Intelligent landslide identification method and system based on hierarchical frame selection and boundary feature fusion

Abstract

The present invention provides a method and system for intelligent landslide identification based on hierarchical frame selection and boundary feature fusion. The method obtains a target remote sensing image dataset, performs hierarchical frame selection on the target remote sensing image dataset, obtains a hierarchical frame selection region set, performs boundary feature extraction on the hierarchical frame selection region set, obtains a boundary feature set for each hierarchical frame selection region, inputs the boundary feature set into a pre-trained boundary feature fusion model for feature fusion processing, generates a fused boundary feature set, performs feature comparison processing based on the fused boundary feature set and a preset landslide morphological feature library, generates a landslide identification feature set, performs regional annotation processing on the target remote sensing image dataset based on the landslide identification feature set, and generates a landslide region identification result. The present invention forms a self-explanatory landslide identification link through topological reconstruction of hierarchical frame selection and feature fusion, avoiding the problem of misjudgment accumulation caused by feature black box transmission in the model.

Description

Technical Field

[0001] The present invention relates to the field of slope identification, and in particular to a landslide intelligent identification method and system based on hierarchical frame selection and boundary feature fusion. Background Art

[0002] In the field of geological disaster monitoring, landslide identification technology, as one of the core application directions of remote sensing image analysis, is mainly used to identify potential landslide areas through surface deformation characteristics and terrain evolution laws. After extracting image features, existing technologies rely on static threshold segmentation or isolated boundary detection algorithms to generate preliminary identification areas, and then perform linear matching verification through a preset landslide morphological feature library. However, due to the fixed segmentation scale, traditional methods are difficult to adapt to the dynamic changes of multi-scale geological features in complex terrain, resulting in insufficient sensitivity in the boundary identification of initial landslides and slow-changing landslides; the simple boundary feature extraction mechanism breaks the correlation between spatial distribution, geometric morphology and surface cover characteristics, resulting in an increase in the misjudgment rate of the transition area between the landslide body and the surrounding terrain; the static feature library matching mode cannot integrate the dynamic evolution of geological conditions in real time, making the identification results susceptible to seasonal changes in surface cover and interference from artificial engineering. Summary of the Invention

[0003] The present invention provides a landslide intelligent identification method and system based on hierarchical frame selection and boundary feature fusion.

[0004] In a first aspect, an embodiment of the present invention provides a method for intelligent landslide identification based on hierarchical frame selection and boundary feature fusion, comprising the following steps:

[0005] Acquire a target remote sensing image dataset, and perform hierarchical frame selection processing on the target remote sensing image dataset to obtain a hierarchical frame selection area set; each hierarchical frame selection area corresponds to a frame selection level identifier;

[0006] Performing boundary feature extraction processing on the hierarchical frame selection area set to obtain a boundary feature set of each hierarchical frame selection area; the boundary feature set includes spatial distribution features, geometric morphology features, and texture association features;

[0007] Inputting the boundary feature set into a pre-trained boundary feature fusion model for feature fusion processing to generate a fused boundary feature set; the fused boundary feature set is used to characterize the correlation between boundary features between different frame selection levels;

[0008] Based on the fused boundary feature set and a preset landslide morphological feature library, feature comparison processing is performed to generate a landslide identification feature set; the landslide identification feature set includes geological deformation features, slope correlation features and surface cover features of the target area;

[0009] The target remote sensing image data set is subjected to region labeling processing according to the landslide identification feature set to generate a landslide region identification result.

[0010] In a second aspect, an embodiment of the present invention provides a computer system, including:

[0011] a memory, wherein the computer program is stored in the memory;

[0012] The processor is used to load the computer program to implement the landslide intelligent identification method based on hierarchical frame selection and boundary feature fusion as described above.

[0013] The intelligent landslide identification method based on hierarchical selection and boundary feature fusion provided by the present invention constructs a remote sensing image processing system with spatial priority identification through a dynamic hierarchical selection mechanism, which effectively solves the problems of incomplete capture of terrain features and blurred boundaries caused by a single segmentation scale in traditional landslide identification methods. Based on the dynamic fusion architecture of the three-dimensional boundary feature set, the nonlinear collaborative calculation of spatial distribution topological relationships, geometric morphological evolution laws and surface texture correlation features is realized for the first time, breaking through the feature fault limitation caused by the isolated analysis of a single boundary dimension in the existing technology. Through the dynamic adjustment mechanism of cross-hierarchical feature association weights, an adaptive matching channel for multi-scale terrain features is established, so that potential landslide bodies at different development stages under complex geological environments can be accurately mapped. The dynamic interaction mechanism of the feature backtracking verification architecture and the landslide morphological feature library realizes the two-way coupling verification of geographic spatial features and geological evolution laws without relying on the historical landslide sample library, significantly improving the recognition sensitivity of primary landslides and hidden landslide boundaries. This method forms a self-explanatory landslide identification link through topological reconstruction of hierarchical frame selection and feature fusion, avoiding the problem of misjudgment accumulation caused by feature black box transmission in existing deep learning models. It reduces the dependence on manual labeling while ensuring the reliability and traceability of geological hazard identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 This is a flow chart of a landslide intelligent identification method based on hierarchical frame selection and boundary feature fusion provided by an embodiment of the present invention.

[0015] Figure 2 It is a schematic diagram of the composition of a computer system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0016] See also Figure 1 , which is a flow chart of a landslide intelligent identification method based on hierarchical frame selection and boundary feature fusion provided by an embodiment of the present invention. The method is executed by a computer system and includes the following steps:

[0017] Step S100: Acquire a target remote sensing image dataset, and perform hierarchical frame selection processing on the target remote sensing image dataset to obtain a hierarchical frame selection area set; each hierarchical frame selection area corresponds to a frame selection level identifier.

[0018] The target remote sensing image dataset refers to a collection of remote sensing image data containing the geographic information of the target area used for intelligent landslide identification. These image data can be obtained through remote sensing equipment such as satellites and drones, and contain rich surface information such as topography, landforms, and vegetation cover. Hierarchical frame selection processing is to perform multi-scale and multi-level regional division of the target remote sensing image dataset to screen out areas where landslides may exist. The hierarchical frame selection area set is a collection of a series of areas obtained after the hierarchical frame selection processing, and each area has its own geographic range and characteristics. The frame selection level identifier is an identifier assigned to each hierarchical frame selection area, which is used to indicate the spatial division priority of the area at different segmentation scales.

[0019] As an implementation method, step S100, obtaining a target remote sensing image dataset, and performing hierarchical frame selection processing on the target remote sensing image dataset to obtain a hierarchical frame selection area set, may specifically include the following steps S110 to S140:

[0020] Step S110: performing multi-scale segmentation processing on the target remote sensing image dataset to obtain a plurality of initial segmentation regions; each initial segmentation region corresponds to a segmentation scale identifier.

[0021] Multiscale segmentation involves segmenting the target remote sensing image dataset at different scales to obtain regions of varying sizes and characteristics. Multiscale segmentation can better adapt to landslide areas of varying sizes and characteristics, improving the accuracy of landslide identification. Initial segmented regions are the basic regional units obtained after multiscale segmentation, each with relatively independent geographic characteristics and attributes. A segmentation scale identifier is assigned to each initial segmented region, indicating the segmentation scale at which the region was obtained and reflecting its size and characteristics.

[0022] As an implementation method, step S110, performing multi-scale segmentation processing on the target remote sensing image dataset to obtain multiple initial segmentation regions, may specifically include the following steps S111 to S114:

[0023] Step S111: Obtain the image resolution and surface cover type distribution information of the target remote sensing image dataset.

[0024] Image resolution refers to the actual ground distance represented by each pixel in a remote sensing image and determines the image's ability to capture detail. The higher the resolution, the richer the image's detail. Land cover type distribution information refers to the distribution of different land cover types (such as vegetation, bare land, and water bodies) within the target area. This information can be obtained through image classification and recognition techniques. Obtaining image resolution and land cover type distribution information provides important information for selecting the subsequent segmentation scale.

[0025] Specifically, image resolution can be obtained by reading image metadata using remote sensing image processing software. For information on the distribution of land cover types, machine learning-based classification algorithms, such as support vector machines (SVM) and random forests (RF), can be used to classify remote sensing images and determine the distribution of different land cover types. For example, when using the SVM algorithm, a certain number of training samples are first selected, each labeled with its land cover type. These samples are then fed into the SVM model for training to produce a classification model. Finally, the target remote sensing image is fed into the trained model for classification prediction, resulting in information on the distribution of land cover types.

[0026] Step S112: determining a segmentation scale range according to the image resolution, and selecting a plurality of candidate segmentation scales within the segmentation scale range based on the surface cover type distribution information.

[0027] The segmentation scale range is a range of possible segmentation scales determined by image resolution, limiting the size of the segmented region. The distribution of land cover types plays an important role in selecting candidate segmentation scales. Different land cover types have varying sensitivities to segmentation scales, so the appropriate segmentation scale should be selected based on the characteristics of the land cover type. Candidate segmentation scales are a set of possible scales selected from the segmentation scale range.

[0028] As an implementation, step S112, determining a segmentation scale range according to the image resolution, and selecting multiple candidate segmentation scales within the segmentation scale range based on the surface cover type distribution information, may specifically include the following steps S1121 to S1126:

[0029] Step S1121: extracting the minimum recognizable unit size contained in the image resolution, and determining the minimum segmentation parameter and the maximum segmentation parameter of the segmentation scale range according to the minimum recognizable unit size.

