Landslide detection method and device, computer equipment, readable storage medium and program product

Abstract

The application relates to a landslide detection method and device, computer equipment, a computer readable storage medium and a computer program product. The method comprises the following steps: acquiring historical landslide data of a target area, performing image correction and spatial registration processing to obtain processed historical landslide data, and constructing a landslide sample set; superimposing multi-source factor data and landslide label data in the channel dimension to obtain a multi-channel stereo image block, performing slice processing to obtain a training sample set and a test sample set; training a preset model based on the training sample set and the test sample set, obtaining a trained preset model under the condition that a preset training condition is met; calculating the performance index of the trained preset model, and acquiring real-time landslide data of the target area under the condition that the performance index meets a preset performance condition, inputting the real-time landslide data into the trained preset model, and obtaining a landslide detection result of the target area. The method can improve the accuracy of landslide detection.

Description

Technical Field

[0001] This application relates to the field of intelligent geological disaster detection technology, and in particular to a landslide detection method, device, computer equipment, computer-readable storage medium, and computer program product. Background Technology

[0002] Landslides are natural phenomena caused by the overall sliding of soil and rock along a shear plane on a slope, resulting from human activities and natural environmental factors. Landslides are among the most frequent, widespread, and destructive geological hazards in the world. Given the current global situation of frequent earthquakes, vegetation destruction, extreme weather events, and unprecedentedly intense human engineering activities, potential landslide bodies have acquired suitable conditions for gestation and development, leading to a continuous increase in the frequency and scale of landslides and posing a growing potential threat to human life and property.

[0003] Therefore, detecting and predicting the spatial location of landslide-prone areas is an important task and means to prevent and mitigate the potential hazards of landslide disasters.

[0004] In related technologies, with the rapid development of artificial intelligence and Earth observation technology, deep learning models have shown excellent performance in many tasks in remote sensing-related fields. However, the application of deep learning models in landslide detection tasks is still limited. From the perspective of landslide feature extraction, existing deep learning models have problems such as insufficient extraction of local and global features of landslides, resulting in inaccurate landslide susceptibility mapping results and ultimately inaccurate detection results. Summary of the Invention

[0005] Therefore, it is necessary to provide a landslide detection method, apparatus, computer equipment, computer-readable storage medium, and computer program product that can improve the accuracy of landslide detection in response to the above-mentioned technical problems.

[0006] Firstly, this application provides a landslide detection method, including:

[0007] Historical landslide data of the target area is acquired, and image correction and spatial registration are performed on the historical landslide data to obtain processed historical landslide data.

[0008] Based on the processed historical landslide data, a landslide sample set is constructed; the landslide sample set includes multi-source factor data and landslide label data.

[0009] The multi-source factor data and the landslide label data are superimposed in the channel dimension to obtain a multi-channel stereo image patch with uniform resolution. The multi-channel stereo image patch is then sliced ​​to obtain a training sample set and a test sample set.

[0010] A preset model is trained based on the training sample set and the test sample set, and the trained preset model is obtained under the condition that the preset training conditions are met.

[0011] Calculate the performance index of the pre-set model after training. If the performance index meets the pre-set performance conditions, obtain real-time landslide data of the target area and input the real-time landslide data into the pre-set model after training to obtain the landslide detection result of the target area.

[0012] In one embodiment, the step of performing image correction and spatial registration processing on the historical landslide data to obtain processed historical landslide data includes:

[0013] Radiometric calibration and atmospheric correction are performed on the image data in the historical landslide data to obtain corrected historical landslide data;

[0014] The corrected historical landslide data are unified into the same spatial reference system, and all unified historical landslide data are sampled to the same spatial resolution to obtain the processed historical landslide data.

[0015] In one embodiment, the step of slicing the multi-channel stereo image block to obtain a training sample set and a test sample set includes:

[0016] Landslide pixels are randomly selected as center points in multi-channel stereo image blocks. The blocks are then expanded in height, width, and channels around the center point to obtain landslide sample blocks.

[0017] Randomly select non-slope pixels within the target area as the center point, and expand the height, width, and channels around the center point to obtain the same number of non-slope sample blocks as the number of landslide sample blocks.

[0018] Based on the landslide sample block and the non-landslide sample block, a training sample set and a test sample set are obtained according to a preset ratio.

[0019] In one embodiment, the performance metrics include at least one of the following: overall accuracy, precision, recall, F1 score, Matthews correlation coefficient, consistency coefficient, or area under the curve of the trained preset model.

[0020] In one embodiment, the trained preset model includes a convolutional coding module, a Transformer coding module, a feature fusion module, and a decoding module; the step of inputting the real-time landslide data into the trained preset model to obtain the landslide detection result of the target area includes:

[0021] The real-time landslide data is input into a convolutional coding module to extract local feature vectors from the real-time landslide data.

