning by adopting multi-stage regression and a classifier, and additionally arranging a corner reflector in an ice and snow area or a low-coherence area so as to improve the identification precision.

Description

Technical Field

[0001] The present invention relates to the field of remote sensing technology and transportation infrastructure monitoring, and in particular to a wide-area remote sensing detection method for highway facilities in high-cold areas based on the fusion of optical remote sensing images and SAR remote sensing images, which can realize multi-source data monitoring, hidden danger detection and safety assessment of highway roads, slopes, bridges and facilities along the roads in high-cold areas. Background Art

[0002] In high-altitude cold regions, highway infrastructure faces multiple challenges such as frozen soil, extreme climate, and snow cover. Common problems include:

[0003] Uneven settlement, frost heave or cracking of the roadbed;

[0004] Slope instability or collapse, with potential risk of landslide or debris flow;

[0005] Snow and ice accumulation causing road damage or bridge fatigue;

[0006] The inspection cycle is long, the cost is high, and manual troubleshooting is difficult.

[0007] At present, the common inspection methods for highway facilities in high-altitude and cold areas mainly include:

[0008] Manual inspection: along accessible or accessible roads, the road surface, slopes, bridges, etc. are inspected manually or with ground equipment. However, due to the influence of geographical environment and weather, the inspection scope is limited and the efficiency is low.

[0009] Traditional single remote sensing methods: such as using only optical images to identify visible features such as road cracks and collapses. However, due to long-term snow and fog in high-altitude areas, the available time window of optical images is limited, which can easily lead to missed detections or false detections; or only using SAR images for deformation monitoring, but the lack of coherence in forest vegetation and frozen soil environments will also affect the accuracy of the results.

[0010] In view of this, there is an urgent need for a comprehensive monitoring method that integrates optical remote sensing images and SAR remote sensing images to achieve large-scale, high-precision detection and hidden danger assessment of highway facilities under high-altitude and harsh climatic conditions, which can not only reduce the cost of manual inspections, but also improve the ability to detect potential risks early. Summary of the invention

[0011] In order to solve the problems of low inspection efficiency, limited monitoring accuracy and severe interference from ice and snow in highway facilities in high-cold areas, this paper proposes a wide-area remote sensing detection method for highway facilities in high-cold areas based on the fusion of optical remote sensing images and SAR remote sensing images. This method can identify and extract roadbed settlement, cracks, landslide hazards and bridge structure deformation under the complementary advantages of multi-source images, providing technical support for highway infrastructure safety management and disaster emergency response.

[0012] To achieve the above object, the present invention provides the following technical solutions, which mainly include:

[0013] The wide-area remote sensing detection method for highway facilities in high-cold areas includes the following steps:

[0014] 1) Data acquisition and preprocessing:

[0015] Obtain optical remote sensing images and SAR remote sensing images covering highways in high-cold areas and areas along the roads, perform orthorectification, radiation correction, atmospheric correction and image enhancement on the optical remote sensing images, perform complex data conversion, multi-view processing, filtering processing and geocoding on the SAR remote sensing images, and obtain pre-processed optical remote sensing images and SAR remote sensing images;

[0016] The preprocessed optical remote sensing image and SAR remote sensing image are jointly registered by using mutual information metric combined with simulated annealing algorithm.

[0017] 2) Training sample construction:

[0018] Based on the existing vector data of highways and surrounding facilities, on-site inspection records or historical hidden dangers, normal highway areas and potential damaged or hidden danger areas are marked on the optical remote sensing images and SAR remote sensing images after joint registration to obtain preliminary training samples.

