A geological disaster automatic identification system and method based on multi-source remote sensing data

Abstract

The present invention discloses a geological disaster automatic identification system and method based on multi-source remote sensing data, which relates to the field of geological disaster identification. The present invention first collects optical remote sensing images, SAR radar images and LIDAR point cloud data of the target area before and after the occurrence of the geological disaster, and obtains standard images, standard SAR and standard point clouds through preprocessing; compares and analyzes the standard images before and after the disaster to determine the disaster area, and analyzes the standard images, standard SAR and standard point clouds of the disaster area respectively to obtain a vegetation comprehensive index, a SAR index and a point cloud index; uses historical disaster data to train a convolutional neural network model to obtain a disaster identification model; finally, inputs the target area related index into the disaster identification model to determine the disaster type. The present invention overcomes the defects of a single data source by fusing multi-source remote sensing data, can accurately identify the type of geological disaster, and provides a key prerequisite for governance work.

Description

Technical Field

[0001] The present invention relates to the field of geological disaster identification, and in particular to a geological disaster automatic identification system and method based on multi-source remote sensing data. Background Art

[0002] Traditional geological hazard identification mainly relies on manual field surveys and topographic mapping, which not only consumes a lot of manpower, material resources and time, but is also restricted by natural conditions such as terrain and climate. In complex environments such as high mountains and canyons and remote areas, data collection efficiency is low and coverage is limited, making it difficult to achieve real-time dynamic monitoring. Single remote sensing data has certain defects. For example, although optical remote sensing can obtain surface information over a large area, it is easily affected by weather and lighting conditions, and the imaging quality is poor in conditions such as clouds and fog, and at night. Although synthetic aperture radar (SAR) remote sensing has the ability to work around the clock and in all weather conditions, it lacks the ability to extract features of subtle terrain changes, making it difficult to meet the needs of high-precision identification.

[0003] How to integrate multi-source remote sensing data to accurately identify disaster types is a problem we need to solve. To this end, we now provide a geological disaster automatic identification system and method based on multi-source remote sensing data. Summary of the Invention

[0004] The purpose of the present invention is to provide a system and method for automatically identifying geological disasters based on multi-source remote sensing data.

[0005] The purpose of the present invention can be achieved by the following technical solution: a method for automatically identifying geological hazards based on multi-source remote sensing data, comprising:

[0006] Collect remote sensing data of the target area before and after the geological disaster occurs. The remote sensing data includes: optical remote sensing images, SAR radar images and LIDAR point cloud data;

[0007] Pre-process the remote sensing data to obtain standard remote sensing data of the target area before and after the geological disaster; standard remote sensing data includes: standard image, standard SAR and standard point cloud;

[0008] Compare and analyze the standard images of the target area before and after the disaster to obtain the disaster area of ​​the target area. Analyze the standard images, standard SAR and standard point clouds corresponding to the disaster area to obtain the vegetation comprehensive index, SAR index and point cloud index.

[0009] Obtain the disaster type of each historical disaster and its corresponding vegetation comprehensive index, SAR index, and point cloud index. Use the vegetation comprehensive index, SAR index, and point cloud index to train the convolutional neural network model to obtain a disaster recognition model.

[0010] The disaster type of the target area is obtained through the vegetation comprehensive index, SAR index, point cloud index and disaster identification model of the target area.

[0011] Preferably, the process of preprocessing the remote sensing data to obtain standard remote sensing data of the target area before and after the geological disaster occurs is:

[0012] The optical remote sensing image and SAR radar image are subjected to radiometric calibration, geometric correction, noise removal and image enhancement respectively, thus obtaining standard image and standard SAR;

[0013] The LIDAR point cloud data is format converted, coordinate converted, noise removed and resampled to obtain a standard point cloud.

[0014] Preferably, the process of analyzing the standard image of the disaster area to obtain the vegetation comprehensive index is:

[0015] Divide the disaster area into several sub-areas, number each sub-area, and then divide each sub-area into two levels to obtain each grand-area, and obtain the vegetation index of each grand-area;

[0016] Preset the vegetation index threshold, and screen the sub-regions based on the comparison results of the vegetation index threshold and the vegetation index of each sub-region to obtain the high-value region;

[0017] The high-value frequency value of each sub-region is obtained by the number of high-value regions in each sub-region. The high-value frequency value of each sub-region is processed by the spatial entropy formula to obtain the chaos degree of the target region.