[0022] The local feature vectors are encoded at location and then input into the Transformer encoding module to extract the global feature vectors of the real-time landslide data.

[0023] The global feature vector is input into the feature fusion module for feature fusion to obtain the fused features;

[0024] The fused features are input into the decoding module to obtain the landslide detection results for the target area.

[0025] In one embodiment, the historical landslide data includes optical image data, mathematical elevation model data, geological element data, and human engineering activity data.

[0026] Secondly, this application also provides a landslide detection device, comprising:

[0027] The processing module is used to acquire historical landslide data of the target area, perform image correction and spatial registration processing on the historical landslide data, and obtain processed historical landslide data.

[0028] A construction module is used to build a landslide sample set based on processed historical landslide data; the landslide sample set includes multi-source factor data and landslide label data;

[0029] The processing module is also used to overlay the multi-source factor data and the landslide label data in the channel dimension to obtain a multi-channel stereo image block with uniform resolution, and to slice the multi-channel stereo image block to obtain a training sample set and a test sample set.

[0030] The training module is used to train a preset model based on the training sample set and the test sample set, and to obtain the trained preset model under the condition of satisfying the preset training conditions.

[0031] The calculation module is used to calculate the performance index of the preset model after training. When the performance index meets the preset performance conditions, it acquires real-time landslide data of the target area and inputs the real-time landslide data into the preset model after training to obtain the landslide detection result of the target area.

[0032] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0033] Historical landslide data of the target area is acquired, and image correction and spatial registration are performed on the historical landslide data to obtain processed historical landslide data.

[0034] Based on the processed historical landslide data, a landslide sample set is constructed; the landslide sample set includes multi-source factor data and landslide label data.

[0035] The multi-source factor data and the landslide label data are superimposed in the channel dimension to obtain a multi-channel stereo image patch with uniform resolution. The multi-channel stereo image patch is then sliced ​​to obtain a training sample set and a test sample set.

[0036] A preset model is trained based on the training sample set and the test sample set, and the trained preset model is obtained under the condition that the preset training conditions are met.

[0037] Calculate the performance index of the pre-set model after training. If the performance index meets the pre-set performance conditions, obtain real-time landslide data of the target area and input the real-time landslide data into the pre-set model after training to obtain the landslide detection result of the target area.

[0038] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:

[0039] Historical landslide data of the target area is acquired, and image correction and spatial registration are performed on the historical landslide data to obtain processed historical landslide data.

[0040] Based on the processed historical landslide data, a landslide sample set is constructed; the landslide sample set includes multi-source factor data and landslide label data.

[0041] The multi-source factor data and the landslide label data are superimposed in the channel dimension to obtain a multi-channel stereo image patch with uniform resolution. The multi-channel stereo image patch is then sliced ​​to obtain a training sample set and a test sample set.

[0042] A preset model is trained based on the training sample set and the test sample set, and the trained preset model is obtained under the condition that the preset training conditions are met.

[0043] Calculate the performance index of the pre-set model after training. If the performance index meets the pre-set performance conditions, obtain real-time landslide data of the target area and input the real-time landslide data into the pre-set model after training to obtain the landslide detection result of the target area.

[0044] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps:

[0045] Historical landslide data of the target area is acquired, and image correction and spatial registration are performed on the historical landslide data to obtain processed historical landslide data.

[0046] Based on the processed historical landslide data, a landslide sample set is constructed; the landslide sample set includes multi-source factor data and landslide label data.

[0047] The multi-source factor data and the landslide label data are superimposed in the channel dimension to obtain a multi-channel stereo image patch with uniform resolution. The multi-channel stereo image patch is then sliced ​​to obtain a training sample set and a test sample set.

[0048] A preset model is trained based on the training sample set and the test sample set, and the trained preset model is obtained under the condition that the preset training conditions are met.

[0049] Calculate the performance index of the pre-set model after training. If the performance index meets the pre-set performance conditions, obtain real-time landslide data of the target area and input the real-time landslide data into the pre-set model after training to obtain the landslide detection result of the target area.

[0050] The aforementioned landslide detection method, apparatus, computer equipment, computer-readable storage medium, and computer program product first acquire historical landslide data of the target area, perform image correction and spatial registration processing on the historical landslide data to obtain processed historical landslide data; based on the processed historical landslide data, construct a landslide sample set; the landslide sample set includes multi-source factor data and landslide label data; superimpose the multi-source factor data and landslide label data in the channel dimension to obtain a multi-channel stereo image block with uniform resolution, and slice the multi-channel stereo image block to obtain a training sample set and a test sample set; train a preset model based on the training sample set and the test sample set, and obtain the trained preset model under preset training conditions; calculate the performance index of the trained preset model, and if the performance index meets preset performance conditions, acquire real-time landslide data of the target area, input the real-time landslide data into the trained preset model, and obtain the landslide detection result of the target area. Thus, through spatial registration and channel overlay of multi-source data, multi-dimensional disaster-causing factors were effectively integrated, enhancing the characterization ability of sample features. Standardized training and testing samples were constructed through slicing, ensuring the standardization and comparability of model training and evaluation. The model was only put into practical application after its performance indicators met the preset conditions, ensuring the reliability and stability of the detection results. Finally, through real-time data input and output, rapid and accurate detection of landslides in the target area was achieved, providing efficient technical support for early identification, dynamic monitoring, and early warning response of landslide disasters, and effectively reducing the false negative and false positive rates. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 This is a diagram illustrating the application environment of a landslide detection method in one embodiment;