[0019] Performing single sample data enhancement and multi-sample data enhancement on the preliminary training samples to form a training set, a test set, and a validation set;

[0020] 3) Deep fusion model construction and training:

[0021] A dual encoder structure including an optical branch and a SAR branch is established, each consisting of a feature extraction module and a dilated spatial pyramid module, and a cross-attention feature fusion module is set between the low-level features and high-level features of the optical branch and the SAR branch;

[0022] The enhanced training samples are input into the dual encoder structure for end-to-end training, and the damaged highway facilities, slope sliding, bridge deformation, etc. are identified through multi-level regression and classifier;

[0023] Iteratively update the model parameters according to the training error to obtain a trained detection model;

[0024] 4) Highway facility detection:

[0025] Obtain the optical remote sensing image and the SAR remote sensing image of the area to be monitored, pre-process them respectively, and jointly register them;

[0026] Using a sliding window or block processing method, the optical remote sensing image to be detected and the SAR remote sensing image to be detected are input into the detection model, and the information including the location and category of highway subgrade settlement, cracks, slope collapse, and bridge deformation is output;

[0027] The output results are spliced ​​and post-processed to obtain a wide-area detection result map of highway facilities in high-altitude and cold regions.

[0028] In a specific embodiment, the step of preprocessing the optical remote sensing image and the SAR remote sensing image includes:

[0029] Perform orthorectification, radiation correction, atmospheric correction and image enhancement on optical remote sensing images;

[0030] Perform complex data conversion, multi-view processing, filtering and geocoding on SAR remote sensing images;

[0031] When the acquired images are covered by ice or snow or have high cloud cover, cloud removal, snow removal or other masking strategies are further used to improve the quality of effective pixels.

[0032] In a specific embodiment, the joint registration step includes:

[0033] Calculate the mutual information between the preprocessed optical remote sensing image and the SAR remote sensing image at different locations;

[0034] Searching for matching points in the distance and directional gradient space by using a simulated annealing algorithm, and performing meshing and optimal triangulation on the optical remote sensing image and the SAR remote sensing image;

[0035] Affine transformation is used to perform fine registration inside the grid to ensure that multi-source images are aligned at the pixel level.

[0036] In a specific embodiment, the training sample construction includes:

[0037] Using highway vector layers, bridge locations, slope distribution, etc. as references, a buffer zone is established in the co-registered optical remote sensing image and SAR remote sensing image.

[0038] Use pixel-level annotation to annotate normal pavement, cracks, collapse, landslide, bridge deformation and other types to form multi-classification labels;

[0039] Single-sample data enhancement includes image flipping, cropping, rotation, adding noise, and random color change; multi-sample data enhancement includes randomly superimposing or mixing two training samples to obtain new training samples.

[0040] In a specific embodiment, the deep fusion model includes:

[0041] The optical branch and the SAR branch are respectively provided with a feature extraction module and a hole space pyramid module, which are used to extract optical high-level features, optical low-level features, SAR high-level features, and SAR low-level features;

[0042] A cross-attention feature fusion module is set at the output end of the optical branch and the SAR branch to fuse the low-level features of the optical and SAR to obtain low-level fusion features, and to fuse the high-level features of the optical and SAR to obtain high-level fusion features;

[0043] After the low-level fusion features are aggregated with the high-level fusion features through the high-level and low-level feature aggregation module, they are input into the multi-level regression and classifier to identify the damage types of highway facilities.

[0044] In a specific embodiment, the atrous spatial pyramid module includes a plurality of atrous convolutions with different atrous rates for multi-scale feature extraction;

[0045] The multi-scale features are concatenated after dimensionality reduction through 1×1 convolution to obtain optical high-level features and SAR high-level features respectively, and the interaction between optical and SAR high-level features is realized in the cross-attention feature fusion module.

[0046] In a specific embodiment, the cross-attention feature fusion module includes:

[0047] Perform 1×1 convolution operations on the optical features and SAR features respectively to obtain the optical value matrix, optical keyword matrix, optical query matrix, as well as the SAR value matrix, SAR keyword matrix, and SAR query matrix;

[0048] The joint attention score is calculated by the optical self-attention score and the SAR self-attention score, and the matrix dot multiplication operation is performed with the optical value matrix and the SAR value matrix respectively to generate the weighted optical features and SAR features;

[0049] The weighted optical features are multiplied again with the SAR features to obtain fused features for subsequent aggregation.

[0050] In a specific embodiment, the highway facility hidden danger information output by the multi-level regression and classifier includes:

[0051] Location and classification of roadbed settlement, cracks, and fault zones;

[0052] Monitoring of areas prone to slope sliding, landslide or debris flow;

[0053] Detection results of bridge beam deformation or accessory structure damage;

[0054] Identification of safety risks of other facilities along the route.