[0030] The minimum recognizable unit size refers to the size of the smallest geographic unit that can be clearly identified at the image resolution. It is the basis for determining the segmentation scale range. The minimum and maximum segmentation parameters are the lower and upper limits of the segmentation scale range, respectively, and they determine the minimum and maximum sizes of the segmented area.

[0031] Specifically, the minimum recognizable unit size can be determined by analyzing the image resolution. For example, if the image resolution is 1 meter per pixel, the minimum recognizable unit size might be 1 meter by 1 meter. Then, based on experience or experimental data, the minimum and maximum segmentation parameters can be determined. For example, if the minimum segmentation parameter is twice the minimum recognizable unit size and the maximum segmentation parameter is 10 times the minimum recognizable unit size, the segmentation scale range is 2 meters by 2 meters to 10 meters by 10 meters.

[0032] Step S1122: Assign a segmentation parameter adjustment weight to each surface cover type based on the regional proportion and spatial distribution continuity of different surface cover types contained in the surface cover type distribution information; the segmentation parameter adjustment weight is used to characterize the differences in sensitivity of different surface cover types to the segmentation scale.

[0033] Regional share refers to the proportion of the area occupied by different land cover types within the target area, and spatial continuity refers to the degree of spatial continuity within the same land cover type. The segmentation parameter adjustment weight assigns a weight to each land cover type based on regional share and spatial continuity, and is used to adjust the segmentation scale. Different land cover types have different sensitivities to segmentation scale. For example, vegetation-covered areas may require a smaller segmentation scale to accurately identify their boundaries and features, while water areas may be less sensitive to segmentation scale.

[0034] Specifically, statistical analysis of the distribution of land cover types can be used to determine the regional proportions and spatial continuity of different land cover types. Then, based on empirical or experimental data, a segmentation parameter adjustment weight is assigned to each land cover type. For example, higher segmentation parameter adjustment weights are assigned to vegetation-covered areas due to their complex boundaries and features, while lower segmentation parameter adjustment weights are assigned to water bodies.

[0035] Step S1123: performing dynamic range expansion processing on the minimum segmentation parameter and the maximum segmentation parameter based on the segmentation parameter adjustment weight to generate an expanded segmentation scale range.

[0036] Dynamic range expansion involves adjusting the minimum and maximum segmentation parameters based on the segmentation parameter weights to expand the range of segmentation scales. The expanded segmentation scale range can better adapt to the segmentation requirements of different land cover types.

[0037] Specifically, the weights of the segmentation parameters can be adjusted based on the segmentation parameters, and the minimum and maximum segmentation parameters can be weighted. For example, for a certain land cover type, its segmentation parameter adjustment weight is 1.2. Then, when expanding the segmentation scale range, the minimum and maximum segmentation parameters can be multiplied by 1.2 respectively to obtain the expanded minimum and maximum segmentation parameters.

[0038] Step S1124: generating an initial candidate segmentation parameter set according to the segmentation parameter interval step within the expanded segmentation scale range.

[0039] The segmentation parameter interval step size refers to the difference between two adjacent segmentation parameters within the expanded segmentation scale range. The initial candidate segmentation parameter set is a set of possible segmentation parameters generated according to the segmentation parameter interval step size.

[0040] Specifically, we can start from the minimum segmentation parameter after expansion and increase the segmentation parameter interval step size until we reach the maximum segmentation parameter after expansion. These segmentation parameters form the initial candidate segmentation parameter set. For example, if the expanded segmentation scale range is 2.4m×2.4m to 12m×12m and the segmentation parameter interval step size is 0.6m, then the initial candidate segmentation parameter set is {2.4m×2.4m, 3m×3m, 3.6m×3.6m, …, 12m×12m}.

[0041] Step S1125: Based on the spatial aggregation of types in the surface cover type distribution information, cluster and screen the segmentation parameters in the initial candidate segmentation parameter set, remove the segmentation parameters corresponding to those with spatial aggregation lower than the preset type aggregation threshold, and generate an optimized candidate segmentation parameter set.

[0042] Type spatial clustering refers to the degree of spatial aggregation of the same land cover type, reflecting whether the distribution of that type is concentrated. Cluster screening involves filtering the initial set of candidate segmentation parameters based on the type spatial clustering, removing those that are inappropriate for the distribution characteristics of that land cover type. The preset type aggregation threshold is a pre-set threshold used to determine whether the segmentation parameters meet the spatial clustering requirements. The optimized candidate segmentation parameter set is the set of segmentation parameters that, after cluster screening, is more suitable for the distribution characteristics of the land cover type.

[0043] Specifically, a clustering analysis algorithm, such as the K-Means algorithm, can be used to cluster the segmentation parameters in the initial candidate segmentation parameter set. First, the spatial aggregation of the land cover type corresponding to each segmentation parameter is calculated. Then, segmentation parameters with a spatial aggregation below a preset type aggregation threshold are removed to obtain an optimized candidate segmentation parameter set. For example, if the preset type aggregation threshold is 0.8 and the spatial aggregation of the land cover type corresponding to a segmentation parameter is 0.7, then this segmentation parameter will be removed.

[0044] Step S1126: Based on the segmentation parameters in the optimized candidate segmentation parameter set and the type boundary clarity in the surface cover type distribution information, each segmentation parameter is subjected to boundary adaptability verification processing, and the segmentation parameters that meet the boundary consistency condition are retained to generate multiple candidate segmentation scales; the boundary consistency condition is used to ensure that the segmentation scale corresponding to the segmentation parameter can completely cover the continuous area of ​​the same surface cover type and that the distribution of boundary turning points conforms to the preset terrain feature constraint rules.

[0045] Type boundary clarity refers to the clarity of the boundaries between the same land cover type and other types. The boundary adaptability verification process verifies the segmentation parameters in the optimized candidate segmentation parameter set based on the type boundary clarity to determine whether they can accurately segment the boundaries of the surface cover types. The boundary consistency condition is a judgment criterion used to ensure that the segmentation scale corresponding to the segmentation parameter can completely cover the continuous area of ​​the same surface cover type, and that the distribution of boundary turning points conforms to the preset terrain feature constraint rules. The candidate segmentation scale is a set of scales suitable for segmentation obtained after the boundary adaptability verification process.

[0046] Specifically, an edge detection algorithm, such as the Canny edge detection algorithm, can be used to perform boundary detection on the segmentation results corresponding to each segmentation parameter to obtain the distribution of boundary turning points. Then, based on the preset terrain feature constraint rules, it is determined whether the distribution of boundary turning points meets the requirements. For example, if the preset terrain feature constraint rules require that the angle of the boundary turning point be within the target range, the angle of the boundary turning point can be calculated, and the segmentation parameters that do not meet the requirements can be removed. The segmentation parameters that meet the boundary consistency conditions are retained, and multiple candidate segmentation scales are generated.

[0047] Step S113: performing image segmentation processing on the target remote sensing image dataset at each candidate segmentation scale to generate an initial segmentation area; the image segmentation processing includes superpixel segmentation processing and edge detection processing.

[0048] Image segmentation involves segmenting the target remote sensing image dataset according to candidate segmentation scales to obtain initial segmented regions. Superpixel segmentation involves segmenting the image into a series of superpixels with similar characteristics. These superpixels can serve as the basic units of the initial segmented regions. Edge detection involves detecting the boundaries between different regions in the image to determine the boundaries of the initial segmented regions.

[0049] Specifically, the Simple Linear Iterative Clustering (SLIC) algorithm can be used for superpixel segmentation. This algorithm first initializes the pixels in the image and then iteratively updates the pixel labels to divide the pixels into different superpixels. Edge detection can use the Canny edge detection algorithm, which detects edge information in the image by calculating the gradient value of the image. For example, for a candidate segmentation scale, the target remote sensing image dataset is input into the SLIC algorithm to obtain the superpixel segmentation results. The superpixel segmentation results are then input into the Canny edge detection algorithm to obtain the boundary information of the initial segmentation area.

[0050] Step S114: performing scale alignment processing on the initial segmentation regions generated under different candidate segmentation scales to obtain multiple initial segmentation regions; each initial segmentation region corresponds to a segmentation scale identifier and a corresponding segmentation parameter set.

[0051] Scale alignment unifies the initial segmented regions generated at different candidate segmentation scales, ensuring they have the same spatial coordinates and scale. The segmentation scale identifier indicates the scale at which the initial segmented region was generated, and the segmentation parameter set contains the specific segmentation parameters for that scale.

[0052] Specifically, a nearest neighbor interpolation algorithm can be used to align the scales of the initial segmented regions at different candidate segmentation scales. This algorithm achieves scale alignment by finding the nearest pixel value to fill the target pixel. For example, for the initial segmented regions generated at two different candidate segmentation scales, their spatial coordinates are unified, and then the nearest neighbor interpolation algorithm is used to adjust the scale of one region to the same as the other region, resulting in the scale-aligned initial segmented regions. Each initial segmented region is assigned a segmentation scale identifier and a corresponding set of segmentation parameters for subsequent processing.

[0053] Step S120: performing region merging processing on the multiple initial segmented regions according to the segmentation scale identifier to generate a merged region set; each merged region in the merged region set includes at least two initial segmented regions.