[0053] Figure 2 This is a flowchart illustrating a landslide detection method in one embodiment;

[0054] Figure 3 This is a structural block diagram of a landslide detection device in one embodiment;

[0055] Figure 4 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0057] It should be noted that the terms "first," "second," etc., used in this application can be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "comprising" and "having," and any variations thereof, used in this application, are intended to cover non-exclusive inclusion. The term "multiple" used in this application refers to two or more. The term "and / or" used in this application refers to one of the embodiments, or any combination of multiple embodiments.

[0058] The landslide detection method provided in this application embodiment can be applied to, for example... Figure 1In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104 or located on the cloud or other network servers. Terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, drones, low-altitude aircraft, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, projection devices, etc. Portable wearable devices can include smartwatches, smart bracelets, head-mounted devices, etc. Head-mounted devices can be virtual reality (VR) devices, augmented reality (AR) devices, smart glasses, etc. Server 104 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0059] In one exemplary embodiment, such as Figure 2 As shown, a landslide detection method is provided, which can be applied to... Figure 1 Taking terminal 102 as an example, the explanation includes the following steps 202 to 210. Wherein:

[0060] Step 202: Obtain historical landslide data for the target area, perform image correction and spatial registration on the historical landslide data, and obtain processed historical landslide data.

[0061] For example, historical landslide data that can characterize the spatial heterogeneity and multidimensional features of the target area are acquired, and image correction and spatial registration are performed on the historical landslide data to obtain processed historical landslide data.

[0062] Step 204: Based on the processed historical landslide data, construct a landslide sample set; the landslide sample set includes multi-source factor data and landslide label data.

[0063] Optionally, for the processed historical landslide data, landslide label data is determined, key influencing factors are screened using feature importance analysis, multi-source factor data is determined, and a landslide sample set is constructed based on the multi-source factor data and landslide label data.

[0064] Key influencing factors include elevation, slope, aspect, lithology, distance from faults, distance from rivers, rainfall, distance from roads, and NDVI (Normalized Difference Vegetation Index).

[0065] In some embodiments, key influencing factors are extracted using spatial analysis tools and normalized to ensure data range consistency.

[0066] In some embodiments, landslide label data may be obtained and produced by searching existing data and combining it with manual determination of the target area, or it may be determined by other means. This application embodiment does not limit this.

[0067] Step 206: The multi-source factor data and landslide label data are superimposed in the channel dimension to obtain a multi-channel stereo image block with uniform resolution. The multi-channel stereo image block is then sliced ​​to obtain the training sample set and the test sample set.

[0068] For example, multi-source factor data and landslide label data are superimposed in the channel dimension to obtain a multi-channel stereo image patch with uniform resolution, and the multi-channel stereo image patch is sliced ​​to obtain a training sample set and a test sample set.

[0069] Step 208: Train the preset model based on the training sample set and the test sample set, and obtain the trained preset model under the condition that the preset training conditions are met.

[0070] The preset model can be a CNN-Transformer fusion model (CTLGNet).

[0071] For example, a preset model is trained based on a training sample set and a test sample set, and the trained preset model is obtained under the condition that the preset training conditions are met.

[0072] The training method for the preset model can be a weighted sum of the cross-entropy loss function and the Dice loss function.

[0073] In some embodiments, the preset model is iteratively trained by weighted summation of cross-entropy loss function and Dice loss function, and the parameters in the preset model are updated so that the loss value continuously decreases until the loss value is at its minimum or the preset number of iterations is reached, thus obtaining the trained preset model.

[0074] Step 210: Calculate the performance index of the trained preset model. If the performance index meets the preset performance conditions, obtain the real-time landslide data of the target area and input the real-time landslide data into the trained preset model to obtain the landslide detection results of the target area.

[0075] Optionally, the performance index of the pre-trained model is calculated. If the performance index meets the pre-set performance conditions, real-time landslide data of the target area is obtained, and the real-time landslide data is input into the pre-trained model to obtain the landslide detection results of the target area.