[0018] Obtain the minimum bounding rectangle of the disaster area, obtain the length and width of the rectangle, obtain the aspect ratio, and obtain the vegetation comprehensive index of the target area through the aspect ratio and chaos degree.

[0019] Preferably, the process of analyzing the standard SAR of the disaster area to obtain the SAR index is:

[0020] Obtain the gray level co-occurrence matrix of the standard SAR at different angles, and obtain the correlation of the gray level co-occurrence matrix at different angles;

[0021] The standard deviation and mean of the gray-level co-occurrence matrix correlation at each angle are taken as the angular wave value and angular mean respectively;

[0022] The angular wave value and the angular mean value are fused to obtain the SAR index.

[0023] Preferably, the process of analyzing the standard point cloud of the disaster area to obtain the point cloud index is as follows:

[0024] By using the standard point clouds before and after the disaster, digital elevation models before and after the disaster are constructed. The digital elevation model before the disaster is recorded as the pre-disaster model, and the digital elevation model after the disaster is recorded as the post-disaster model.

[0025] Randomly select two points in the pre-disaster model and randomly generate multiple paths between the two points. At the same time, select the same two points in the post-disaster model and generate the same paths.

[0026] Take points at equal distances on each path as marking points, obtain the elevation difference between each marking point on each path, and obtain the elevation set corresponding to each path;

[0027] Based on the number of positive and negative values ​​in the elevation set corresponding to each path, the elevation ratio is obtained, and the average of the elevation ratios of all paths is taken as the elevation ratio indicator value;

[0028] Select several points on the pre-disaster model as analysis points and obtain the elevation difference of each analysis point;

[0029] Obtain the spatial distance between the two analysis points, obtain a spatial distance set, preset a spatial distance threshold, and obtain a spatial weight matrix through the spatial distance set and the spatial distance threshold;

[0030] The elevation difference of each analysis point is merged through the spatial weight matrix to obtain the elevation difference indicator value;

[0031] The elevation ratio indication value and the elevation difference indication value are fused to obtain the point cloud index.

[0032] Preferably, the process of training the convolutional neural network model using the vegetation comprehensive index, SAR index, and point cloud index to obtain the disaster recognition model is as follows:

[0033] The vegetation comprehensive index, SAR index and point cloud index corresponding to each historical disaster are combined to obtain the identification vector corresponding to each disaster;

[0034] Based on the disaster type corresponding to each historical disaster, the identification vector is labeled to obtain a label identification vector;

[0035] The label recognition vector is input into the convolutional neural network model for training, and the weights and biases of each neuron are continuously adjusted based on the preset loss function to obtain a disaster recognition model.

[0036] Preferably, the process of obtaining the disaster type of the target area through the vegetation comprehensive index, SAR index, point cloud index and disaster identification model of the target area is:

[0037] Combine the vegetation comprehensive index, SAR index, and point cloud index of the target area in sequence to obtain a target recognition vector;

[0038] Input the target identification vector into the disaster identification model to obtain the disaster discrimination vector;

[0039] The disaster type corresponding to the largest element in the disaster discrimination vector is taken as the disaster type of the target area.

[0040] Preferably, a system for automatically identifying geological hazards based on multi-source remote sensing data includes:

[0041] Data acquisition module: collects remote sensing data of the target area, remote sensing data of various historical disasters, and the disaster types of various historical disasters;

[0042] Preprocessing module: preprocess the remote sensing data of the target area and the remote sensing data corresponding to each historical disaster to obtain the standard remote sensing data corresponding to the target area and each historical disaster;

[0043] Data analysis module: Analyze standard remote sensing data to obtain vegetation comprehensive index, SAR index and point cloud index;

[0044] Model training module: Use the vegetation comprehensive index, SAR index, and point cloud index corresponding to each historical disaster to train the convolutional neural network model to obtain a disaster recognition model;