[0054] The region merging process is to merge the initial segmented regions with similar features to reduce the number of regions and improve the integrity and coherence of the regions. The merged region set is a set of regions obtained after the region merging process, and each merged region contains at least two initial segmented regions. Specifically, a rule-based merging algorithm can be used to merge regions based on the segmentation scale identifier and the characteristics of the initial segmented regions (such as terrain, vegetation cover, etc.). For example, for adjacent initial segmented regions with the same segmentation scale identifier, if their terrain features and vegetation cover types are similar, they are merged into one merged region. The terrain similarity and vegetation cover similarity of the two initial segmented regions can be calculated, and the merging operation can be performed when the similarity is higher than the preset merging threshold.

[0055] Step S130: performing region screening processing on the merged region set to obtain a candidate region set; each candidate region in the candidate region set meets a preset landslide candidate condition; the landslide candidate condition includes a region area threshold, region terrain continuity, and region texture consistency.

[0056] Region screening involves selecting regions from the merged region set that meet landslide candidate criteria to reduce the workload of subsequent processing. The candidate region set is a set of regions identified as potential landslide sites after the region screening process. The landslide candidate criteria are a set of criteria used to determine whether a region is likely to host a landslide, including a region area threshold, regional terrain continuity, and regional texture consistency. Specifically, the region area threshold can be set based on experience or experimental data. For example, a region area threshold of 100 square meters can be set, and merged regions with an area smaller than this threshold will be removed. Regional terrain continuity can be determined by analyzing the region's topographical undulations and slope variations. Terrain analysis algorithms, such as the slope calculation algorithm and the aspect calculation algorithm, are used to calculate the slope and aspect within the region. When the slope and aspect variations are within a target range, the region is considered to have continuous terrain. Regional texture consistency can be determined by calculating regional texture features, such as the gray-level co-occurrence matrix (GLCM) features. When the texture features are within a target range, the region is considered to have consistent texture. Merged regions that meet the landslide candidate criteria are screened to obtain the candidate region set.

[0057] Step S140: Based on the distribution density and area overlap of the candidate area set, the candidate area set is hierarchically divided to generate a hierarchical frame selection area set; each hierarchical frame selection area corresponds to a frame selection level identifier, and the frame selection level identifier is used to indicate the spatial division priority of the hierarchical frame selection area at different segmentation scales.

[0058] Hierarchical partitioning divides candidate regions into different levels based on their distribution density and overlap, achieving multi-scale spatial partitioning. A hierarchical bounding box region set is a collection of regions at different levels obtained after hierarchical partitioning. The bounding box level identifier indicates the spatial partitioning priority of the hierarchical bounding box region at different segmentation scales. Regions with higher priorities are considered earlier in subsequent processing.

[0059] Specifically, a density clustering algorithm, such as the DBSCAN algorithm, can be used to perform cluster analysis on the candidate region set, and the distribution density of the region can be obtained based on the clustering results. The regional overlap can be determined by calculating the overlapping area between the candidate regions. Based on the distribution density and regional overlap, the candidate regions are divided into different levels. For example, candidate regions with high distribution density and large regional overlap are divided into higher levels, while candidate regions with low distribution density and small regional overlap are divided into lower levels. Each level of the frame selection area is assigned a frame selection level identifier for subsequent processing.

[0060] Step S200: performing boundary feature extraction processing on the hierarchical frame selection area set to obtain a boundary feature set of each hierarchical frame selection area; the boundary feature set includes spatial distribution features, geometric morphology features and texture association features.

[0061] Boundary feature extraction involves extracting boundary features from each hierarchical selection region for subsequent feature fusion and landslide identification. The boundary feature set, derived from the boundary feature extraction process, describes the boundary characteristics of the hierarchical selection region, including spatial distribution features, geometric morphological features, and texture-related features.

[0062] Spatial distribution features describe the spatial distribution of the hierarchical selection area, such as its location, direction, and range. Geometric features describe the geometric shape of the hierarchical selection area, such as its perimeter, area, and shape index. Texture association features describe the texture characteristics of the hierarchical selection area, such as its roughness and contrast.

[0063] Specifically, the spatial distribution features can be obtained by calculating the center point coordinates, the position and size of the circumscribed rectangle of the hierarchical selection area. The geometric morphological features can be obtained by calculating the geometric parameters of the region, such as the perimeter, area, and shape index. The texture association features can be obtained by calculating the gray-level co-occurrence matrix (GLCM) features of the region, such as contrast, correlation, energy, etc. For example, for a certain hierarchical selection area, the boundary is detected using a boundary detection algorithm, and then the center point coordinates of the boundary are calculated as part of the spatial distribution features, the perimeter and area of ​​the boundary are calculated as part of the geometric morphological features, and the GLCM features of the boundary area are calculated as part of the texture association features to obtain the boundary feature set of the hierarchical selection area.

[0064] Step S300: Input the boundary feature set into a pre-trained boundary feature fusion model for feature fusion processing to generate a fused boundary feature set; the fused boundary feature set is used to characterize the correlation between boundary features between different frame selection levels.

[0065] Feature fusion combines different features within a boundary feature set to generate a fused boundary feature set that characterizes the correlations between boundary features at different selection levels. A pretrained boundary feature fusion model is a pretrained model for feature fusion that learns the correlations and interactions between different features, thereby generating more effective fused features.

[0066] Specifically, the pre-trained boundary feature fusion model can employ a deep learning model, such as a convolutional neural network (CNN) or a recurrent neural network (RNN). Taking a CNN as an example, the model consists of an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer. The input layer receives a boundary feature set as input. The convolutional layer extracts local information of the features using convolution kernels. The pooling layer reduces the dimensionality of the features. The fully connected layer integrates the extracted features, and the output layer outputs a fused boundary feature set. During the training phase, the model is trained using a large amount of historical data, and its parameters are adjusted to enable it to learn the associations and interactions between different features. During the inference phase, the boundary feature set is input into the trained model to obtain a fused boundary feature set.

[0067] As an implementation method, the boundary feature fusion model is trained through the following steps S301 to S3010:

[0068] Step S301: Acquire a historical remote sensing image sample set and a corresponding annotated landslide area set; perform hierarchical frame selection processing on the historical remote sensing image sample set to obtain a sample hierarchical framed area set; perform boundary feature extraction processing on the sample hierarchical framed area set to obtain a sample boundary feature set.

[0069] The historical remote sensing image sample set is a collection of historical remote sensing image data. These image data can come from different times and locations and contain rich landslide information. The labeled landslide area set is a set obtained by labeling the landslide areas in the historical remote sensing image sample set. The labeled information may include the location and range of the landslide area. The sample level frame selection area set is a set of a series of areas obtained by performing level frame selection processing on the historical remote sensing image sample set. Each area corresponds to a frame selection level identifier. The sample boundary feature set is a set of features that describes the boundary features of the sample level frame selection area obtained by performing boundary feature extraction processing on the sample level frame selection area set.

[0070] Specifically, the historical remote sensing image sample set can be downloaded through the remote sensing data platform or related database. The method for performing hierarchical frame selection processing on the historical remote sensing image sample set is the same as the method described in step S100. The method for performing boundary feature extraction processing on the sample hierarchical frame selection area set is the same as the method described in step S200. For example, 100 historical remote sensing images are downloaded from the remote sensing data platform as the historical remote sensing image sample set, and hierarchical frame selection processing is performed on each image to obtain a sample hierarchical frame selection area set, and then boundary feature extraction processing is performed on each sample hierarchical frame selection area to obtain a sample boundary feature set.

[0071] Step S302: Initialize the feature fusion network structure; the feature fusion network structure includes a multi-level feature association module, a dynamic weight generation module and a nonlinear mapping module.

[0072] The feature fusion network architecture is used for feature fusion. It consists of multiple modules, each with different functions. The multi-level feature association module is used to learn the feature association relationships between selected regions at different levels. The dynamic weight generation module is used to generate dynamic weights for different features. The nonlinear mapping module is used to perform nonlinear transformations on features to enhance their expressiveness.

[0073] Specifically, the multi-level feature association module can use an attention mechanism, such as a multi-head attention mechanism, to learn the feature association relationship between the selected areas at different levels. The dynamic weight generation module can use a fully connected layer to generate dynamic weights for different features based on the input features. The nonlinear mapping module can use an activation function, such as the ReLU function, to perform nonlinear transformations on the features. For example, when initializing the feature fusion network structure, the number of heads in the multi-level feature association module is set to 8, the number of neurons in the fully connected layer of the dynamic weight generation module is set to 64, and the nonlinear mapping module uses the ReLU activation function.

[0074] Step S303: Input the sample boundary feature set into the multi-level feature association module in the feature fusion network structure to generate a first associated feature sequence of the sample.

[0075] The multi-level feature association module receives a set of sample boundary features as input and generates a sample first-association feature sequence by learning the feature association relationships between selected regions at different levels. This sequence describes the feature association relationships between selected regions at different levels.

[0076] Specifically, the sample boundary feature set is input into the multi-level feature association module. The multi-head attention mechanism of the multi-level feature association module processes the input features, calculates the attention scores between the different levels of framed areas, and then performs weighted summation of the features according to the attention scores to obtain the sample first associated feature sequence. For example, for the boundary features of each level of framed area in the sample boundary feature set, the multi-level feature association module calculates the attention scores between it and the boundary features of other level framed areas, performs weighted summation of the features with higher attention scores, obtains the first associated feature corresponding to the level framed area, and forms the sample first associated feature sequence with the first associated features of all level framed areas.

[0077] Step S304: inputting the first associated feature sequence of the sample into the dynamic weight generation module, and combining the sample geometric features and texture associated features to generate the sample dynamic weight adjustment coefficient.