[0076] Among them, the preset performance condition is that the performance index of the preset model after training is better than that of traditional machine learning models (such as random forest, support vector machine, etc.), and the real-time landslide data is data with the same structure as the historical landslide data but collected in different time periods.

[0077] In some embodiments, image correction and spatial registration are performed before real-time landslide data are input into a pre-trained model.

[0078] In some embodiments, the landslide detection results are vectorized and spatially aggregated to remove false detection patches that are too small. The trained preset model is then screened and corrected by combining elevation, slope and geological background information. A landslide detection result distribution map is generated through raster reclassification and regional statistical analysis.

[0079] In the aforementioned landslide detection method, historical landslide data of the target area is acquired, and image correction and spatial registration are performed on the historical landslide data to obtain processed historical landslide data. Based on the processed historical landslide data, a landslide sample set is constructed, which includes multi-source factor data and landslide label data. The multi-source factor data and landslide label data are superimposed along the channel dimension to obtain a multi-channel stereo image patch with uniform resolution. The multi-channel stereo image patch is then sliced ​​to obtain a training sample set and a test sample set. A preset model is trained based on the training sample set and the test sample set. Under preset training conditions, the trained preset model is obtained. The performance index of the trained preset model is calculated. Under preset performance conditions, real-time landslide data of the target area is acquired, and the real-time landslide data is input into the trained preset model to obtain the landslide detection result of the target area. Thus, through spatial registration and channel overlay of multi-source data, multi-dimensional disaster-causing factors were effectively integrated, enhancing the characterization ability of sample features. Standardized training and testing samples were constructed through slicing, ensuring the standardization and comparability of model training and evaluation. The model was only put into practical application after its performance indicators met the preset conditions, ensuring the reliability and stability of the detection results. Finally, through real-time data input and output, rapid and accurate detection of landslides in the target area was achieved, providing efficient technical support for early identification, dynamic monitoring, and early warning response of landslide disasters, and effectively reducing the false negative and false positive rates.

[0080] In an exemplary embodiment, image correction and spatial registration processing are performed on historical landslide data to obtain processed historical landslide data. This includes: performing radiometric calibration and atmospheric correction processing on the image data in the historical landslide data to obtain corrected historical landslide data; unifying the corrected historical landslide data to the same spatial reference system; and sampling all unified historical landslide data to the same spatial resolution to obtain processed historical landslide data.

[0081] In practice, radiometric calibration and atmospheric correction are performed on the image data in the historical landslide data to obtain the corrected historical landslide data. The corrected historical landslide data are then unified into the same spatial reference system, and all unified historical landslide data are sampled to the same spatial resolution to obtain the processed historical landslide data.

[0082] In some embodiments, radiometric calibration and atmospheric correction are performed using ENVI software, and multispectral and panchromatic band fusion is achieved through principal component transformation (PCA) and Gram-Schmidt fusion method.

[0083] In other embodiments, the processed historical landslide data is subjected to feature normalization.

[0084] In the above embodiments, radiometric calibration and atmospheric correction effectively eliminate the interference of sensor noise, atmospheric scattering and absorption on the radiometric information of the image, restore the true reflectivity of the ground objects, and significantly improve the data quality and spectral fidelity of optical images. By unifying multi-source data to the same spatial reference system and resampling to a consistent spatial resolution, the heterogeneity problem of inconsistent coordinate systems and mismatched resolutions among data from different sources (such as optical images, digital elevation models, and geological vector maps) is solved.

[0085] In an exemplary embodiment, slicing a multi-channel stereo image block to obtain a training sample set and a test sample set includes: randomly selecting landslide pixels as center points in the multi-channel stereo image block, and expanding the block in height, width, and channels around the center point to obtain landslide sample blocks; randomly selecting non-landslide pixels as center points within a target area, and expanding the block in height, width, and channels around the center point to obtain the same number of non-landslide sample blocks; and dividing the block into a training sample set and a test sample set according to a preset ratio based on the landslide sample blocks and non-landslide sample blocks.

[0086] In practice, landslide pixels are randomly selected from the multi-channel stereo image block as center points and labels, and expanded in height, width, and channels to obtain landslide sample blocks; non-landslide pixels are randomly selected from the target area as center points, and expanded in height, width, and channels with the center points as the center to obtain the same number of non-landslide sample blocks as the landslide sample blocks; based on the landslide sample blocks and non-landslide sample blocks, the training sample set and the test sample set are obtained according to a preset ratio (e.g., 7:3).

[0087] In some embodiments, the training and testing sample sets are expanded using data augmentation strategies, such as random rotation, translation, or mirroring of the images.

