A landslide hazard identification method, system, medium and device

Abstract

This invention provides a method for identifying landslide hazards, relating to the field of landslide hazard identification technology. The method includes: acquiring multi-source image data of a target area, including optical remote sensing images, SAR images, DEM data, normalized difference vegetation index (DDE), and land cover type; performing channel overlay processing on the multi-source image data to obtain a multi-channel image; dividing the multi-channel image into multiple sub-images, and inputting each sub-image into a landslide hazard prediction model for landslide prediction, obtaining the probability of a landslide hazard existing in the sub-region corresponding to each sub-image; identifying sub-images with probabilities greater than a preset probability threshold, where the sub-region corresponding to these sub-images is a region in the target area with a landslide hazard; and identifying sub-images with probabilities greater than a preset probability threshold, where these sub-images are regions in the target area with a landslide hazard. This method solves the problem of low accuracy in landslide hazard identification results in existing technologies.

Description

Technical Field

[0001] This invention relates to the field of landslide hazard identification technology, and in particular to a landslide hazard identification method, system, medium and equipment. Background Technology

[0002] Landslides are geological hazards caused by the sliding of soil and rock masses down a slope under the influence of gravity, triggered by factors such as river erosion, groundwater activity, rainfall, earthquakes, and artificial slope cutting. Landslides account for over 60% of all geological hazards nationwide, and statistics show that nearly 80% of major geological disasters occur outside known hazard sites. Therefore, comprehensive, large-scale, and efficient landslide hazard identification is an urgent need and of significant practical importance for landslide disaster prevention and control.

[0003] Traditional manual on-site investigation of landslide hazards is inefficient, costly, and extremely resource-intensive. In recent years, with the development of computer technology, deep learning methods have received widespread attention in the field of landslide identification, effectively extracting deep-level features and improving the accuracy of landslide identification. Based on the spatial autocorrelation assumption, landslide occurrence is closely related to the surrounding environment. Traditional machine learning methods can only express linear relationships between data and struggle to consider the environmental information surrounding the landslide point. Methods based on landslide hazard prediction models can take into account the contextual information of the landslide, possess stronger feature extraction capabilities, and can express complex nonlinear relationships between landslide information. Currently, deep learning methods have some applications in landslide identification. Data-level landslide identification methods are mostly based on optical remote sensing imagery, integrating multimodal data such as Digital Elevation Model (DEM) data and Synthetic Aperture Radar (SAR) imagery. However, using only DEM or SAR imagery is insufficient to comprehensively depict landslide hazard information, focusing primarily on extracting areas where landslides have already occurred. Deep learning models based on Convolutional Neural Networks (CNNs), such as VGGNet, UNet, and ResNet, have achieved some success in landslide identification. However, these models have large network depths and a large number of parameters, making them prone to overfitting when identifying landslide hazards in areas with limited data, and they also require significant computational power.

[0004] Currently, the main data sources for landslide identification include SAR imagery, DEM (Digital Elevation Model), and optical remote sensing. SAR imagery is typically used in landslide identification by employing Interferometric Synthetic Aperture Radar (InSAR) technology to obtain surface deformation rate maps and then using thresholding methods to distinguish landslides. However, this approach is subjective and prone to missed or false positives. Because landslides have unique topographic geometric representations, and DEM data can fully reflect these geometric features, DEM data plays a crucial auxiliary role in landslide identification and is widely used in landslide identification research. However, acquiring high-resolution DEM data is costly. Optical remote sensing imagery reflects the color, shape, and texture of landslides, but it is limited to information on already occurred landslides and cannot identify potential landslides. However, relying solely on the above data is insufficient to comprehensively describe landslide hazards, resulting in low accuracy in landslide hazard identification results. Summary of the Invention

[0005] This invention provides a method, system, medium, and equipment for identifying landslide hazards, with the aim of solving the problem of low accuracy in landslide hazard identification results in existing technologies.