[0055] In a specific embodiment, the step of correcting seasonal ice and snow interference in high-cold areas is also included:

[0056] While acquiring the image to be detected, determine the snow and ice coverage based on meteorological and snowmelt information;

[0057] Mask or remove the snow-covered areas to avoid misjudging the snow-covered areas as surface cracks or collapses.

[0058] Feature extraction and model detection are performed in the remaining valid area to improve detection accuracy.

[0059] In a specific embodiment, for areas in high-cold regions where the surface scattering coherence is insufficient, the step of deploying corner reflectors is also included:

[0060] Install movable or lightweight corner reflectors near slopes or important bridges with high monitoring requirements to enhance the coherence of SAR image echo signals;

[0061] The high-precision deformation monitoring results of the corresponding pixels of the corner reflector are combined with the optical features to further improve the early identification capability of potential hazards.

[0062] It can be seen from the above technical solution that compared with the prior art, the above technical solution has the following advantages:

[0063] Multi-source data complementation improves detection accuracy

[0064] This method uses the fusion of optical remote sensing images and SAR remote sensing images, making full use of the ability of optical data to capture texture, color, shape and other features in the visible light / near-infrared bands, as well as the advantages of SAR remote sensing in all-weather, cloud-penetrating and rain-penetrating, and detecting tiny surface deformations.

[0065] Multi-source data complementarity can effectively reduce missed detections or false detections caused by a single data source (such as the interference of long-term snow / cloud cover on optical images and the impact of surface scattering differences on SAR coherence) in harsh high-altitude environments, and significantly improve the accuracy and stability of highway facility detection.

[0066] Adapt to high-cold environment and enhance the ability to suppress ice and snow interference

[0067] The method introduces cloud removal, snow removal or other masking strategies specifically for periods of ice and snow cover and high cloud cover, and makes corrections based on seasonal snowmelt information.

[0068] In areas with severe ice and snow cover or insufficient coherence, lightweight corner reflectors can be deployed to improve the quality of SAR imaging, thereby enabling stable data acquisition and detailed interpretation even in high-altitude and cold areas.

[0069] Automation and wide-scale application reduce the cost of manual inspection

[0070] By establishing a technical process of "data acquisition and preprocessing - model training - automatic identification - splicing / post-processing", we can achieve automated batch detection of various hidden dangers along the highway (roadbed settlement, cracks, landslides, bridge deformation, etc.).

[0071] Compared with the traditional method that relies on manual investigation or single visual interpretation, it can greatly shorten the investigation cycle, reduce manpower investment, and has higher promotion value on large-scale routes in high-altitude and cold areas.

[0072] Visualization and quantitative output to assist public management and decision-making

[0073] Through multi-level regression and classifier output, different types of hidden dangers (such as landslides, debris flow prone areas, bridge component damage, etc.) are classified and located to form an intuitive distribution map.

[0074] It provides the transportation department or emergency management department with the technical basis for quickly discovering hidden dangers and allocating resources in a prioritized manner, thereby improving the scientificity and timeliness of preventive maintenance and disaster emergency response. DETAILED DESCRIPTION

[0075] The technical solutions in the embodiments of the present invention are described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0076] Embodiment 1

[0077] The wide-area remote sensing detection method for highway facilities in high-cold areas includes the following steps:

[0078] 1) Data acquisition and preprocessing:

[0079] Obtain optical remote sensing images and SAR remote sensing images covering highways in high-cold areas and areas along the roads, perform orthorectification, radiation correction, atmospheric correction and image enhancement on the optical remote sensing images, perform complex data conversion, multi-view processing, filtering processing and geocoding on the SAR remote sensing images, and obtain pre-processed optical remote sensing images and SAR remote sensing images;

[0080] The preprocessed optical remote sensing image and SAR remote sensing image are jointly registered by using mutual information metric combined with simulated annealing algorithm.