[0088] In the above embodiments, expansion centered on landslide pixels ensures that the core area of ​​each positive sample block contains effective landslide information, avoiding the problem of the model failing to learn meaningful features due to the low proportion of landslide pixels in the samples. At the same time, an equal number of non-landslide sample blocks are generated, effectively solving the inherent problem of severe imbalance between positive and negative samples in landslide detection and preventing the model from favoring the majority class and ignoring the minority class of landslides. The random selection strategy enhances the spatial representativeness and diversity of the samples, avoiding oversampling bias in specific areas or types of landslides. The training set and test set are divided according to a preset ratio, ensuring that the data used for model training and performance evaluation are independent of each other.

[0089] In one exemplary embodiment, the performance metrics include at least one of the following: overall accuracy, precision, recall, F1 score, Matthews correlation coefficient, consistency coefficient, or area under the curve of the trained preset model.

[0090] In practice, performance metrics include at least one of the following: overall accuracy (OA), precision, recall, F1 score, Matthews correlation coefficient (MCC), Kappa coefficient, or area under curve (AUC) of the pre-trained model.

[0091] In some embodiments, performance metrics are calculated by constructing a two-dimensional confusion matrix.

[0092] For example, as shown in Table 1, in the landslide detection confusion matrix, the rows of the matrix represent the true values, the columns of the matrix represent the predicted values, TP (True Positive) represents the number of pixels correctly predicted as landslides by the model, TN (True Negative) represents the number of pixels correctly predicted as non-landslides by the model, FP (False Positive) represents the number of pixels incorrectly predicted as landslides by the model, and FN (False Negative) represents the number of pixels incorrectly predicted as non-landslides by the model.

[0093] Table 1. Landslide Detection Confusion Matrix

[0094]

[0095] Overall accuracy (OA) represents the model's classification accuracy for all samples, which is the ratio of correctly predicted pixels to the total number of pixels. The calculation formula is shown below:

[0096]

[0097] Precision and recall are two commonly used metrics for evaluating classification results. Precision refers to the proportion of correctly predicted landslide pixels among all pixels predicted as landslides. It measures the model's accuracy in predicting landslides and is crucial for reducing false positives. Recall, on the other hand, refers to the proportion of correctly predicted landslide pixels among all true landslide pixels. It reflects the model's ability to capture true landslide pixels and is essential for reducing false negatives. The formulas for calculating precision and recall are shown below:

[0098]

[0099]

[0100] The F1 score is used to comprehensively evaluate the model's classification results. The F1 score is the harmonic mean of precision and recall. A higher F1 score indicates that the model has achieved a good balance between precision and recall, and the overall classification results of the model are better. The specific calculation formula is shown below:

[0101]

[0102] The Matthews correlation coefficient (MCC) effectively measures the correlation between actual and predicted values. It considers all elements in the confusion matrix and provides relatively stable performance even with imbalanced sample distributions. A value of 1 indicates the best predictive performance, 0 indicates random prediction, and -1 represents a completely inaccurate prediction. The specific calculation formula is shown below:

[0103]

[0104] The Kappa coefficient provides a method for evaluating model performance while considering random factors, and it is a widely used metric for evaluating classification models. The Kappa coefficient ranges from -1 to 1, where 1 represents perfect consistency, 0 represents random classification, and -1 represents complete disagreement. The specific calculation formula is shown below:

[0105]

[0106]

[0107] Where n is the total number of pixels and OA is the overall precision.

[0108] The ROC curve is a graphical tool that shows the relationship between the True Positive Rate and the False Positive Rate at different classification thresholds. The closer the ROC curve is to the top left corner, the more reliable the landslide susceptibility results. The AUC value is the area under the ROC curve, ranging from 0 to 1. The higher the AUC value, the better the model performance.

[0109] In the above embodiments, by using at least one of the following as model performance evaluation indicators—overall accuracy, precision, recall, F1 score, Matthews correlation coefficient, consistency coefficient, and area under the curve—a multi-dimensional and comprehensive quantitative evaluation of the landslide detection model is achieved.

[0110] In an exemplary embodiment, the trained preset model includes a convolutional coding module, a Transformer coding module, a feature fusion module, and a decoding module. Inputting real-time landslide data into the trained preset model to obtain landslide detection results for the target area includes: inputting the real-time landslide data into the convolutional coding module to extract local feature vectors from the real-time landslide data; encoding the local feature vectors at their positions and then inputting them into the Transformer coding module to extract global feature vectors from the real-time landslide data; inputting the global feature vectors into the feature fusion module to perform feature fusion and obtain fused features; and inputting the fused features into the decoding module to decode and obtain the landslide detection results for the target area.

[0111] In practice, the pre-trained model includes a convolutional coding module, a Transformer coding module, a feature fusion module, and a decoding module. The convolutional coding module is used to extract local features of landslide data, while the Transformer coding module is used to extract global features of landslide data.