[0045] Output module: The vegetation comprehensive index, SAR index and point cloud index of the target area are input into the disaster identification model to obtain the disaster type of the target area.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] The present invention collects optical remote sensing images, SAR radar images and LIDAR point cloud data of the target area and analyzes them separately to obtain a vegetation comprehensive index, SAR index and point cloud index; the vegetation comprehensive index, SAR index and point cloud index corresponding to historical disasters are combined to obtain an identification vector, and the identification vector is labeled based on the disaster type of each historical disaster to obtain a label identification vector, and a convolutional neural network model is trained by the label identification vector to obtain a disaster identification model; finally, the vegetation comprehensive index, SAR index and point cloud index of the target area are input into the disaster identification model to obtain the disaster type of the target area, thereby realizing the fusion of multi-source remote sensing data, improving the accuracy of disaster identification, and providing a key prerequisite for governance work. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments described in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0049] Figure 1 This is a schematic diagram of the present invention. DETAILED DESCRIPTION

[0050] like Figure 1 As shown, a method for automatic identification of geological hazards based on multi-source remote sensing data includes:

[0051] Collect remote sensing data of the target area before and after the geological disaster occurs. The remote sensing data includes: optical remote sensing images, SAR radar images and LIDAR point cloud data;

[0052] Pre-process the remote sensing data to obtain standard remote sensing data of the target area before and after the geological disaster; standard remote sensing data includes: standard image, standard SAR and standard point cloud;

[0053] The optical remote sensing image and SAR radar image are subjected to radiometric calibration, geometric correction, noise removal and image enhancement respectively, thus obtaining standard image and standard SAR;

[0054] Perform format conversion, coordinate conversion, noise removal and resampling on LIDAR point cloud data to obtain standard point cloud;

[0055] Compare and analyze the standard images of the target area before and after the disaster to obtain the disaster area of ​​the target area. Analyze the standard images, standard SAR and standard point clouds corresponding to the disaster area to obtain the vegetation comprehensive index, SAR index and point cloud index.

[0056] The process of analyzing the standard image of the disaster area and obtaining the comprehensive vegetation index is as follows:

[0057] The disaster area is divided into several sub-areas and each sub-area is numbered. The number is represented by i, i=1,2...n, and n represents the total number of sub-areas. Each sub-area is then divided into two levels to obtain each grand-area, and the vegetation index of each grand-area is obtained.

[0058] In detail, vegetation index The calculation formula is:

[0059] ;

[0060] in, represents the reflectivity of the near-infrared band in the Sun region, represents the reflectivity of the red light band in the Sun region;

[0061] Specifically, the leaf cell structure of plants has high reflectivity to near-infrared light and strong absorption to red light, while non-vegetation objects have low reflectivity to near-infrared light. When there are more plants in the sun area, the vegetation index is larger, and vice versa.

[0062] Preset the vegetation index threshold, and screen the sub-regions based on the comparison results of the vegetation index threshold and the vegetation index of each sub-region to obtain the high-value region;

[0063] In detail, if the vegetation index of a certain grandchild region is higher than the vegetation index threshold, the grandchild region is marked as a high-value region;

[0064] The number of high-value areas in each sub-area is used to obtain the high-value frequency value of each sub-area. The high-value frequency value of each sub-area is processed by the spatial entropy formula to obtain the chaos degree of the target area.

[0065] Specifically, the number of high-value areas corresponding to the sub-region is divided by the number of high-value areas corresponding to the disaster region to obtain the high-value frequency value corresponding to the sub-region;

[0066] In detail, the spatial entropy formula is:

[0067] ;

[0068] in, Represents the degree of chaos, Represents the high frequency value of the sub-region numbered i;

[0069] Get the minimum bounding rectangle of the disaster area, get the length and width of the rectangle, and get the aspect ratio;

[0070] Specifically, the length of the rectangle is divided by the width of the rectangle to obtain the aspect ratio;