[0081] 2) Training sample construction:

[0082] Based on the existing vector data of highways and surrounding facilities, on-site inspection records or historical hidden dangers, normal highway areas and potential damaged or hidden danger areas are marked on the optical remote sensing images and SAR remote sensing images after joint registration to obtain preliminary training samples.

[0083] Performing single sample data enhancement and multi-sample data enhancement on the preliminary training samples to form a training set, a test set, and a validation set;

[0084] 3) Deep fusion model construction and training:

[0085] A dual encoder structure including an optical branch and a SAR branch is established, each consisting of a feature extraction module and a dilated spatial pyramid module, and a cross-attention feature fusion module is set between the low-level features and high-level features of the optical branch and the SAR branch;

[0086] The enhanced training samples are input into the dual encoder structure for end-to-end training, and the damaged highway facilities, slope sliding, bridge deformation, etc. are identified through multi-level regression and classifier;

[0087] Iteratively update the model parameters according to the training error to obtain a trained detection model;

[0088] 4) Highway facility detection:

[0089] Obtain the optical remote sensing image and the SAR remote sensing image of the area to be monitored, pre-process them respectively, and jointly register them;

[0090] Using a sliding window or block processing method, the optical remote sensing image to be detected and the SAR remote sensing image to be detected are input into the detection model, and the information including the location and category of highway subgrade settlement, cracks, slope collapse, and bridge deformation is output;

[0091] The output results are spliced ​​and post-processed to obtain a wide-area detection result map of highway facilities in high-altitude and cold regions.

[0092] In a specific embodiment, the step of preprocessing the optical remote sensing image and the SAR remote sensing image includes:

[0093] Perform orthorectification, radiation correction, atmospheric correction and image enhancement on optical remote sensing images;

[0094] Perform complex data conversion, multi-view processing, filtering and geocoding on SAR remote sensing images;

[0095] When the acquired images are covered by ice or snow or have high cloud cover, cloud removal, snow removal or other masking strategies are further used to improve the quality of effective pixels.

[0096] In a specific embodiment, the joint registration step includes:

[0097] Calculate the mutual information between the preprocessed optical remote sensing image and the SAR remote sensing image at different locations;

[0098] Searching for matching points in the distance and directional gradient space by using a simulated annealing algorithm, and performing meshing and optimal triangulation on the optical remote sensing image and the SAR remote sensing image;

[0099] Affine transformation is used to perform fine registration inside the grid to ensure that multi-source images are aligned at the pixel level.

[0100] In a specific embodiment, the training sample construction includes:

[0101] Using highway vector layers, bridge locations, slope distribution, etc. as references, a buffer zone is established in the co-registered optical remote sensing image and SAR remote sensing image.

[0102] Use pixel-level annotation to annotate normal pavement, cracks, collapse, landslide, bridge deformation and other types to form multi-classification labels;

[0103] Single-sample data enhancement includes image flipping, cropping, rotation, adding noise, and random color change; multi-sample data enhancement includes randomly superimposing or mixing two training samples to obtain new training samples.

[0104] In a specific embodiment, the deep fusion model includes:

[0105] The optical branch and the SAR branch are respectively provided with a feature extraction module and a hole space pyramid module, which are used to extract optical high-level features, optical low-level features, SAR high-level features, and SAR low-level features;

[0106] A cross-attention feature fusion module is set at the output end of the optical branch and the SAR branch to fuse the low-level features of the optical and SAR to obtain low-level fusion features, and to fuse the high-level features of the optical and SAR to obtain high-level fusion features;

[0107] After the low-level fusion features are aggregated with the high-level fusion features through the high-level and low-level feature aggregation module, they are input into the multi-level regression and classifier to identify the damage types of highway facilities.

[0108] In a specific embodiment, the atrous spatial pyramid module includes a plurality of atrous convolutions with different atrous rates for multi-scale feature extraction;

[0109] The multi-scale features are concatenated after dimensionality reduction through 1×1 convolution to obtain optical high-level features and SAR high-level features respectively, and the interaction between optical and SAR high-level features is realized in the cross-attention feature fusion module.