[0112] In some embodiments, real-time landslide data is input to a convolutional coding module to extract local feature vectors from the real-time landslide data; the local feature vectors are then encoded at their locations and input to a Transformer coding module to extract global feature vectors from the real-time landslide data; the global feature vectors are then input to a feature fusion module to perform feature fusion and obtain fused features; and the fused features are then input to a decoding module to decode and obtain the landslide detection results for the target area.

[0113] In some embodiments, the convolutional coding module includes a convolutional layer with a kernel size of 3×3, a batch normalization layer, and a ReLU activation function. The first convolutional layer aims to extract lower-level local landslide features, while the second convolutional layer further extracts more abstract and higher-level local landslide features. The batch normalization layer and the activation function layer play a crucial role in normalizing the data and introducing nonlinearity. The last convolutional unit contains a convolutional layer with a kernel size of 3×3, a ReLU activation function, and a max-pooling layer.

[0114] For example, real-time landslide data is The following operations are performed to calculate the local feature vector X. p Local feature vectors are used to characterize the multi-scale spatial texture and spectral features of real-time landslide data.

[0115]

[0116] For local eigenvectors X p Perform positional encoding to obtain a feature sequence that carries positional information. .

[0117] The Transformer encoding module consists of traditional ViT structures such as multi-head self-attention, layer normalization, multilayer perceptron, and skip connections. The multilayer perceptron is a fully connected layer with a GELU activation function, and the encoder part is mainly responsible for extracting global features of the landslide.

[0118] Input to the Transformer encoding module In this case, the local feature vector after position encoding is processed, and the specific formula is as follows:

[0119]

[0120]

[0121] Where l represents the number of Transformer encoders, set to 2, while MLP, LN, and MSA represent multi-head self-attention, layer normalization, and multilayer perceptron, respectively, with the number of heads set to 4.

[0122] A token-like alternative called Sequence Pooling (SeqPool) is introduced, which ensures that the output sequence contains relevant information from various regions of the input image by pooling the output sequence of the tokens.

[0123] In some embodiments, if the output of the last Transformer encoder is a global feature vector ,in Represents batch size. For sequence length, To embed the dimension, sequence pooling will be achieved through a series of transformations. Global feature vectors are used to characterize the long-distance dependencies and global semantic information of real-time landslide data.

[0124] Specifically, the global feature vector input fusion module performs fusion. It will be input into a linear transformation layer The Softmax activation function is used to generate the output. :

[0125] T)

[0126] Generate an importance weight for each input token. and then with Multiplication is used to perform sequence pooling and output the result. The formula is as follows:

[0127]

[0128] Then, As a substitute for tokens, it is passed to the decoding module of the pre-trained model. The decoding module is a classifier, which is a fully connected layer with a Softmax activation function and is responsible for generating landslide susceptibility values, i.e., the spatial probability of a landslide occurring.

[0129] In some embodiments, the decoding module is also used for pixel-based segmentation and boundary restoration.

[0130] In the above embodiments, the convolutional coding module first extracts local feature vectors from real-time landslide data, which can effectively capture fine spatial structural information such as the edges and textures of the landslide. After positional encoding, the Transformer coding module further extracts global feature vectors and uses a self-attention mechanism to capture global semantic information such as the shape, size, spatial distribution, and long-distance dependencies of the landslide. The feature fusion module jointly optimizes local and global features in spatial and channel dimensions, enabling the model to take into account both landslide details and overall context. Finally, the decoding module achieves pixel-level segmentation and boundary restoration, outputting high-precision landslide detection results.

[0131] In one exemplary embodiment, historical landslide data includes optical imagery data, mathematical elevation model data, geological element data, and human engineering activity data.

[0132] In practice, historical landslide data includes optical image data, mathematical elevation model data, geological element data, and human engineering activity data.

[0133] In some embodiments, optical imagery data includes satellite imagery of the target area, mathematical elevation model data includes topographic data of the target area, geological element data includes lithology, fault and drainage distribution maps, and human engineering data includes road distances.

[0134] In the above embodiments, optical image data provides information on land cover, vegetation conditions, and spectral characteristics of landslide bodies; topographic factors such as elevation, slope, and aspect derived from digital elevation model data reveal the topographic conditions for landslide occurrence; geological element data (lithology, faults, and drainage distribution) reflects the control effect of regional geological structure and soil properties on landslide stability; and human engineering activity data (such as distance from roads) quantifies the disturbance impact of human activities on slopes.

[0135] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0136] Based on the same inventive concept, this application also provides a landslide detection device for implementing the landslide detection method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more landslide detection device embodiments provided below can be found in the limitations of the landslide detection method described above, and will not be repeated here.

[0137] In one exemplary embodiment, such as Figure 3 As shown, a landslide detection device is provided, including: a processing module 301, a construction module 302, a training module 303, and a calculation module 304, wherein:

[0138] The processing module is used to acquire historical landslide data of the target area, perform image correction and spatial registration processing on the historical landslide data, and obtain processed historical landslide data.