[0006] To achieve the above objectives, the present invention provides a method for identifying landslide hazards, comprising:

[0007] Step 1: Acquire multi-source image data of the target area. The multi-source image data includes optical remote sensing images, SAR images, DEM data, normalized vegetation index, and land cover type.

[0008] Step 2: Perform channel overlay processing on the multi-source image data to obtain a multi-channel image;

[0009] Step 3: Divide the multi-channel image into multiple sub-images, and input each sub-image into the landslide hazard prediction model to predict landslides, so as to obtain the probability that there is a landslide hazard in the sub-region corresponding to each sub-image; the sub-region is a part of the target area.

[0010] Step 4: Identify the sub-images whose probability is greater than a preset probability threshold. The sub-regions corresponding to these sub-images are the areas in the target region where there is a risk of landslide.

[0011] The landslide hazard prediction model includes: a first convolutional layer, a second convolutional layer, a third convolutional layer, a channel attention module, a first pooling layer, a second pooling layer, a Dropout layer, a fully connected layer, and a softmax layer;

[0012] The output of the first convolutional layer is connected to the input of the channel attention module. The output of the channel attention module is connected to the input of the first pooling layer. The output of the first pooling layer is connected to the input of the second convolutional layer. The output of the second convolutional layer is connected to the input of the second pooling layer. The output of the second pooling layer is connected to the input of the third convolutional layer. The output of the third convolutional layer is connected to the input of the Dropout layer. The output of the Dropout layer is connected to the input of the fully connected layer. The output of the fully connected layer is connected to the input of the softmax layer.

[0013] Furthermore, the sub-images are input one by one into the landslide hazard prediction model through a sliding window prediction method to predict landslides.

[0014] Furthermore, after each sub-image is input into the first convolutional layer for convolution processing, it passes through the channel attention module, the first pooling layer, the second convolutional module, the second pooling layer, the Dropout layer, the fully connected layer, and the softmax layer in sequence to perform landslide prediction, calculating the probability that the sub-region corresponding to each sub-image has a landslide risk.

[0015] Furthermore, the channel attention module includes:

[0016] The global average pooling layer, the fourth convolutional layer, and the fifth convolutional layer are connected in sequence;

[0017] For each of the multiple sub-images, the sub-image after passing through the first convolutional layer is input into the global average pooling layer, the fourth convolutional layer, and the fifth convolutional layer to obtain the number of channels. The sub-image after passing through the first convolutional layer is then multiplied with the number of channels by weights to assign weights to each channel in the sub-image.

[0018] Furthermore, after the first convolutional layer, there is a first batch of normalized layers. The input of the first batch of normalized layers is connected to the output of the first convolutional layer, and the output of the first batch of normalized layers is connected to the input of the channel attention module.

[0019] After the second convolutional layer, there is a second batch normalization layer. The input of the second batch normalization layer is connected to the output of the second convolutional layer, and the output of the second batch normalization layer is connected to the input of the second pooling layer.

[0020] After the third convolutional layer, there is a third batch normalization layer. The input of the third batch normalization layer is connected to the output of the third convolutional layer, and the output of the third batch normalization layer is connected to the input of the Dropout layer.

[0021] Furthermore, after step 4, the following is also included:

[0022] Within the target area, the landslide hazard zone is visualized.

[0023] The present invention also provides a landslide hazard identification device, comprising:

[0024] The acquisition module is used to acquire multi-source image data of the target area. The multi-source image data includes optical remote sensing images, SAR images, DEM data, normalized difference vegetation index (NDVI), and land cover type.

[0025] The processing module is used to perform channel overlay processing on multi-source image data to obtain multi-channel images;

[0026] The prediction module is used to segment multi-channel images into multiple sub-images and input each sub-image into the landslide hazard prediction model to predict landslides, thereby obtaining the probability that a sub-region corresponding to each sub-image has a landslide hazard; the sub-region is a part of the target region.

[0027] The identification module is used to identify sub-images with a probability greater than a preset probability threshold. The sub-region corresponding to the sub-image is the area in the target region where there is a risk of landslide.