[0110] In a specific embodiment, the cross-attention feature fusion module includes:

[0111] Perform 1×1 convolution operations on the optical features and SAR features respectively to obtain the optical value matrix, optical keyword matrix, optical query matrix, as well as the SAR value matrix, SAR keyword matrix, and SAR query matrix;

[0112] The joint attention score is calculated by the optical self-attention score and the SAR self-attention score, and the matrix dot multiplication operation is performed with the optical value matrix and the SAR value matrix respectively to generate the weighted optical features and SAR features;

[0113] The weighted optical features are multiplied again with the SAR features to obtain fused features for subsequent aggregation.

[0114] In a specific embodiment, the highway facility hidden danger information output by the multi-level regression and classifier includes:

[0115] Location and classification of roadbed settlement, cracks, and fault zones;

[0116] Monitoring of areas prone to slope sliding, landslide or debris flow;

[0117] Detection results of bridge beam deformation or accessory structure damage;

[0118] Identification of safety risks of other facilities along the route.

[0119] In a specific embodiment, the step of correcting seasonal ice and snow interference in high-cold areas is also included:

[0120] While acquiring the image to be detected, determine the snow and ice coverage based on meteorological and snowmelt information;

[0121] Mask or remove the snow-covered areas to avoid misjudging the snow-covered areas as surface cracks or collapses.

[0122] Feature extraction and model detection are performed in the remaining valid area to improve detection accuracy.

[0123] In a specific embodiment, for areas in high-cold regions where the surface scattering coherence is insufficient, the step of deploying corner reflectors is also included:

[0124] Install movable or lightweight corner reflectors near slopes or important bridges with high monitoring requirements to enhance the coherence of SAR image echo signals;

[0125] The high-precision deformation monitoring results of the corresponding pixels of the corner reflector are combined with the optical features to further improve the early identification capability of potential hazards.

[0126] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0127] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A wide-area remote sensing detection method for highway facilities in high-cold areas, characterized in that: The following steps are involved: 1) Data acquisition and preprocessing: Obtain optical remote sensing images and SAR remote sensing images covering highways in high-cold areas and areas along the roads, perform orthorectification, radiation correction, atmospheric correction and image enhancement on the optical remote sensing images, perform complex data conversion, multi-view processing, filtering processing and geocoding on the SAR remote sensing images, and obtain pre-processed optical remote sensing images and SAR remote sensing images; The preprocessed optical remote sensing image and SAR remote sensing image are jointly registered by using mutual information metric combined with simulated annealing algorithm. 2) Training sample construction: Based on the existing vector data of highways and surrounding facilities, on-site inspection records or historical hidden dangers, normal highway areas and potential damaged or hidden danger areas are marked on the optical remote sensing images and SAR remote sensing images after joint registration to obtain preliminary training samples. Performing single sample data enhancement and multi-sample data enhancement on the preliminary training samples to form a training set, a test set, and a validation set; 3) Deep fusion model construction and training: A dual encoder structure including an optical branch and a SAR branch is established, each consisting of a feature extraction module and a dilated spatial pyramid module, and a cross-attention feature fusion module is set between the low-level features and high-level features of the optical branch and the SAR branch; The enhanced training samples are input into the dual encoder structure for end-to-end training, and the damaged highway facilities, slope sliding, bridge deformation, etc. are identified through multi-level regression and classifier; Iteratively update the model parameters according to the training error to obtain a trained detection model; 4) Highway facility detection: Obtain the optical remote sensing image and the SAR remote sensing image of the area to be monitored, pre-process them respectively, and jointly register them; Using a sliding window or block processing method, the optical remote sensing image to be detected and the SAR remote sensing image to be detected are input into the detection model, and the information including the location and category of highway subgrade settlement, cracks, slope collapse, and bridge deformation is output; The output results are spliced ​​and post-processed to obtain a wide-area detection result map of highway facilities in high-altitude and cold regions.

2. The method according to claim 1, characterized in that The step of preprocessing the optical remote sensing image and the SAR remote sensing image comprises: Perform orthorectification, radiation correction, atmospheric correction and image enhancement on optical remote sensing images; Perform complex data conversion, multi-view processing, filtering and geocoding on SAR remote sensing images; When the acquired images are covered by ice or snow or have high cloud cover, cloud removal, snow removal or other masking strategies are further used to improve the quality of effective pixels.