[0071] Specifically, landslides are usually tongue-shaped, fan-shaped, or semicircular, extending from the landslide wall to the slope foot, but the difference between the length and width of their minimum circumscribed rectangle is small; while debris flows have obvious gully dependence, and their flow and accumulation areas extend along the gully, presenting a long strip or ribbon shape. The length and width of the minimum circumscribed rectangle differ greatly, so compared with landslides, the length-to-width ratio is larger;

[0072] The comprehensive vegetation index of the target area is obtained by the aspect ratio and disorder;

[0073] Specifically, the chaos degree HLD and the aspect ratio Normalized descendant input formula:

[0074] ;

[0075] Get the comprehensive vegetation index of the target area ,in and They are the chaos degree HLD and aspect ratio The corresponding weight impact factor;

[0076] Specifically, the geological disasters in this embodiment include debris flow and landslide. The distribution of destroyed plants in the disaster area is different for debris flow and landslide. In the disaster area caused by debris flow, the destroyed plants are distributed more evenly, and the corresponding chaos is also greater. In the disaster area caused by landslide, the destroyed plants are distributed more concentratedly, so the corresponding chaos is less than that of debris flow.

[0077] By fusing the aspect ratio and the degree of disorder, the vegetation comprehensive index is obtained, so that the vegetation coefficient can more accurately reflect the type of geological disasters. The smaller the vegetation comprehensive index, the higher the probability that the disaster area is a landslide, and the larger the vegetation comprehensive index, the higher the probability that the disaster area is a debris flow.

[0078] The process of analyzing the standard SAR of the disaster area and obtaining the SAR index is as follows:

[0079] Obtain the gray level co-occurrence matrix of the standard SAR at different angles, and obtain the correlation of the gray level co-occurrence matrix at different angles;

[0080] The standard deviation and mean of the gray-level co-occurrence matrix correlation at each angle are taken as the angular wave value and angular mean respectively;

[0081] Specifically, because SAR images of landslides have strong texture, the correlation of the gray-level co-occurrence matrix in a specific direction is significantly higher, while the correlation of the gray-level co-occurrence matrix in other angles is lower. However, the texture of debris flows shows mixed directionality, and the correlation of the gray-level co-occurrence matrix in each direction is higher.

[0082] Therefore, the SAR image produced by debris flow has a smaller angular wave value and a larger angular mean value; the SAR image produced by landslide has a larger angular wave value and a smaller angular mean value.

[0083] The angular wave value and the angular mean value are fused to obtain the SAR index;

[0084] In detail, the angular wave value and angular mean After normalization, substitute into the formula:

[0085] ;

[0086] Get SAR index ,in and The angular wave values ​​are and angular mean The corresponding weight impact factor;

[0087] Specifically, the larger the SAR index, the higher the probability that the SAR image corresponds to a debris flow, and the smaller the SAR index, the higher the probability that the SAR image corresponds to a landslide;

[0088] The process of analyzing the standard point cloud of the disaster area and obtaining the point cloud index is as follows:

[0089] Using the standard point clouds before and after the disaster, digital elevation models before and after the disaster are constructed. The digital elevation model before the disaster is recorded as the pre-disaster model, and the digital elevation model after the disaster is recorded as the post-disaster model.

[0090] Randomly select two points in the pre-disaster model and randomly generate multiple paths between the two points. At the same time, select the same two points in the post-disaster model and generate the same paths.

[0091] Take points at equal distances on each path as marking points, obtain the elevation difference between each marking point on each path, and obtain the elevation set corresponding to each path;

[0092] Specifically, for a certain marked point on a certain path, the elevation value corresponding to the post-disaster model is subtracted from the elevation value corresponding to the pre-disaster model to obtain the elevation difference of the marked point. The elevation values ​​of all marked points on the path are obtained to obtain the elevation set corresponding to the path.

[0093] Based on the number of positive and negative values ​​in the elevation set corresponding to each path, the elevation ratio is obtained, and the average of the elevation ratios of all paths is taken as the elevation ratio indicator value;

[0094] Specifically, the number of positive and negative values ​​in the elevation set corresponding to a path is obtained, and the number of positive values ​​is divided by the number of negative values ​​to obtain the elevation ratio corresponding to the path;

[0095] Specifically, different elevation ratios can reflect different disaster types. Landslides exhibit a bimodal distribution, with the sliding zone and the accumulation zone having similar areas. Therefore, the number of positive and negative values ​​in the elevation set corresponding to landslides is similar, and the elevation ratio is close to 1. Debris flows also exhibit a bimodal distribution, but the area of ​​the upstream erosion zone is larger than that of the downstream accumulation zone. Therefore, the number of negative values ​​in the elevation set corresponding to debris flows is relatively large, and the elevation ratio is less than 1.