[0078] The dynamic weight generation module receives the first associated feature sequence of the sample as input and combines the sample's geometric features and texture-related features to generate the sample's dynamic weight adjustment coefficients. The sample dynamic weight adjustment coefficients are a set of coefficients used to adjust the weights of different features during the feature fusion process.

[0079] Specifically, the first associated feature sequence of the sample, along with the sample's geometric features and texture-related features, is input into the fully connected layer of the dynamic weight generation module. The fully connected layer performs a linear transformation on the input features and then uses an activation function (such as the Sigmoid function) to map the output values ​​to the range [0, 1] to obtain the sample's dynamic weight adjustment coefficient. For example, if the fully connected layer has an input feature dimension of 128 and an output feature dimension of 1, the input features are linearly transformed and then the output values ​​are mapped to the range [0, 1] using the Sigmoid function to obtain the sample's dynamic weight adjustment coefficient.

[0080] Step S305: performing weighted aggregation processing on the sample texture-related features based on the sample dynamic weight adjustment coefficient to generate the sample initial fused texture features.

[0081] Weighted aggregation processing is to perform a weighted summation of the sample texture-related features based on the sample dynamic weight adjustment coefficient to generate the sample initial fused texture features. The sample initial fused texture features are a set of features that fuse the texture-related features of the selected areas at different levels.

[0082] Specifically, the sample dynamic weight adjustment coefficient is multiplied element-by-element by the sample texture-related feature, and then the multiplication results are summed to obtain the sample initial fusion texture feature.

[0083] Step S306: performing feature splicing processing on the sample's initial fused texture features and the sample's first associated feature sequence to generate a predicted fused feature set.

[0084] The feature concatenation process is to concatenate the sample's initial fused texture features with the sample's first associated feature sequence to generate a predicted fused feature set. The predicted fused feature set is a feature set that combines the sample's initial fused texture features with the sample's first associated feature sequence.

[0085] Specifically, the sample's initial fused texture features and the sample's first associated feature sequence are concatenated in terms of feature dimensions to obtain a predicted fused feature set. For example, if the sample's initial fused texture features have a dimension of 3 and the sample's first associated feature sequence has a dimension of 10, concatenating them in terms of feature dimensions yields a predicted fused feature set with a dimension of 13.

[0086] Step S307: extracting standard spatial distribution features, standard geometric features and standard texture association features from the set of annotated landslide areas, and constructing a standard fusion feature template.

[0087] Standard spatial distribution features, standard geometric features, and standard texture association features are a set of standard features extracted from the set of annotated landslide areas to describe the characteristics of the landslide area. The standard fusion feature template is a template constructed based on these standard features and is used to evaluate the accuracy of the predicted fusion feature set.

[0088] Specifically, standard spatial distribution features can be extracted by calculating the coordinates of the center point of the annotated landslide area, the position and size of the circumscribed rectangle, and so on. Standard geometric morphological features can be extracted by calculating geometric parameters such as the perimeter, area, and shape index of the annotated landslide area. Standard texture association features can be extracted by calculating the gray-level co-occurrence matrix (GLCM) features of the annotated landslide area, such as contrast, correlation, and energy. These standard features are fused to construct a standard fusion feature template. For example, for each landslide area in the set of annotated landslide areas, its center point coordinates, perimeter, area, GLCM features, etc. are calculated, and these features are fused to construct a standard fusion feature template.

[0089] Step S308: Calculating the distribution difference, morphological matching and texture consistency index between the predicted fusion feature set and the standard fusion feature template.

[0090] Distribution difference refers to the degree of difference in spatial distribution between the predicted fusion feature set and the standard fusion feature template. Morphological matching refers to the degree of matching between the predicted fusion feature set and the standard fusion feature template in geometric morphology. Texture consistency index refers to the degree of consistency between the predicted fusion feature set and the standard fusion feature template in texture features.

[0091] Specifically, the distribution difference can be calculated using a distance measurement method such as Euclidean distance or Manhattan distance to calculate the distance between the predicted fusion feature set and the standard fusion feature template in terms of spatial distribution features. The morphological matching degree can be calculated using a shape matching algorithm, such as the Hausdorff distance algorithm, to calculate the distance between the predicted fusion feature set and the standard fusion feature template in terms of geometric morphological features. The texture consistency index can be calculated using a similarity measurement method such as correlation coefficient or cosine similarity to calculate the similarity between the predicted fusion feature set and the standard fusion feature template in terms of texture-related features. For example, the Euclidean distance is used to calculate the distance between the predicted fusion feature set and the standard fusion feature template in terms of spatial distribution features, the Hausdorff distance algorithm is used to calculate their distance in terms of geometric morphological features, and the cosine similarity is used to calculate their similarity in terms of texture-related features.

[0092] Step S309: Generate a model optimization loss value according to the weighted combination result of the distribution difference, morphological matching and texture consistency indicators.

[0093] The model optimization loss value is calculated based on the weighted combination of distribution difference, morphological matching and texture consistency indicators, and is used to evaluate the degree of difference between the model's prediction results and the standard fusion feature template.

[0094] Specifically, different weights are assigned to the distribution difference, morphological matching, and texture consistency indicators. For example, the weight of distribution difference is 0.3, the weight of morphological matching is 0.3, and the weight of texture consistency is 0.4. The distribution difference, morphological matching, and texture consistency indicators are then multiplied by their corresponding weights, and the results are added together to obtain the model optimization loss value. For example, if the distribution difference is 0.2, the morphological matching is 0.3, and the texture consistency index is 0.8, the model optimization loss value = 0.3 × 0.2 + 0.3 × 0.3 + 0.4 × (1-0.8) = 0.19.

[0095] Step S3010: Based on the model optimization loss value, the network parameters of the multi-level feature association module, the dynamic weight generation module and the nonlinear mapping module are jointly iteratively optimized until the feature deviation between the predicted fusion feature set and the standard fusion feature template is less than the preset optimization threshold, thereby obtaining a trained boundary feature fusion model.

[0096] Joint iterative optimization updates the network parameters of the multi-level feature association module, dynamic weight generation module, and nonlinear mapping module based on the model's optimization loss to improve the model's predictive accuracy. The preset optimization threshold is a pre-set threshold used to determine whether the model has converged. When the feature deviation between the predicted fusion feature set and the standard fusion feature template is less than the preset optimization threshold, the model is considered converged and training is complete.

[0097] Specifically, a gradient descent algorithm, such as stochastic gradient descent (SGD) or adaptive moment estimation (Adam), is used to calculate the gradients of the network parameters of the multi-level feature association module, dynamic weight generation module, and nonlinear mapping module based on the model optimization loss value. The network parameters are then updated based on the gradients. This process is repeated until the feature deviation between the predicted fusion feature set and the standard fusion feature template is less than a preset optimization threshold. For example, if the preset optimization threshold is 0.01, the Adam algorithm is used to update the network parameters. When the feature deviation between the predicted fusion feature set and the standard fusion feature template is less than 0.01, the model is considered converged, resulting in a trained boundary feature fusion model.

[0098] As an implementation method, step S300, inputting the boundary feature set into a pre-trained boundary feature fusion model for feature fusion processing to generate a fused boundary feature set, may specifically include the following steps S310 to S360:

[0099] Step S310: performing multi-level feature association analysis on the spatial distribution features in the boundary feature set to generate a first association feature sequence; the first association feature sequence is used to characterize the spatial distribution association strength between different frame selection levels.

[0100] Multi-level feature association analysis analyzes the spatial distribution features in a set of boundary features to learn the strength of spatial distribution associations between different selection levels. The first association feature sequence is a set of feature sequences that describes the strength of spatial distribution associations between different selection levels.

[0101] Specifically, an attention mechanism, such as a multi-head attention mechanism, is used to process the spatial distribution features in the boundary feature set. The multi-head attention mechanism calculates the attention scores of the spatial distribution features between different frame selection levels, and then performs weighted summation on the features according to the attention scores to obtain a first associated feature sequence. For example, for the spatial distribution features of each frame selection level in the boundary feature set, the multi-head attention mechanism calculates the attention scores between it and the spatial distribution features of other frame selection levels, performs weighted summation on the features with higher attention scores, obtains the first associated feature corresponding to the frame selection level, and forms the first associated features of all frame selection levels into a first associated feature sequence.

[0102] Step S320: performing morphological similarity measurement on the geometric features in the boundary feature set to generate a geometric similarity map; the geometric similarity map is used to indicate the geometric differences of different frame selection levels.

[0103] The morphological similarity metric compares the geometric features in a set of boundary features to calculate the geometric differences between different selection levels. The geometric similarity map is a graph that visually displays the geometric differences between different selection levels.

[0104] Specifically, a shape matching algorithm, such as the Hausdorff distance algorithm, is used to calculate the distances between geometric features at different selection levels. The calculated distances are combined into a matrix, which is the geometric similarity map. For example, for each geometric feature at a selection level in the boundary feature set, the Hausdorff distance algorithm is used to calculate the distance between it and the geometric features at other selection levels. These distances are combined into a matrix, where each element in the matrix represents the degree of geometric difference between the two selection levels.

[0105] Step S330: Perform cross-feature matching based on the first associated feature sequence and the geometric morphology similarity map to generate a dynamic weight adjustment coefficient; the dynamic weight adjustment coefficient is used to balance the influence of the spatial distribution and geometric morphology of different selection levels on the fusion result.