[0028] The present invention also provides a computer-readable storage medium for storing a computer program, which, when executed, is used to implement the above-mentioned landslide hazard identification method.

[0029] The present invention also provides a landslide hazard identification device for implementing the above-mentioned landslide hazard identification method, comprising:

[0030] Memory and processor;

[0031] Memory is used to store computer programs;

[0032] A processor is used to execute computer programs stored in memory.

[0033] The above-described solution of the present invention has the following beneficial effects:

[0034] This invention acquires multi-source image data including optical remote sensing images, SAR images, DEM data, Normalized Difference Vegetation Index (NDVI), and land cover type. The multi-source image data is then overlaid to obtain multi-channel images. These multi-channel images are divided into multiple sub-images by rows and columns, and each sub-image is input into a landslide hazard prediction model for landslide prediction. The probability of a landslide hazard existing in the sub-region corresponding to each sub-image is obtained. Sub-images with probabilities greater than a preset probability threshold are identified, and the sub-regions corresponding to these sub-images are designated as target areas with landslide hazards. Furthermore, the landslide hazard prediction model used in this invention has fewer parameters, is more lightweight, has faster training and convergence speed, and is less dependent on computational power, resulting in higher efficiency. The data used is also easier to acquire, making it suitable for large-scale landslide hazard identification. This solves the problem of low accuracy in landslide hazard identification results caused by existing technologies not fully considering the multi-source data characteristics of landslide areas.

[0035] Other beneficial effects of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0036] Figure 1 This is a flowchart illustrating an embodiment of the present invention;

[0037] Figure 2 This is a structural diagram of the landslide hazard prediction model in an embodiment of the present invention;

[0038] Figure 3 This is a structural diagram of the channel attention module in an embodiment of the present invention;

[0039] Figure 4 The ROC curve obtained in an embodiment of the present invention;

[0040] Figure 5 The PR curve obtained in an embodiment of the present invention;

[0041] Figure 6 This is a visualization result of landslide hazard identification in an embodiment of the present invention. Detailed Implementation

[0042] To make the technical problems, solutions, and advantages of this invention clearer, a detailed description will be provided below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0043] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0044] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a locking connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0045] Furthermore, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0046] This invention addresses existing problems by providing a method, system, medium, and equipment for identifying landslide hazards.

[0047] like Figure 1 As shown, an embodiment of the present invention provides a method for identifying landslide hazards, including:

[0048] Step 1: Acquire multi-source image data of the target area. The multi-source image data includes optical remote sensing images, SAR images, DEM data, normalized vegetation index, and land cover type.

[0049] Step 2: Perform channel overlay processing on the multi-source image data to obtain a multi-channel image;

[0050] Step 3: Divide the multi-channel image into multiple sub-images, and input each sub-image into the landslide hazard prediction model for landslide prediction to obtain the probability that a landslide hazard exists in the sub-region corresponding to each sub-image; the sub-region is a part of the target area.

[0051] Step 4: Identify sub-images with a probability greater than a preset probability threshold. The sub-region corresponding to the sub-image is the area in the target region where there is a risk of landslide.

[0052] The landslide hazard identification method of the present invention will be specifically described below using a county in southwestern my country as an example.

[0053] Taking a county in southwestern my country as the study area, the mountainous area accounts for more than 90% of the total area of ​​the county and is a potential landslide disaster area. 231 points in the area were selected as samples of potential landslide points on gentle slopes.

[0054] This invention acquires optical remote sensing images (Landsat), Sentinel-1 SAR images, DEM, NDVI, and land cover type of the study area. All of these data are publicly available. In this embodiment, a uniform resolution of 30m is set. An InSAR deformation rate map is obtained using the SAR image, with the spatial resolution set to 30m and Nan values ​​removed. The optical remote sensing images (RGB bands), InSAR deformation rate map, DEM, NDVI, and land cover type are unified into a single geographic coordinate system, and a multi-channel image is generated using channel overlay. Channel overlay processing refers to stitching together the pixel matrices of the multi-source image data to obtain the multi-channel image.