[0096] Therefore, the smaller the elevation ratio indicator value is, the higher the probability that the geological disaster is a debris flow;

[0097] Select several points on the pre-disaster model as analysis points and obtain the elevation difference of each analysis point;

[0098] Obtain the spatial distance between the two analysis points, obtain a spatial distance set, preset a spatial distance threshold, and obtain a spatial weight matrix through the spatial distance set and the spatial distance threshold;

[0099] Specifically, the analysis points are numbered, and the number is represented by v, v = 1, 2...g, g represents the total number of analysis points, and the value in the spatial weight matrix Represents the weight value between the analysis point numbered e and the analysis point numbered r;

[0100] Specifically, if the spatial distance between the analysis point numbered e and the analysis point numbered r is less than the spatial distance threshold, then ,on the contrary, , and so on, thus obtaining the spatial weight matrix;

[0101] The elevation difference of each analysis point is merged through the spatial weight matrix to obtain the elevation difference indicator value;

[0102] In detail, through the formula:

[0103] ;

[0104] Get the elevation difference indication value ,in represents the mean value of the elevation difference of all analysis points, Represents the elevation difference of the analysis point numbered e, Represents the elevation difference of the analysis point numbered r;

[0105] Specifically, due to the large impact range of debris flow, the distance between the scouring area and the accumulation area is far. When the two analysis points are close, that is, the weight value between the two analysis points is 1, the difference between the two points and the mean of the elevation difference of all analysis points is positive or negative, so the elevation difference indicator value is larger;

[0106] The landslide has a small impact range. When two analysis points are close to each other, the difference between the two points and the mean of the elevation difference of all analysis points is in a chaotic state. It may be both positive and negative, or one positive and the other negative. Therefore, the elevation difference indicator value smaller;

[0107] Therefore, the smaller the elevation difference indication value, the greater the probability that the geological disaster is a landslide;

[0108] The elevation ratio indicator value and the elevation difference indicator value are fused to obtain the point cloud index;

[0109] Specifically, a number of elevation difference indicator value intervals are preset, and different elevation difference indicator value intervals correspond to different adjustment factors. The smaller the elevation difference indicator value, the larger the corresponding adjustment factor. The elevation difference indicator value is matched with the elevation difference indicator value interval to obtain the corresponding adjustment factor. The adjustment factor is multiplied by the elevation ratio indicator value to obtain the point cloud index.

[0110] Specifically, the smaller the point cloud index is, the greater the probability of debris flow is;

[0111] Obtain the types of historical disasters and their corresponding vegetation comprehensive index, SAR index, and point cloud index. Use the vegetation comprehensive index, SAR index, and point cloud index to train the convolutional neural network model to obtain a disaster recognition model.

[0112] The vegetation comprehensive index, SAR index and point cloud index corresponding to each historical disaster are combined to obtain the identification vector corresponding to each disaster;

[0113] Based on the disaster type corresponding to each historical disaster, the identification vector is labeled to obtain a label identification vector;

[0114] The label recognition vector is input into the convolutional neural network model for training. The weights and biases of each neuron are continuously adjusted based on the preset loss function to obtain a disaster recognition model.

[0115] Specifically, each disaster type is numbered, and the number is represented by c, c = 1, 2...q, q represents the number of disaster types, and the preset loss function is:

[0116] ;

[0117] Where SSZ represents the loss value, Indicates whether the current sample belongs to the disaster type numbered c. If so, =1, otherwise, Represents the probability that the current sample predicted by the model belongs to the disaster type numbered c;

[0118] Specifically, the loss value of the sample is obtained based on the loss function. Based on the loss value of the sample, the weights and biases of each neuron are continuously adjusted using the back-propagation mechanism to continuously reduce the loss value. At the same time, the early stopping method is used to prevent the model from overfitting, thereby obtaining a disaster recognition model.