[0106] Cross-feature matching matches the first-correlated feature sequence with the geometric similarity map to generate dynamic weight adjustment coefficients. Dynamic weight adjustment coefficients are a set of coefficients used to adjust the weights of the spatial distribution and geometric forms of different frame selection levels during the feature fusion process.

[0107] Specifically, the first correlation feature sequence and the geometric similarity map are input into a fully connected layer. The fully connected layer performs a linear transformation on the input features and then uses an activation function (such as a sigmoid function) to map the output values ​​to the range [0, 1] to obtain the dynamic weight adjustment coefficient. For example, the input feature dimension of the fully connected layer is the dimension of the first correlation feature sequence plus the dimension of the geometric similarity map, and the output feature dimension is 1. After the input features are linearly transformed, the output value is mapped to the range [0, 1] using a sigmoid function to obtain the dynamic weight adjustment coefficient.

[0108] Step S340: performing adaptive weighted aggregation processing on the texture-related features in the boundary feature set according to the dynamic weight adjustment coefficient to generate an initial fused texture feature.

[0109] Adaptive weighted aggregation involves weighting and summing the texture-related features in the boundary feature set according to a dynamic weight adjustment coefficient to generate an initial fused texture feature. This initial fused texture feature is a set of features that fuses texture-related features from different frame selection levels. Specifically, the dynamic weight adjustment coefficient is element-wise multiplied with the texture-related features in the boundary feature set, and the multiplication results are summed to generate the initial fused texture feature.

[0110] Step S350: performing feature splicing processing on the initial fused texture feature and the first associated feature sequence to generate a cross-level splicing feature.

[0111] Feature concatenation involves concatenating the initial fused texture features with the first associated feature sequence to generate cross-level concatenated features. Cross-level concatenated features are a set of features that are a fusion of the initial fused texture features and the first associated feature sequence. Specifically, the initial fused texture features and the first associated feature sequence are concatenated along the feature dimension to generate the cross-level concatenated features.

[0112] Step S360: Perform nonlinear mapping processing on the cross-level splicing features to eliminate feature redundancy and enhance the consistency of feature expression across levels, and generate a fused boundary feature set; each feature in the fused boundary feature set contains fused expression information of spatial distribution, geometric form and texture association.

[0113] Nonlinear mapping is a nonlinear transformation of cross-level splicing features to eliminate feature redundancy and enhance cross-level feature expression consistency. The fused boundary feature set is a set of features obtained after nonlinear mapping that integrates spatial distribution, geometric form, and texture association information.

[0114] Specifically, an activation function, such as the ReLU function, is used to perform a nonlinear transformation on the cross-level spliced ​​features. The ReLU function sets negative values ​​in the input features to 0 and retains positive values, thereby eliminating feature redundancy. After processing with the ReLU function, the expression consistency of the features is enhanced, generating a fused boundary feature set. For example, for the cross-level spliced ​​features [1, -2, 3, 4, -5, 6], after ReLU processing, the fused boundary feature set [1, 0, 3, 4, 0, 6] is obtained.

[0115] Step S400: performing feature comparison processing based on the fused boundary feature set and a preset landslide morphological feature library to generate a landslide identification feature set; the landslide identification feature set includes geological deformation features, slope correlation features and surface cover features of the target area.

[0116] Feature matching involves comparing the fused boundary feature set with a pre-defined landslide morphological feature library to identify features that match the landslide's morphological characteristics. The landslide morphological feature library is a database of various landslide morphological features, constructed using historical data and expert knowledge. The landslide identification feature set, derived from feature matching, is a set of features that describe the potential for landslides in the target area. These features include geological deformation characteristics, slope-related characteristics, and surface cover characteristics.

[0117] Specifically, each feature in the fused boundary feature set is compared to a feature in the landslide morphological feature library for similarity. When the similarity exceeds a preset matching threshold, the feature is considered a match with the landslide morphological feature. These matched features are then combined into a landslide identification feature set. For example, if the preset matching threshold is 0.8 and a feature in the fused boundary feature set has a similarity of 0.9 to a feature in the landslide morphological feature library, the feature is considered a match with the landslide morphological feature and is added to the landslide identification feature set.

[0118] As an embodiment, step S400 performs feature comparison processing based on the fused boundary feature set and a preset landslide morphological feature library to generate a landslide identification feature set, which may specifically include the following steps S410 to S450:

[0119] Step S410: obtaining a standard landslide morphological feature set from a landslide morphological feature library; the standard landslide morphological feature set includes geological deformation characteristics, slope correlation characteristics, and surface cover characteristics of a typical landslide area.

[0120] The standard landslide morphological feature set is a set of standard features extracted from the landslide morphological feature library to describe the characteristics of typical landslide areas. These features can be obtained by analyzing and summarizing a large amount of historical landslide data.

[0121] Specifically, the characteristics of typical landslide areas related to the current target area are screened from the landslide morphological feature library to form a standard landslide morphological feature set. For example, if the current target area is a mountainous area, the geological deformation characteristics, slope correlation characteristics, and surface cover characteristics of typical landslide areas in mountainous areas are screened from the landslide morphological feature library to form a standard landslide morphological feature set.

[0122] Step S420: performing feature analysis processing on each fused boundary feature in the fused boundary feature set to extract target geological deformation parameters, target slope parameters, and target surface cover parameters.

[0123] Feature parsing involves analyzing each fused boundary feature in the fused boundary feature set to extract the target geological deformation parameters, target slope parameters, and target surface cover parameters contained therein. These parameters describe the geological deformation, slope, and surface cover of the target area.

[0124] Specifically, feature extraction algorithms, such as principal component analysis (PCA) or independent component analysis (ICA), are used to reduce the dimensionality of the fused boundary features and extract their key features. These key features are then used to derive the target geological deformation parameters, target slope parameters, and target land cover parameters. For example, PCA can be used to reduce the dimensionality of the fused boundary features to obtain key features. Based on these key features, the target geological deformation parameter can be determined to be land subsidence, the target slope parameter to be average slope, and the target land cover parameter to be vegetation coverage.

[0125] Step S430: Calculate similarity between the target geological deformation parameter, the target slope parameter and the target surface cover parameter and the corresponding parameters in the standard landslide morphological feature set to generate geological deformation similarity, slope correlation similarity and surface cover similarity.

[0126] Similarity calculation compares the target geological deformation parameters, target slope parameters, and target surface cover parameters with the corresponding parameters in the standard landslide morphological feature set to determine the degree of similarity between them. Geological deformation similarity, slope correlation similarity, and surface cover similarity are indicators that describe the degree of similarity between the target area's geological deformation, slope, and surface cover, respectively, and the standard landslide morphological feature set.

[0127] Specifically, a similarity measurement method, such as cosine similarity or Euclidean distance, is used to calculate the similarity between the target geological deformation parameters and the geological deformation parameters in the standard landslide morphological feature set to obtain the geological deformation similarity. Similarly, the similarity between the target slope parameters and the slope parameters in the standard landslide morphological feature set is calculated to obtain the slope association similarity. The similarity between the target surface cover parameters and the surface cover parameters in the standard landslide morphological feature set is calculated to obtain the surface cover similarity. For example, the similarity between the target geological deformation parameters and the geological deformation parameters in the standard landslide morphological feature set is calculated using cosine similarity. When the similarity is 0.8, the geological deformation similarity is considered to be 0.8.

[0128] Step S440: Determine the landslide identification confidence corresponding to the fused boundary feature based on the weighted sum of the geological deformation similarity, the slope correlation similarity, and the surface cover similarity.

[0129] The landslide identification confidence level is a probability value used to determine whether a landslide exists in the target area corresponding to the fused boundary features. The landslide identification confidence level is obtained by taking the weighted sum of the geological deformation similarity, slope correlation similarity, and surface cover similarity.

[0130] Specifically, different weights are assigned to geological deformation similarity, slope correlation similarity, and land cover similarity. For example, the weight for geological deformation similarity is 0.4, the weight for slope correlation similarity is 0.3, and the weight for land cover similarity is 0.3. The geological deformation similarity, slope correlation similarity, and land cover similarity are then multiplied by their corresponding weights, and the results are added together to obtain the landslide identification confidence level. For example, if the geological deformation similarity is 0.8, the slope correlation similarity is 0.7, and the land cover similarity is 0.6, the landslide identification confidence level is 0.4 × 0.8 + 0.3 × 0.7 + 0.3 × 0.6 = 0.71.

[0131] Step S450: Determine the feature set of the target area corresponding to the fused boundary features whose landslide identification confidence is greater than a preset confidence threshold as the landslide identification feature set.

[0132] The preset confidence threshold is a pre-set threshold used to determine whether a landslide exists in the target area corresponding to the fused boundary feature. When the landslide identification confidence exceeds the preset confidence threshold, the target area corresponding to the fused boundary feature is considered to have a high probability of a landslide, and the feature set corresponding to the target area is determined as the landslide identification feature set.

[0133] Specifically, each fused boundary feature in the fused boundary feature set is traversed and its corresponding landslide identification confidence is calculated. Fused boundary features with landslide identification confidence greater than a preset confidence threshold are selected, and the feature set of the target area corresponding to each fused boundary feature is added to the landslide identification feature set. For example, if the preset confidence threshold is 0.7 and the corresponding landslide identification confidence for a fused boundary feature is 0.8, the feature set of the target area corresponding to this fused boundary feature is added to the landslide identification feature set.