3. The method according to claim 1 or 2, characterized in that: The joint registration step comprises: Calculate the mutual information between the preprocessed optical remote sensing image and the SAR remote sensing image at different locations; Searching for matching points in the distance and directional gradient space by using a simulated annealing algorithm, and performing meshing and optimal triangulation on the optical remote sensing image and the SAR remote sensing image; Affine transformation is used to perform fine registration inside the grid to ensure that multi-source images are aligned at the pixel level.

4. The method according to any one of claims 1 or 2, characterized in that The training sample construction includes: Using highway vector layers, bridge locations, slope distribution, etc. as references, a buffer zone is established in the co-registered optical remote sensing image and SAR remote sensing image. Use pixel-level annotation to annotate normal pavement, cracks, collapse, landslide, bridge deformation and other types to form multi-classification labels; Single-sample data enhancement includes image flipping, cropping, rotation, adding noise, and random color change; multi-sample data enhancement includes randomly superimposing or mixing two training samples to obtain new training samples.

5. The method according to any one of claims 1 or 2, characterized in that The deep fusion model includes: The optical branch and the SAR branch are respectively provided with a feature extraction module and a hole space pyramid module, which are used to extract optical high-level features, optical low-level features, SAR high-level features, and SAR low-level features; A cross-attention feature fusion module is set at the output end of the optical branch and the SAR branch to fuse the low-level features of the optical and SAR to obtain low-level fusion features, and to fuse the high-level features of the optical and SAR to obtain high-level fusion features; After the low-level fusion features are aggregated with the high-level fusion features through the high-level and low-level feature aggregation module, they are input into the multi-level regression and classifier to identify the damage types of highway facilities.

6. The method according to claim 5, characterized in that The dilated spatial pyramid module includes a plurality of dilated convolutions with different dilation rates for multi-scale feature extraction; The multi-scale features are concatenated after dimensionality reduction through 1×1 convolution to obtain optical high-level features and SAR high-level features respectively, and the interaction between optical and SAR high-level features is realized in the cross-attention feature fusion module.

7. The method according to claim 6, characterized in that The cross-attention feature fusion module includes: Perform 1×1 convolution operations on the optical features and SAR features respectively to obtain the optical value matrix, optical keyword matrix, optical query matrix, as well as the SAR value matrix, SAR keyword matrix, and SAR query matrix; The joint attention score is calculated by the optical self-attention score and the SAR self-attention score, and the matrix dot multiplication operation is performed with the optical value matrix and the SAR value matrix respectively to generate the weighted optical features and SAR features; The weighted optical features are multiplied again with the SAR features to obtain fused features for subsequent aggregation.

8. The method according to any one of claims 1, 2, 6 and 7, characterized in that: The highway facility hidden danger information output by the multi-level regression and classifier includes: Location and classification of roadbed settlement, cracks, and fault zones; Monitoring of areas prone to slope sliding, landslide or debris flow; Detection results of bridge beam deformation or accessory structure damage; Identification of safety risks of other facilities along the route.

9. The method according to any one of claims 1, 2, 6 and 7, characterized in that: It also includes steps to correct seasonal ice and snow disturbances in high-altitude areas: While acquiring the image to be detected, determine the snow and ice coverage based on meteorological and snowmelt information; Mask or remove the snow-covered areas to avoid misjudging the snow-covered areas as surface cracks or collapses. Feature extraction and model detection are performed in the remaining valid area to improve detection accuracy.

10. The method according to any one of claims 1, 2, 6 and 7, characterized in that: For areas in high-altitude cold regions where the surface scattering coherence is insufficient, the steps of laying out corner reflectors are also included: Install movable or lightweight corner reflectors near slopes or important bridges with high monitoring requirements to enhance the coherence of SAR image echo signals; The high-precision deformation monitoring results of the corresponding pixels of the corner reflector are combined with the optical features to further improve the early identification capability of potential hazards.