[0119] The disaster type of the target area is obtained through the vegetation comprehensive index, SAR index, point cloud index and disaster identification model of the target area;

[0120] Combine the vegetation comprehensive index, SAR index, and point cloud index of the target area in sequence to obtain a target recognition vector;

[0121] Input the target identification vector into the disaster identification model to obtain the disaster discrimination vector;

[0122] In detail, each value in the disaster discrimination vector represents the probability value of the disaster identification model determining that the target area belongs to each disaster type;

[0123] The disaster type corresponding to the largest element in the disaster discrimination vector is taken as the disaster type of the target area;

[0124] A system for automatic identification of geological hazards based on multi-source remote sensing data, comprising:

[0125] Data acquisition module: collects remote sensing data of the target area, remote sensing data of various historical disasters, and the disaster types of various historical disasters;

[0126] Preprocessing module: preprocess the remote sensing data of the target area and the remote sensing data corresponding to each historical disaster to obtain the standard remote sensing data corresponding to the target area and each historical disaster;

[0127] Data analysis module: Analyze standard remote sensing data to obtain vegetation comprehensive index, SAR index and point cloud index;

[0128] Model training module: Use the vegetation comprehensive index, SAR index, and point cloud index corresponding to each historical disaster to train the convolutional neural network model to obtain a disaster recognition model;

[0129] Output module: Input the vegetation comprehensive index, SAR index and point cloud index of the target area into the disaster identification model to obtain the disaster type of the target area;

[0130] The above description is merely a preferred embodiment of the present invention and does not constitute any form of limitation to the present invention. Although the present invention has been disclosed as a preferred embodiment as above, it is not intended to limit the present invention. Any person skilled in the art can make some changes or modifications to equivalent embodiments using the technical contents disclosed above without departing from the scope of the technical solution of the present invention. However, any modification or equivalent replacement of the above embodiments made according to the technical essence of the present invention without departing from the content of the technical solution of the present invention still falls within the scope of the technical solution of the present invention.

Claims

1. A method for automatic identification of geological hazards based on multi-source remote sensing data, characterized in that: include: Collect remote sensing data of the target area before and after the geological disaster occurs. The remote sensing data includes: optical remote sensing images, SAR radar images and LIDAR point cloud data; Pre-process the remote sensing data to obtain standard remote sensing data of the target area before and after the geological disaster; standard remote sensing data includes: standard image, standard SAR and standard point cloud; Compare and analyze the standard images of the target area before and after the disaster to obtain the disaster area of ​​the target area. Analyze the standard images, standard SAR and standard point clouds corresponding to the disaster area to obtain the vegetation comprehensive index, SAR index and point cloud index. Obtain the disaster type of each historical disaster and its corresponding vegetation comprehensive index, SAR index, and point cloud index. Use the vegetation comprehensive index, SAR index, and point cloud index to train the convolutional neural network model to obtain a disaster recognition model. The disaster type of the target area is obtained through the vegetation comprehensive index, SAR index, point cloud index and disaster identification model of the target area; The process of analyzing the standard SAR of the disaster area and obtaining the SAR index is as follows: Obtain the gray level co-occurrence matrix of the standard SAR at different angles, and obtain the correlation of the gray level co-occurrence matrix at different angles; The standard deviation and mean of the gray-level co-occurrence matrix correlation at each angle are taken as the angular wave value and angular mean respectively; The angular wave value and the angular mean value are fused to obtain the SAR index; The process of analyzing the standard point cloud of the disaster area and obtaining the point cloud index is as follows: By using the standard point clouds before and after the disaster, digital elevation models before and after the disaster are constructed. The digital elevation model before the disaster is recorded as the pre-disaster model, and the digital elevation model after the disaster is recorded as the post-disaster model. Randomly select two points in the pre-disaster model and randomly generate multiple paths between the two points. At the same time, select the same two points in the post-disaster model and generate the same paths. Take points at equal distances on each path as marking points, obtain the elevation difference between each marking point on each path, and obtain the elevation set corresponding to each path; Based on the number of positive and negative values ​​in the elevation set corresponding to each path, the elevation ratio is obtained, and the average of the elevation ratios of all paths is taken as the elevation ratio indicator value; Select several points on the pre-disaster model as analysis points and obtain the elevation difference of each analysis point; Obtain the spatial distance between the two analysis points, obtain a spatial distance set, preset a spatial distance threshold, and obtain a spatial weight matrix through the spatial distance set and the spatial distance threshold; The elevation difference of each analysis point is merged through the spatial weight matrix to obtain the elevation difference indicator value; The elevation ratio indication value and the elevation difference indication value are fused to obtain the point cloud index.