[0134] As an embodiment, step S450 determines the feature set of the target area corresponding to the fused boundary features whose landslide identification confidence is greater than a preset confidence threshold as the landslide identification feature set, which may specifically include the following steps S451 to S455:

[0135] Step S451: Obtain the frame selection level identifier and spatial position information corresponding to the fusion boundary feature set.

[0136] The selection level identifier is used to indicate the spatial division priority of the hierarchical selection area corresponding to the fused boundary feature at different segmentation scales. The spatial location information is used to describe the geographical location of the target area corresponding to the fused boundary feature.

[0137] Specifically, the selection level identifier and spatial location information corresponding to each fused boundary feature are extracted from the fused boundary feature set. For example, for a fused boundary feature in the fused boundary feature set, its corresponding selection level identifier is "Level 3" and its spatial location information is "Longitude 110°, Latitude 30°."

[0138] Step S452: determining the spatial division priority of the target area according to the frame selection level identifier, and performing spatial clustering processing on the target area according to the spatial position information to generate a set of candidate landslide areas.

[0139] Spatial priority is used to determine the dependency of different target areas in the processing order. Spatial clustering is to cluster target areas with similar spatial locations to generate a set of candidate landslide areas.

[0140] Specifically, the spatial division priority of each target area is determined according to the frame selection level identification, and the target areas with higher priority are processed first. Then, a spatial clustering algorithm, such as the DBSCAN algorithm, is used to cluster the target areas according to the spatial location information. The DBSCAN algorithm clusters the target areas with similar spatial locations into one category based on the distance and density between the target areas, and generates a set of candidate landslide areas. For example, for the target area set {area 1, area 2, area 3, area 4}, the spatial division priority of area 1 is determined to be the highest according to the frame selection level identification, and it is processed first. Then, the DBSCAN algorithm is used to cluster areas 1, 2, 3, and 4. For example, areas 1 and 2 are spatially close and clustered into one category, and areas 3 and 4 are spatially close and clustered into one category, generating a set of candidate landslide areas {{area 1, area 2}, {area 3, area 4}}.

[0141] As an embodiment, step S452 determines the spatial division priority of the target area according to the frame selection level identifier, and performs spatial clustering processing on the target area according to the spatial location information to generate a set of candidate landslide areas. Specifically, the following steps S4521 to S4528 may be included:

[0142] Step S4521: extracting the level division rules corresponding to the frame selection level identifier, where the level division rules include the decreasing relationship of spatial coverage between different frame selection levels and the level weight allocation strategy.

[0143] The hierarchical division rules are determined based on the selection level identifiers. They clarify the spatial coverage of different selection levels and the weights assigned to each level. The decreasing spatial coverage relationship reflects the gradual decrease in spatial extent as the selection level increases, reflecting the spatial division process from macro to micro perspectives. The hierarchical weighting strategy assigns different levels of importance, reflecting their priority in subsequent processing.

[0144] Specifically, the selected level identifier can be matched with a pre-set level division rule table to extract the corresponding rules. For example, if the selected level identifier is "Level 2", the level division rule table can be used to find the decreasing ratio of spatial coverage between this level and other levels (such as Level 1 and Level 3), as well as the weight coefficient of Level 2 in the overall processing. For example, the level division rule table stipulates that the spatial coverage of Level 2 is 50% of Level 1, and the level weight of Level 2 is 0.3.

[0145] Step S4522: Determine the spatial division priority corresponding to each frame selection level based on the decreasing relationship of spatial coverage and the level weight allocation strategy; the spatial division priority is used to indicate the dependency of different frame selection levels in the processing order.

[0146] Spatial prioritization is determined by a combination of decreasing spatial coverage and a weighted distribution strategy. This determines the order in which different selection levels are processed. Higher-priority levels are processed first, helping to ensure precise identification of target areas from a macro to micro perspective throughout the overall processing.

[0147] Specifically, for each frame selection level, a comprehensive priority index can be calculated based on its spatial coverage and level weight. For example, the weighted average method can be used to weight the spatial coverage decrease ratio and the level weight. For example, the spatial coverage decrease ratio of level 2 is 0.5, the level weight is 0.3, the spatial coverage decrease ratio of level 3 is 0.2, and the level weight is 0.5. Then the comprehensive priority index of level 2 is 0.5×0.3=0.15, and the comprehensive priority index of level 3 is 0.2×0.5=0.1. It can be seen that the spatial division priority of level 2 is higher than that of level 3.

[0148] Step S4523: Hierarchical sorting is performed on the target areas based on their spatial priority, generating a sorted sequence of target areas. Hierarchical sorting sorts the target areas by their spatial priority, placing higher-priority target areas first and lower-priority target areas last, forming an ordered sequence. This sorting helps subsequent processing proceed in a reasonable order, improving processing efficiency and accuracy.

[0149] During specific operations, you can use a sorting algorithm, such as the quick sort algorithm. For a set containing multiple target areas, sort the target areas based on the spatial priority of the selected level to which each target area belongs. For example, if target area A belongs to level 2, target area B belongs to level 3, and target area C belongs to level 1, the sorted target area sequence based on the previously calculated priority is [target area C, target area A, target area B].

[0150] Step S4524: performing spatial proximity analysis on the sorted target region sequence according to the coordinate distribution characteristics and region density characteristics in the spatial position information to generate a set of adjacent region clusters.

[0151] Spatial proximity analysis analyzes the coordinate distribution and regional density characteristics of target areas to identify spatially adjacent target areas and cluster them. A neighboring region cluster set is a collection of region clusters obtained through spatial proximity analysis. The target areas within each cluster are spatially adjacent.

[0152] Specifically, the DBSCAN (density-based spatial clustering application) algorithm can be used for spatial proximity analysis. The algorithm first defines a neighborhood radius and a minimum point count threshold. For each target area in the sorted target area sequence, the number of points within the neighborhood radius is counted with it as the center. If the number of points exceeds the minimum point count threshold, these adjacent target areas are clustered into a cluster. For example, the neighborhood radius is set to 10 meters and the minimum point count threshold is 3. For target area D, there are target areas E and target areas F within its 10-meter neighborhood, and the minimum point count threshold is met, then target areas D, E, and F will be clustered into a neighborhood cluster.

[0153] Step S4525: Performing hierarchical consistency verification on the set of adjacent region clusters, and screening region clusters that meet the hierarchical division rules as preliminary candidate region clusters.

[0154] Hierarchical consistency verification checks whether each cluster in the set of neighboring clusters complies with the hierarchical partitioning rules. Only clusters that meet the rules are selected as preliminary candidate clusters, which helps ensure the rationality of the hierarchical structure of the regions subsequently processed.

[0155] Specifically, hierarchical consistency verification can be performed from two aspects. On the one hand, check whether the frame selection level to which the target areas within the regional cluster belong conforms to the decreasing relationship of spatial coverage and the hierarchical weight allocation strategy. On the other hand, check whether the overall hierarchical attributes of the regional cluster match the hierarchical division rules. For example, if the target areas within a regional cluster belong to levels 2 and 3 respectively, but there are too many target areas in level 2, which does not conform to the spatial coverage and weight ratio of the level in the hierarchical division rules, then the regional cluster will be excluded.

[0156] Step S4526: Based on the spatial distribution consistency of the geological deformation characteristics and slope-related characteristics of the preliminary candidate region clusters, the preliminary candidate region clusters are merged to generate a merged candidate region cluster.

[0157] The regional merging process combines clusters with similar characteristics based on the spatial consistency of their geologic deformation and slope-related characteristics. This consistency reflects the similarity of the regions in terms of geology and topography. Merging clusters with similar characteristics reduces redundancy and improves the efficiency of subsequent processing.

[0158] Specifically, the similarity of geological deformation characteristics and slope-related characteristics between preliminary candidate regional clusters can be calculated. The similarity can be calculated using a cosine similarity algorithm or a Euclidean distance algorithm. When the similarity of geological deformation characteristics and slope-related characteristics of two regional clusters are both higher than a preset similarity threshold, the two regional clusters are merged. For example, if the preset similarity threshold is 0.8, the similarity of geological deformation characteristics of regional cluster G and regional cluster H is 0.85, and the similarity of slope-related characteristics is 0.9, then regional cluster G and regional cluster H are merged into one merged candidate regional cluster.

[0159] Step S4527: Based on the spatial boundary overlap and surface coverage continuity of the merged candidate region cluster, the merged candidate region cluster is subjected to boundary optimization processing to eliminate boundary conflict areas and generate smooth candidate regions.

[0160] Boundary optimization adjusts the boundaries of candidate region clusters to eliminate conflicting areas and ensure continuity of surface cover. Spatial boundary overlap reflects the overlap between the boundaries of candidate region clusters, while surface cover continuity reflects the degree of coherence of the surface cover types within the regions.

[0161] In specific operations, a boundary smoothing algorithm, such as the Laplace smoothing algorithm, can be used. This algorithm reduces boundary mutations by smoothing the boundary points of the merged candidate region clusters. For boundary conflict areas, adjustments can be made based on the spatial boundary overlap to reasonably divide the overlapping parts. At the same time, the surface cover continuity is checked, and areas with incoherent surface cover types are corrected. For example, if there is boundary overlap between merged candidate region cluster I and merged candidate region cluster J, the boundary is smoothed using the Laplace smoothing algorithm, and the overlapping part is divided into two region clusters according to a set ratio based on the spatial boundary overlap. At the same time, areas with incoherent surface cover types are corrected to generate smoothed candidate regions.