[0055] In this embodiment of the invention, existing landslide hazard points in the study area are used as positive samples. To estimate the balance of the sample set, negative samples are randomly generated at a 1:1 ratio of positive to negative samples. In practical applications, landslides account for a very small proportion of the entire map, and the rest can be considered as non-landslide areas. Therefore, in the experiment, negative samples may be sampled in landslide hazard areas or areas where landslides have already occurred. This is a low-probability event, and a small number of noisy samples can improve the robustness of the model.

[0056] This invention employs a multi-scale strategy, using a patch-based sampling method, selecting 16... 16, 20 20, 24 24 Three scales, and resampled to a smaller scale 16 16. Based on the spatial autocorrelation assumption, the occurrence of landslides is closely related to the surrounding environment, and the multi-scale strategy can fully estimate the multi-scale contextual environmental information around the landslide.

[0057] Specifically, before inputting the landslide hazard prediction model, it is also necessary to clean and augment the image data to remove samples containing outliers (NaN).

[0058] The embodiments of the present invention use five methods—horizontal symmetrical flipping, vertical symmetrical flipping, rotation of 90 degrees, rotation of 180 degrees, and rotation of 270 degrees—to expand the sample size by 5 times.

[0059] Specifically, such as Figure 2 As shown, the landslide hazard prediction model constructed in this embodiment of the invention includes:

[0060] The first convolutional layer, the second convolutional layer, and the third convolutional layer used for convolution processing;

[0061] Channel attention module for channel weighting;

[0062] The first pooling layer and the second pooling layer are used for downsampling operations;

[0063] Dropout layer and fully connected layer used to prevent overfitting of landslide hazard prediction models;

[0064] Softmax layer used for classification;

[0065] The output of the first convolutional layer is connected to the input of the channel attention module. The output of the channel attention module is connected to the input of the first pooling layer. The output of the first pooling layer is connected to the input of the second convolutional layer. The output of the second convolutional layer is connected to the input of the second pooling layer. The output of the second pooling layer is connected to the input of the third convolutional layer. The output of the third convolutional layer is connected to the input of the Dropout layer. The output of the Dropout layer is connected to the input of the fully connected layer. The output of the fully connected layer is connected to the input of the softmax layer.

[0066] Specifically, each sub-image is input into the first convolutional layer for convolution processing, and then landslide prediction is performed through the channel attention module, the first pooling layer, the second convolutional module, the second pooling layer, the Dropout layer, the fully connected layer, and the softmax layer connected in sequence. The probability that there is a landslide hazard in the sub-region corresponding to each sub-image is calculated.

[0067] Specifically, in this embodiment of the invention, the first convolutional layer, the second convolutional layer, and the third convolutional layer all employ 3... A convolutional kernel of size 3 is used, with the stride and padding of the feature map both set to 1; the window size and stride of the first and second pooling layers are set to 2 for double downsampling; the dropout layer parameter is set to 0.5; and finally, a softmax layer is used for classification.

[0068] Specifically, such as Figure 3 As shown, the channel attention module includes:

[0069] The global average pooling layer, the fourth convolutional layer, and the fifth convolutional layer are connected in sequence;

[0070] For each of the multiple sub-images, the sub-image after passing through the first convolutional layer is input into the global average pooling layer, the fourth convolutional layer, and the fifth convolutional layer to obtain the number of channels. The sub-image after passing through the first convolutional layer is then multiplied with the number of channels by weights to assign weights to each channel in the sub-image, thereby establishing the correlation between multiple channels and strengthening important features.

[0071] Specifically, after the first convolutional layer, there is a first batch of normalized layers. The input of the first batch of normalized layers is connected to the output of the first convolutional layer, and the output of the first batch of normalized layers is connected to the input of the channel attention module.

[0072] After the second convolutional layer, there is a second batch normalization layer. The input of the second batch normalization layer is connected to the output of the second convolutional layer, and the output of the second batch normalization layer is connected to the input of the second pooling layer.