[0139] A construction module is used to build a landslide sample set based on processed historical landslide data; the landslide sample set includes multi-source factor data and landslide label data.

[0140] The processing module is further configured to overlay the multi-source factor data and the landslide label data in the channel dimension to obtain a multi-channel stereo image block with uniform resolution, and to slice the multi-channel stereo image block to obtain a training sample set and a test sample set.

[0141] The training module is used to train a preset model based on the training sample set and the test sample set, and to obtain the trained preset model under the condition of satisfying the preset training conditions.

[0142] The calculation module is used to calculate the performance index of the preset model after training. When the performance index meets the preset performance conditions, it acquires real-time landslide data of the target area and inputs the real-time landslide data into the preset model after training to obtain the landslide detection result of the target area.

[0143] In one exemplary embodiment, the above-described processing module is further configured to:

[0144] Radiometric calibration and atmospheric correction were performed on the image data in the historical landslide data to obtain the corrected historical landslide data.

[0145] The corrected historical landslide data are unified into the same spatial reference system, and all unified historical landslide data are sampled to the same spatial resolution to obtain the processed historical landslide data.

[0146] In one exemplary embodiment, the above-described processing module is further configured to:

[0147] Landslide pixels are randomly selected as center points in multi-channel stereo image blocks. The blocks are then expanded in height, width, and channels around the center point to obtain landslide sample blocks.

[0148] Randomly select non-slope pixels within the target area as the center point, and expand the height, width, and channels around the center point to obtain the same number of non-slope sample blocks as the number of landslide sample blocks.

[0149] Based on landslide sample blocks and non-landslide sample blocks, training sample sets and test sample sets are obtained according to a preset ratio.

[0150] In one exemplary embodiment, the performance metrics include at least one of the following: overall accuracy, precision, recall, F1 score, Matthews correlation coefficient, consistency coefficient, or area under the curve of the trained preset model.

[0151] In one exemplary embodiment, the above-described computing module is further configured to:

[0152] Real-time landslide data is input into a convolutional coding module to extract local feature vectors from the real-time landslide data.

[0153] After the local feature vectors are positionally encoded, they are input into the Transformer encoding module to extract the global feature vectors of the real-time landslide data.

[0154] The global feature vector is input into the feature fusion module for feature fusion to obtain the fused features.

[0155] The fused features are input into the decoding module to obtain the landslide detection results for the target area.

[0156] In one exemplary embodiment, historical landslide data includes optical image data, mathematical elevation model data, geological element data, and human engineering activity data.

[0157] Each module in the aforementioned landslide detection device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the computer device's memory as software, so that the processor can call and execute the corresponding operations of each module.

[0158] In one exemplary embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 4 As shown, the computer device includes a processor, memory, input / output interfaces, a communication interface, a display unit, and an input device. The processor, memory, and input / output interfaces are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The input / output interfaces are used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, Near Field Communication (NFC), or other technologies. When the computer program is executed by the processor, it implements a landslide detection method.

[0159] The display unit of this computer device is used to form a visually visible image and can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of this computer device can be a touch layer covering the display screen, or buttons, a trackball, or a touchpad set on the casing of the computer device, or an external keyboard, touchpad, or mouse, etc.

[0160] Those skilled in the art will understand that Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0161] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0162] Historical landslide data of the target area is acquired, and image correction and spatial registration are performed on the historical landslide data to obtain processed historical landslide data.

[0163] Based on the processed historical landslide data, a landslide sample set is constructed; the landslide sample set includes multi-source factor data and landslide label data.

[0164] The multi-source factor data and the landslide label data are superimposed in the channel dimension to obtain a multi-channel stereo image patch with uniform resolution. The multi-channel stereo image patch is then sliced ​​to obtain a training sample set and a test sample set.

[0165] A preset model is trained based on the training sample set and the test sample set, and the trained preset model is obtained under the condition that the preset training conditions are met.

[0166] Calculate the performance index of the pre-set model after training. If the performance index meets the pre-set performance conditions, obtain real-time landslide data of the target area and input the real-time landslide data into the pre-set model after training to obtain the landslide detection result of the target area.

[0167] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0168] Historical landslide data of the target area is acquired, and image correction and spatial registration are performed on the historical landslide data to obtain processed historical landslide data.

[0169] Based on the processed historical landslide data, a landslide sample set is constructed; the landslide sample set includes multi-source factor data and landslide label data.

[0170] The multi-source factor data and the landslide label data are superimposed in the channel dimension to obtain a multi-channel stereo image patch with uniform resolution. The multi-channel stereo image patch is then sliced ​​to obtain a training sample set and a test sample set.

[0171] A preset model is trained based on the training sample set and the test sample set, and the trained preset model is obtained under the condition that the preset training conditions are met.