[0162] Step S4528: performing feature matching verification processing on the smooth candidate area and the landslide identification feature set, retaining the area whose matching degree meets the preset landslide morphological constraint conditions, and generating a candidate landslide area set.

[0163] Feature matching verification compares the smooth candidate region with the landslide identification feature set to check whether the smooth candidate region's features match the landslide identification features. The preset landslide morphology constraint is a pre-defined criterion used to determine whether the smooth candidate region has landslide characteristics.

[0164] Specifically, the degree of match between the smooth candidate region and each feature in the landslide identification feature set can be calculated. This degree of match can be calculated using a feature similarity calculation method, such as cosine similarity calculation based on feature vectors. When the degree of match between the smooth candidate region and the landslide identification feature exceeds a preset landslide morphological constraint, the region is considered likely to contain a landslide and is retained as a candidate landslide region. For example, if the preset landslide morphological constraint is a match greater than 0.7, and the degree of match between smooth candidate region K and a feature in the landslide identification feature set is 0.8, then smooth candidate region K is added to the candidate landslide region set.

[0165] Step S453: performing a regional merging verification process on each candidate landslide region in the candidate landslide region set to obtain a verified merged region; the regional merging verification process includes regional overlap verification and geological continuity verification.

[0166] Regional merging verification further verifies the candidate landslide regions in the candidate landslide region set to ensure that the merged regions are spatially reasonable and geologically continuous. Regional overlap verification checks the overlap between candidate landslide regions, and geological continuity verification checks whether the merged regions are geologically coherent.

[0167] Specifically, for every two candidate landslide areas in the candidate landslide area set, their regional overlap is calculated. The regional overlap can be obtained by calculating the ratio of the overlapping area of ​​the two areas to the total area. If the regional overlap is higher than the preset overlap threshold, the two areas are considered to be merged. At the same time, geological continuity verification is performed. By analyzing geological information such as geological deformation characteristics and slope correlation characteristics, it is determined whether the regional geology after the merger is continuous. For example, if the preset overlap threshold is 0.3, the regional overlap of candidate landslide area L and candidate landslide area M is 0.4, and the geological continuity verification is passed, then candidate landslide area L and candidate landslide area M are merged into a verification merged area.

[0168] Step S454: performing a secondary comparison process on the feature set corresponding to the verification merged area and the standard landslide morphological feature set to generate a secondary comparison similarity.

[0169] The secondary comparison process involves comparing the feature set corresponding to the verified merged area with the standard landslide morphological feature set to more accurately determine whether the verified merged area is a landslide area. The secondary comparison similarity is an indicator of the degree of similarity between the verified merged area and the standard landslide morphological feature set.

[0170] In specific operations, a similar similarity calculation method as in step S430 can be used to calculate similarities between the geological deformation characteristics, slope-related characteristics, and surface cover characteristics of the verification merged area and the corresponding characteristics in the standard landslide morphological characteristic set. These similarities are then combined to obtain a secondary comparison similarity. For example, a weighted average method can be used to assign different weights to the geological deformation characteristic similarity, slope-related similarity, and surface cover similarity to calculate the secondary comparison similarity.

[0171] Step S455: If the secondary comparison similarity is greater than the preset secondary threshold, the feature set corresponding to the verification merged area is added to the landslide identification feature set.

[0172] The preset quadratic threshold is a pre-set criterion used to determine whether the merged region meets the characteristics of a landslide. When the quadratic similarity is greater than the preset quadratic threshold, the merged region is considered a landslide region, and its corresponding feature set is added to the landslide identification feature set.

[0173] Specifically, the calculated quadratic comparison similarity is compared with a preset quadratic threshold. For example, if the preset quadratic threshold is 0.7 and the quadratic comparison similarity of the verification merged region N is 0.8, then the feature set corresponding to the verification merged region N is added to the landslide identification feature set.

[0174] Step S500: performing region labeling processing on the target remote sensing image dataset according to the landslide identification feature set to generate a landslide region identification result.

[0175] Regional labeling involves marking areas of potential landslides on a target remote sensing image dataset based on a set of landslide identification features, thereby generating landslide region identification results. Regional labeling can intuitively demonstrate which parts of the target area are likely to experience landslides, providing important evidence for subsequent geological disaster prevention and control.

[0176] As an embodiment, step S500, performing region annotation processing on the target remote sensing image dataset according to the landslide identification feature set to generate a landslide region identification result, may specifically include the following steps S510 to S570:

[0177] Step S510: performing spatial continuity analysis on the geological deformation features in the landslide identification feature set to generate a deformation continuous area distribution map; the deformation continuous area distribution map is used to identify target areas with consistent surface deformation trends and coherent spatial distribution.

[0178] Spatial continuity analysis analyzes the geological deformation characteristics within a landslide identification feature set to identify areas with consistent surface deformation trends and spatial coherence. A continuous deformation area distribution map is a visualization that graphically displays these target areas with continuous deformation characteristics.

[0179] Specifically, spatial clustering algorithms, such as the DBSCAN algorithm, can be used to cluster geological deformation features. This algorithm clusters spatially adjacent regions with similar deformation trends based on the spatial distribution and similarity of geological deformation features. Each cluster is then marked on the target remote sensing image dataset to form a distribution map of continuous deformation regions. For example, in a target remote sensing image dataset of a mountainous area, clustering geological deformation features using the DBSCAN algorithm revealed several regions with consistent surface deformation trends and spatial proximity. These regions were then marked on the image to generate a distribution map of continuous deformation regions.

[0180] Step S520: performing a superposition analysis on the deformation continuous region distribution map and the slope correlation feature to extract candidate deformation regions where the slope change overlaps with the deformation continuous region.

[0181] Overlay analysis involves overlaying the distribution map of continuous deformation areas with slope-related features to identify areas where slope changes overlap with continuous deformation areas. These overlapping areas are considered candidate deformation areas. Overlapping slope changes with continuous deformation areas may indicate an area is more susceptible to landslides.

[0182] For specific operations, the overlay analysis function of the Geographic Information System (GIS) software can be used. The deformation continuous area distribution map and the slope-related feature layer are imported into the GIS software and overlaid. The software automatically identifies areas where slope changes overlap with deformation continuous areas and extracts them as candidate deformation areas. For example, in the GIS software, by overlaying the deformation continuous area distribution map and the slope-related feature layer, some areas are found to have both continuous surface deformation and obvious slope changes. These areas are extracted as candidate deformation areas.

[0183] Step S530: performing texture pattern matching processing based on the texture association features of the candidate deformation regions to generate a texture consistency region set; the regions in the texture consistency region set meet the preset landslide texture evolution law.

[0184] Texture pattern matching is a process that analyzes the texture correlation characteristics of candidate deformation regions to identify areas that meet pre-defined landslide texture evolution patterns. These patterns are texture variation characteristics summarized based on historical landslide data and geological knowledge.

[0185] Specifically, a template matching algorithm can be used for texture pattern matching. First, a texture template is constructed based on the preset landslide texture evolution law. Then, the texture-associated features of the candidate deformation region are matched with the texture template. When the matching degree between the texture features of the candidate deformation region and the texture template is higher than the preset matching threshold, the region is considered to meet the landslide texture evolution law and is added to the texture consistency region set. For example, if the preset matching threshold is 0.7 and the matching degree between the texture features of the candidate deformation region P and the texture template is 0.8, then the candidate deformation region P is added to the texture consistency region set.

[0186] Step S540: performing boundary closure verification processing on the texture consistency region set, and screening regions with a boundary turning point density lower than a preset closure threshold as closed candidate regions.

[0187] Boundary closure verification checks the boundary closure of each region in a set of texture-consistent regions. Boundary turning point density is a measure of boundary closure, representing the ratio of the number of turning points on a boundary to the boundary length. The preset closure threshold is a pre-defined criterion used to determine whether a region's boundary is closed.

[0188] Specifically, the density of boundary turning points for each region in the set of texture-consistent regions can be calculated. By analyzing the region's boundaries, counting the number of turning points on the boundary, and calculating the boundary length, the boundary turning point density is obtained. Regions with a boundary turning point density below a preset closure threshold are selected as closed candidate regions. For example, if the preset closure threshold is 0.1 and the boundary turning point density of texture-consistent region Q is 0.08, then texture-consistent region Q is selected as a closed candidate region.

[0189] Step S550: performing coverage type matching processing on the spatial extent of the closed candidate area and the surface coverage features in the target remote sensing image dataset, excluding interference areas whose coverage types do not match the landslide surface types, and generating an initial landslide annotation area.

[0190] Land cover matching involves comparing the spatial extent of closed candidate regions with the surface cover characteristics in the target remote sensing image dataset, identifying and eliminating areas where the land cover type does not match the landslide surface type. The landslide surface type is a landslide-related surface cover type derived from geological knowledge and historical data.

[0191] Specifically, the spatial extent of the closed candidate regions is compared to the surface cover classification layer in the target remote sensing image dataset. For each closed candidate region, its cover type is checked to see if it matches the landslide surface type. If not, the region is excluded. For example, if the landslide surface type is typically bare ground or sparse vegetation, and a closed candidate region has a cover type of dense forest, then this region is excluded, ultimately generating the initial landslide annotated region.

[0192] Step S560: performing multi-scale context consistency correction processing on the initial landslide annotated area, fusing the boundary feature correlations of different frame selection levels, and generating a corrected landslide annotated area.