[0073] After the third convolutional layer, there is a third batch normalization layer. The input of the third batch normalization layer is connected to the output of the third convolutional layer, and the output of the third batch normalization layer is connected to the input of the Dropout layer.

[0074] Batch normalization is performed by using the first, second, and third normalization layers to accelerate model convergence.

[0075] The sample set is input into the landslide hazard prediction model for training; the input sample size is 16. 16 7 (16) (16 pixels, 7 channels), using binary cross-entropy as the loss function, the formula is as follows:

[0076]

[0077] in, The total number of samples, For the first The true label of a binary sample is 0 (non-slide) or 1 (slide). For the first The output of each binary sample is the probability that it belongs to the label y.

[0078] The trained landslide hazard prediction model was tested on a test set, and the test results are as follows: Figure 4 and Figure 5 As shown.

[0079] In this embodiment of the invention, multiple units of size 16 are obtained by dividing the data into rows and columns. The multi-channel image was cropped into 16 sub-images using a sliding window prediction method. Each 16-bit sub-image is input into the first convolutional layer for convolution processing. Then, it passes through a channel attention module, a first pooling layer, a second convolutional module, a second pooling layer, a third convolutional layer, a dropout layer, a fully connected layer, and a softmax layer in sequence to predict landslides. The probability of each sub-image having a landslide risk is calculated, and the images are stitched together to obtain the prediction result for the whole image.

[0080] The row and column segmentation method is as follows: Let the size of the sub-image be n. n (in this invention, it is 16) 16) The number of rows and columns of the multi-channel image, a b) Calculate how many sub-images can be divided into, padding with 0s where rows and columns are insufficient, and then traverse from row to column to divide the sub-images.

[0081] Specifically, the preset probability threshold is 0.8. If a sub-image is identified with a probability greater than 0.8, the sub-region corresponding to that sub-image is considered to be an area in the target area with a potential landslide hazard. If a sub-image has a probability less than 0.8, the sub-region corresponding to that sub-image is considered to be an area in the target area without a potential landslide hazard.

[0082] Specifically, after step 4, the following is also included:

[0083] Within the target area, the landslide hazard area is visualized to obtain a visualization result image of the landslide hazard identification, as shown below. Figure 6 As shown.

[0084] This invention employs a multi-scale strategy and multi-source data for ablation experiments. During the training of the landslide hazard prediction model, the number of positive and negative samples used is 213, the batch size is set to 16, and the training batch epoch is 40. The model is evaluated using four metrics: Accuracy, Recall, Precision, and F1-Score. The confusion matrix of the model's prediction results is calculated, and the aforementioned four metrics are then used to calculate the results.

[0085] In the confusion matrix, TP represents correctly predicted positive samples, FP represents incorrectly predicted positive samples, FN represents correctly predicted negative samples, and TN represents incorrectly predicted negative samples.

[0086] Accuracy measures the precision of classification and is the ratio of the number of correctly classified samples to the total number of samples. The formula is as follows:

[0087]

[0088] Recall is the probability of a correct prediction among positive samples. Its calculation formula is:

[0089]

[0090] Precision is the accuracy rate, representing the probability that a sample predicted as positive is correctly predicted. Its calculation formula is:

[0091]

[0092] The F1-Score is the harmonic mean of recall and precision, measuring the overall performance of a model. Its formula is:

[0093]

[0094] The results of the ablation experiment are shown in Table 1 below. Among them, multi-source means that in addition to basic optical remote sensing images, InSAR deformation rate, and DEM, NDVI and land cover type are added; multi-scale means that a multi-scale sampling strategy is used; and attention means that a channel attention module is added.

[0095]

[0096] Table 1

[0097] Experimental results show that the additional multi-source data and multi-scale strategy adopted in this embodiment of the invention can effectively improve the accuracy of landslide identification. Ablation experiments show that the multi-scale sampling strategy can expand the sample set to a certain extent and alleviate the problem of overfitting caused by the small number of landslide sample points.