2. The method for automatic identification of geological hazards based on multi-source remote sensing data according to claim 1, characterized in that: The process of preprocessing remote sensing data to obtain standard remote sensing data of the target area before and after the geological disaster occurs is as follows: The optical remote sensing image and SAR radar image are subjected to radiometric calibration, geometric correction, noise removal and image enhancement respectively, thus obtaining standard image and standard SAR; The LIDAR point cloud data is format converted, coordinate converted, noise removed and resampled to obtain a standard point cloud.

3. The method for automatic identification of geological hazards based on multi-source remote sensing data according to claim 2, characterized in that: The process of analyzing the standard image of the disaster area and obtaining the comprehensive vegetation index is as follows: Divide the disaster area into several sub-areas, number each sub-area, and then divide each sub-area into two levels to obtain each grand-area, and obtain the vegetation index of each grand-area; Preset the vegetation index threshold, and screen the sub-regions based on the comparison results of the vegetation index threshold and the vegetation index of each sub-region to obtain the high-value region; The high-value frequency value of each sub-region is obtained by the number of high-value regions in each sub-region. The high-value frequency value of each sub-region is processed by the spatial entropy formula to obtain the chaos degree of the target region. Obtain the minimum bounding rectangle of the disaster area, obtain the length and width of the rectangle, obtain the aspect ratio, and obtain the vegetation comprehensive index of the target area through the aspect ratio and chaos degree.

4. The method for automatic identification of geological hazards based on multi-source remote sensing data according to claim 3, characterized in that: The process of training the convolutional neural network model using the vegetation comprehensive index, SAR index, and point cloud index to obtain the disaster recognition model is as follows: The vegetation comprehensive index, SAR index and point cloud index corresponding to each historical disaster are combined to obtain the identification vector corresponding to each disaster; Based on the disaster type corresponding to each historical disaster, the identification vector is labeled to obtain a label identification vector; The label recognition vector is input into the convolutional neural network model for training, and the weights and biases of each neuron are continuously adjusted based on the preset loss function to obtain a disaster recognition model.

5. The method for automatic identification of geological hazards based on multi-source remote sensing data according to claim 4, characterized in that: The process of obtaining the disaster type of the target area through the vegetation comprehensive index, SAR index, point cloud index and disaster identification model of the target area is as follows: Combine the vegetation comprehensive index, SAR index, and point cloud index of the target area in sequence to obtain a target recognition vector; Input the target identification vector into the disaster identification model to obtain the disaster discrimination vector; The disaster type corresponding to the largest element in the disaster discrimination vector is taken as the disaster type of the target area.

6. A system for automatically identifying geological hazards based on multi-source remote sensing data according to any one of claims 1 to 5, characterized in that: include: Data acquisition module: collects remote sensing data of the target area, remote sensing data of various historical disasters, and the disaster types of various historical disasters; Preprocessing module: preprocess the remote sensing data of the target area and the remote sensing data corresponding to each historical disaster to obtain the standard remote sensing data corresponding to the target area and each historical disaster; Data analysis module: Analyze standard remote sensing data to obtain vegetation comprehensive index, SAR index and point cloud index; Model training module: Use the vegetation comprehensive index, SAR index, and point cloud index corresponding to each historical disaster to train the convolutional neural network model to obtain a disaster recognition model; Output module: Input the vegetation comprehensive index, SAR index and point cloud index of the target area into the disaster model to obtain the disaster type of the target area.

Images