[0172] Calculate the performance index of the pre-set model after training. If the performance index meets the pre-set performance conditions, obtain real-time landslide data of the target area and input the real-time landslide data into the pre-set model after training to obtain the landslide detection result of the target area.

[0173] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, performs the following steps:

[0174] Historical landslide data of the target area is acquired, and image correction and spatial registration are performed on the historical landslide data to obtain processed historical landslide data.

[0175] Based on the processed historical landslide data, a landslide sample set is constructed; the landslide sample set includes multi-source factor data and landslide label data.

[0176] The multi-source factor data and the landslide label data are superimposed in the channel dimension to obtain a multi-channel stereo image patch with uniform resolution. The multi-channel stereo image patch is then sliced ​​to obtain a training sample set and a test sample set.

[0177] A preset model is trained based on the training sample set and the test sample set, and the trained preset model is obtained under the condition that the preset training conditions are met.

[0178] Calculate the performance index of the pre-set model after training. If the performance index meets the pre-set performance conditions, obtain real-time landslide data of the target area and input the real-time landslide data into the pre-set model after training to obtain the landslide detection result of the target area.

[0179] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0180] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0181] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0182] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A landslide detection method, characterized in that, The method includes: Historical landslide data of the target area is acquired, and image correction and spatial registration are performed on the historical landslide data to obtain processed historical landslide data. Based on the processed historical landslide data, a landslide sample set is constructed; the landslide sample set includes multi-source factor data and landslide label data. The multi-source factor data and the landslide label data are superimposed in the channel dimension to obtain a multi-channel stereo image patch with uniform resolution. The multi-channel stereo image patch is then sliced ​​to obtain a training sample set and a test sample set. A preset model is trained based on the training sample set and the test sample set, and the trained preset model is obtained under the condition that the preset training conditions are met. Calculate the performance index of the pre-set model after training. If the performance index meets the pre-set performance conditions, obtain real-time landslide data of the target area and input the real-time landslide data into the pre-set model after training to obtain the landslide detection result of the target area.

2. The method according to claim 1, characterized in that, The image correction and spatial registration processing of the historical landslide data to obtain processed historical landslide data includes: Radiometric calibration and atmospheric correction are performed on the image data in the historical landslide data to obtain corrected historical landslide data; The corrected historical landslide data are unified into the same spatial reference system, and all unified historical landslide data are sampled to the same spatial resolution to obtain the processed historical landslide data.

3. The method according to claim 1, characterized in that, The process of slicing multi-channel stereo image blocks to obtain training and testing sample sets includes: Landslide pixels are randomly selected as center points in multi-channel stereo image blocks. The blocks are then expanded in height, width, and channels around the center point to obtain landslide sample blocks. Randomly select non-slope pixels within the target area as the center point, and expand the height, width, and channels around the center point to obtain the same number of non-slope sample blocks as the number of landslide sample blocks. Based on the landslide sample block and the non-landslide sample block, a training sample set and a test sample set are obtained according to a preset ratio.

4. The method according to claim 1, characterized in that, The performance metrics include at least one of the following: overall accuracy, precision, recall, F1 score, Matthews correlation coefficient, consistency coefficient, or area under the curve of the trained preset model.

5. The method according to claim 1, characterized in that, The trained preset model includes a convolutional coding module, a Transformer coding module, a feature fusion module, and a decoding module; the step of inputting the real-time landslide data into the trained preset model to obtain the landslide detection results of the target area includes: The real-time landslide data is input into a convolutional coding module to extract local feature vectors from the real-time landslide data. The local feature vectors are encoded at location and then input into the Transformer encoding module to extract the global feature vectors of the real-time landslide data. The global feature vector is input into the feature fusion module for feature fusion to obtain the fused features; The fused features are input into the decoding module to obtain the landslide detection results for the target area.

6. The method according to claim 1, characterized in that, The historical landslide data includes optical image data, mathematical elevation model data, geological element data, and human engineering activity data.

7. A landslide detection device, characterized in that, The device includes: The processing module is used to acquire historical landslide data of the target area, perform image correction and spatial registration processing on the historical landslide data, and obtain processed historical landslide data. A construction module is used to build a landslide sample set based on processed historical landslide data; the landslide sample set includes multi-source factor data and landslide label data; The processing module is also used to overlay the multi-source factor data and the landslide label data in the channel dimension to obtain a multi-channel stereo image block with uniform resolution, and to slice the multi-channel stereo image block to obtain a training sample set and a test sample set. The training module is used to train a preset model based on the training sample set and the test sample set, and to obtain the trained preset model under the condition of satisfying the preset training conditions. The calculation module is used to calculate the performance index of the preset model after training. When the performance index meets the preset performance conditions, it acquires real-time landslide data of the target area and inputs the real-time landslide data into the preset model after training to obtain the landslide detection result of the target area.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images