[0098] The experimental results show that, compared with existing convolutional neural networks, the landslide hazard prediction model in this embodiment has a smaller parameter count of only 7.37M, making the network more lightweight, faster in training and convergence, less dependent on computing power, and more efficient. Furthermore, the data used is easier to acquire, making it suitable for large-scale landslide hazard identification. It achieves high identification accuracy with only about 200 positive sample points, thus solving the problem of low accuracy in existing landslide hazard identification technologies.

[0099] This invention also provides a landslide hazard identification device, comprising:

[0100] The acquisition module is used to acquire multi-source image data of the target area. The multi-source image data includes optical remote sensing images, SAR images, DEM data, normalized difference vegetation index (NDVI), and land cover type.

[0101] The processing module is used to perform channel overlay processing on multi-source image data to obtain multi-channel images;

[0102] The prediction module is used to segment multi-channel images into multiple sub-images and input each sub-image into the landslide hazard prediction model to predict landslides and obtain the probability that there is a landslide hazard in the sub-region corresponding to each sub-image.

[0103] The identification module is used to identify sub-images with a probability greater than a preset probability threshold. The sub-region corresponding to the sub-image is the area in the target region where there is a risk of landslide.

[0104] This invention also provides a computer-readable storage medium for storing a computer program, which, when executed, is used to implement the aforementioned landslide hazard identification method.

[0105] This invention also provides a landslide hazard identification device for implementing the above-mentioned landslide hazard identification method, comprising:

[0106] Memory and processor;

[0107] Memory is used to store computer programs;

[0108] A processor is used to execute computer programs stored in memory.

[0109] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for identifying landslide hazards, characterized in that, include: Step 1: Acquire multi-source image data of the target area, including optical remote sensing images, SAR images, DEM data, normalized vegetation index, and land cover type. Step 2: Perform channel overlay processing on the multi-source image data to obtain a multi-channel image; Step 3: Divide the multi-channel image into multiple sub-images, and input each of the sub-images into the landslide hazard prediction model for landslide prediction to obtain the probability that a landslide hazard exists in the sub-region corresponding to each sub-image; the sub-region is a part of the target region; Step 4: Identify the sub-images whose probability is greater than a preset probability threshold. The sub-regions corresponding to these sub-images are the areas in the target region where there is a risk of landslide. The landslide hazard prediction model includes: a first convolutional layer, a second convolutional layer, a third convolutional layer, a channel attention module, a first pooling layer, a second pooling layer, a Dropout layer, a fully connected layer, and a softmax layer; The output of the first convolutional layer is connected to the input of the channel attention module. The output of the channel attention module is connected to the input of the first pooling layer. The output of the first pooling layer is connected to the input of the second convolutional layer. The output of the second convolutional layer is connected to the input of the second pooling layer. The output of the second pooling layer is connected to the input of the third convolutional layer. The output of the third convolutional layer is connected to the input of the Dropout layer. The output of the Dropout layer is connected to the input of the fully connected layer. The output of the fully connected layer is connected to the input of the softmax layer. The multiple sub-images are input one by one into the landslide hazard prediction model to predict landslides, and the probability that a sub-region corresponding to each sub-image has a landslide hazard is obtained, including: The multi-channel image was cropped to 16 using a sliding window prediction method. Each 16-bit sub-image is input into the first convolutional layer for convolution processing. Then, it passes through a channel attention module, a first pooling layer, a second convolutional module, a second pooling layer, a third convolutional layer, a dropout layer, a fully connected layer, and a softmax layer in sequence to predict landslides and calculate the probability that each sub-image has a potential landslide risk.

2. The landslide hazard identification method according to claim 1, characterized in that, The sub-images are input one by one into the landslide hazard prediction model to predict landslides using a sliding window prediction method.

3. The landslide hazard identification method according to claim 1, characterized in that, After each sub-image is input into the first convolutional layer for convolution processing, it passes through the channel attention module, the first pooling layer, the second convolutional layer, the third convolutional layer, the Dropout layer, the fully connected layer, and the softmax layer in sequence to perform landslide prediction, calculating the probability that the sub-region corresponding to each sub-image has a landslide hazard.

4. The landslide hazard identification method according to claim 3, characterized in that, The channel attention module includes: The global average pooling layer, the fourth convolutional layer, and the fifth convolutional layer are connected in sequence; For each of the multiple sub-images, the sub-image after passing through the first convolutional layer is input into the global average pooling layer, the fourth convolutional layer, and the fifth convolutional layer to obtain the number of channels. The sub-image after passing through the first convolutional layer is then multiplied with the number of channels by weights to assign weights to each channel in the sub-image.

5. The landslide hazard identification method according to claim 3, characterized in that, After the first convolutional layer, there is a first batch of normalized layers. The input of the first batch of normalized layers is connected to the output of the first convolutional layer, and the output of the first batch of normalized layers is connected to the input of the channel attention module. A second batch of normalized layers is provided after the second convolutional layer. The input of the second batch of normalized layers is connected to the output of the second convolutional layer, and the output of the second batch of normalized layers is connected to the input of the second pooling layer. A third batch normalization layer is provided after the third convolutional layer. The input of the third batch normalization layer is connected to the output of the third convolutional layer, and the output of the third batch normalization layer is connected to the input of the Dropout layer.

6. The landslide hazard identification method according to claim 3, characterized in that, Step 4 is followed by: In the target area, the landslide hazard area is visualized.

7. A landslide hazard identification device, characterized in that, include: The acquisition module is used to acquire multi-source image data of the target area, including optical remote sensing images, SAR images, DEM data, normalized difference vegetation index (NDVI), and land cover type. The processing module is used to perform channel overlay processing on the multi-source image data to obtain a multi-channel image; The prediction module is used to segment the multi-channel image into multiple sub-images, and input each of the multiple sub-images into the landslide hazard prediction model to predict landslides, thereby obtaining the probability that a sub-region corresponding to each sub-image has a landslide hazard; the sub-region is a part of the target region; The identification module is used to identify sub-images with a probability greater than a preset probability threshold, and the sub-region corresponding to the sub-image is the area in the target region where there is a landslide hazard. The landslide hazard prediction model includes: a first convolutional layer, a second convolutional layer, a third convolutional layer, a channel attention module, a first pooling layer, a second pooling layer, a Dropout layer, a fully connected layer, and a softmax layer; The output of the first convolutional layer is connected to the input of the channel attention module. The output of the channel attention module is connected to the input of the first pooling layer. The output of the first pooling layer is connected to the input of the second convolutional layer. The output of the second convolutional layer is connected to the input of the second pooling layer. The output of the second pooling layer is connected to the input of the third convolutional layer. The output of the third convolutional layer is connected to the input of the Dropout layer. The output of the Dropout layer is connected to the input of the fully connected layer. The output of the fully connected layer is connected to the input of the softmax layer. The multiple sub-images are input one by one into the landslide hazard prediction model to predict landslides, and the probability that a sub-region corresponding to each sub-image has a landslide hazard is obtained, including: The multi-channel image was cropped to 16 using a sliding window prediction method. Each 16-bit sub-image is input into the first convolutional layer for convolution processing. Then, it passes through a channel attention module, a first pooling layer, a second convolutional module, a second pooling layer, a third convolutional layer, a dropout layer, a fully connected layer, and a softmax layer in sequence to predict landslides and calculate the probability that each sub-image has a potential landslide risk.

8. A computer-readable storage medium for storing a computer program, characterized in that, The computer program is executed to implement the landslide hazard identification method according to any one of claims 1-6.

9. A landslide hazard identification device, used to implement the landslide hazard identification method according to any one of claims 1-6, characterized in that, include: Memory and processor; The memory is used to store computer programs; The processor is used to execute computer programs stored in the memory.

Images