[0193] Multi-scale contextual consistency correction is a process that modifies the initial landslide annotation by comprehensively considering the correlation of boundary features at different selection levels to improve the accuracy and consistency of the annotation. Boundary features at different selection levels may contain landslide information at different scales. Fusion of these features can provide a more comprehensive description of the landslide area.

[0194] As an embodiment, step S560 performs multi-scale context consistency correction processing on the initial landslide annotated area, integrates the boundary feature correlations of different frame selection levels, and generates a corrected landslide annotated area. Specifically, the following steps S561 to S568 may be included:

[0195] Step S561: extracting the frame selection level identifier and boundary feature correlation parameters corresponding to the initial landslide marked area.

[0196] The frame selection level identifier is used to indicate the frame selection level to which the initial landslide annotation area belongs, and the boundary feature association parameter describes the boundary feature association relationship between the area and other frame selection levels.

[0197] Specifically, the frame selection level identifier and boundary feature correlation parameters are extracted from the relevant data of the initial landslide annotated area. For example, by querying the database or data file, the frame selection level identifier corresponding to the initial landslide annotated area is obtained as "Level 3", and its boundary feature correlation parameters with other levels, such as the boundary overlap ratio with Level 2, are obtained.

[0198] Step S562: Obtain a set of boundary features of adjacent levels in the level selection area set according to the level selection level identifier.

[0199] The adjacent layer refers to the layer adjacent to the frame-selected layer to which the initial landslide annotation area belongs. Obtaining the boundary feature set of the adjacent layer can provide more information for subsequent context association analysis.

[0200] Specifically, the layer selection level identifier is used to search for adjacent layers within the layer selection area set. For example, if the initial landslide annotation area belongs to the selection level "Level 3," the boundary feature sets of Levels 2 and 4 are searched. For each selected region in Levels 2 and 4, the boundary features are extracted to form the boundary feature sets of the adjacent layers.

[0201] Step S563: performing context association analysis on the boundary feature sets of adjacent levels to generate cross-level boundary constraints; the cross-level boundary constraints are used to limit the spatial offset amplitude of boundary features of different levels.

[0202] Contextual association analysis analyzes boundary feature sets from adjacent levels to identify correlations between boundary features at different levels, thereby generating cross-level boundary constraints. Cross-level boundary constraints ensure spatial consistency between boundary features at different levels, preventing excessive offsets. Specifically, machine learning algorithms, such as support vector machines (SVMs), can be used to analyze boundary feature sets from adjacent levels. This algorithm learns the relationships between boundary features and builds a classification model to determine whether the spatial offsets of boundary features at different levels are reasonable. Based on the model's output, cross-level boundary constraints are generated. For example, an SVM algorithm analyzing boundary feature sets from levels 2 and 3 reveals that the spatial offset of boundary features at level 3 relative to those at level 2 cannot exceed a set distance. This distance is then used as a cross-level boundary constraint.

[0203] Step S564: dynamically adjusting the boundary of the initial landslide marked area based on the cross-level boundary constraint conditions to generate an adjusted boundary area.

[0204] The dynamic adjustment process is to adjust the boundaries of the initial landslide marking area according to the cross-level boundary constraints so that it meets the spatial consistency requirements of the boundary characteristics of different levels.

[0205] Specifically, each boundary point in the initial landslide annotation area is checked to see if it satisfies the cross-level boundary constraint. If not, the boundary point's position is adjusted based on the constraint. For example, according to the cross-level boundary constraint, the offset of the boundary feature at level 3 relative to the boundary feature at level 2 cannot exceed 5 meters. If a boundary point in the initial landslide annotation area is found to be offset by more than 5 meters, the boundary point is moved in a direction that satisfies the constraint, generating an adjusted boundary area.

[0206] Step S565: Perform geological continuity verification on the adjusted boundary area to detect the distribution of topographic mutation points and fault zones within the area.

[0207] Verifying geological continuity involves examining the geological conditions within the adjusted boundary area to identify any topographical abrupt changes and fault zones. These abrupt changes and fault zones may affect the stability of the landslide area and therefore require testing.

[0208] Specifically, existing terrain analysis algorithms, such as slope analysis and curvature analysis, can be used to analyze the adjusted boundary area. Slope analysis can identify areas with significant changes in terrain slope, while curvature analysis can detect the degree of terrain curvature. These analyses can be used to identify the locations of topographic abrupt changes and fault zones. For example, if slope analysis reveals a sudden increase in slope in a certain area within the adjusted boundary area, this area can be marked as a topographic abrupt change.

[0209] Step S566: performing region segmentation processing on the adjusted boundary region according to the distribution of the terrain mutation points and the fault zones to generate multiple sub-regions.

[0210] Regional segmentation is the process of dividing the adjusted boundary area into multiple sub-regions based on the distribution of topographic abrupt points and fault zones. This segmentation can more accurately describe the geological structure of the landslide area and provide more detailed information for subsequent processing.

[0211] The specific operation is to split the adjusted boundary area into multiple parts using the terrain mutation points and fault zones as boundaries. For example, if a fault zone is found passing through the adjusted boundary area, the area will be split into two sub-areas along the fault zone.

[0212] Step S567: Perform a secondary matching process of the surface cover features and the texture association features independently on each sub-region, and select the sub-regions with a matching degree higher than a preset consistency sub-threshold as valid landslide sub-regions.

[0213] The secondary matching process is to re-match the surface cover characteristics and texture correlation characteristics of each sub-area to check whether they are consistent with the landslide characteristics. The preset consistency sub-threshold is a pre-set standard used to determine whether a sub-area is a valid landslide sub-area.

[0214] Specifically, the degree of match between the surface cover features and texture-related features of each sub-region and the landslide features can be calculated. The degree of match can be calculated using a feature similarity calculation method, such as cosine similarity calculation based on feature vectors. When the degree of match of a sub-region is higher than a preset consistency sub-threshold, the sub-region is considered to be a valid landslide sub-region. For example, if the preset consistency sub-threshold is 0.6 and the degree of match between the surface cover features and texture-related features of sub-region R and the landslide features is 0.7, then sub-region R is considered to be a valid landslide sub-region.

[0215] Step S568: topologically merge the spatial ranges of the effective landslide sub-regions to eliminate overlaps and gaps, and generate a revised landslide marked region.

[0216] The topological merging process is to merge the spatial extents of the effective landslide sub-areas, eliminate overlaps and gaps, and make the revised landslide marking area more continuous and reasonable in space.

[0217] In specific operations, topological operations, such as the union operation, can be used to merge the spatial extents of valid landslide subregions. Overlapping portions are retained, and gaps are filled. For example, if there are two valid landslide subregions S and T with partially overlapping spatial extents, a union operation can be used to merge them into a single continuous region, eliminating the overlapping portions and generating a revised landslide annotated region.

[0218] Step S570: generating a landslide area recognition result according to the spatial coordinate mapping relationship between the corrected landslide marked area and the target remote sensing image dataset.

[0219] The spatial coordinate mapping relationship refers to the coordinate correspondence between the corrected landslide annotated area and the target remote sensing image dataset. Through this mapping relationship, the corrected landslide annotated area can be accurately annotated on the target remote sensing image dataset to generate the final landslide area identification result.

[0220] Specifically, based on the corrected spatial coordinates of the landslide annotated area, the corresponding location is found on the target remote sensing image dataset and the area is marked. For example, if the corrected coordinates of the landslide annotated area are (x1, y1) to (x2, y2), the corresponding pixel range is found on the target remote sensing image dataset and marked as the landslide area, generating the landslide area recognition result.

[0221] As an implementation manner, after generating the landslide area identification result, the method provided by the embodiment of the present invention may further include the following steps S580 to S5130:

[0222] Step S580: performing feature backtracking verification processing on the target area in the landslide area identification result, and extracting the fusion boundary feature set and the landslide identification feature set corresponding to the target area.

[0223] The feature backtracking verification process rechecks the target area in the landslide area identification results and extracts its corresponding fused boundary feature set and landslide identification feature set. This helps further confirm whether the target area is truly a landslide area and improves the accuracy of the identification results.

[0224] Specifically, for each target area in the landslide area identification results, the corresponding fused boundary feature set and landslide identification feature set are searched in the relevant data recorded in the previous processing process. For example, the fused boundary feature set and landslide identification feature set corresponding to target area A are obtained by querying a database or data file.

[0225] Step S590: performing secondary dynamic matching processing on the fused boundary feature set and the standard landslide morphological feature set in the landslide morphological feature library to generate a regional matching confidence set.

[0226] Secondary dynamic matching involves comparing the fused boundary feature set with the standard landslide morphological feature set again. The matching degree between the fused boundary feature set and the standard landslide morphological feature set is calculated for each target region, thereby generating a regional matching confidence set. The regional matching confidence set reflects the degree of similarity between each target region and the standard landslide morphological feature set.

[0227] In practice, a similar method to the previous feature comparison process can be used to calculate the similarity between the fused boundary feature set and the standard landslide morphological feature set. For example, using the feature vector-based cosine similarity calculation method, for each target region's fused boundary feature, the cosine similarity is calculated between that feature and each feature in the standard landslide morphological feature set. These similarities are combined to obtain the regional match confidence for that target region. For all target regions, the calculated regional match confidences are combined to form a regional match confidence set.

[0228] Step S5100: Filter the target areas in the landslide area identification results based on the area matching confidence set, remove the misjudged areas with confidence lower than the preset backtracking threshold, and generate an optimized landslide area set.

[0229] The screening process involves filtering the target areas in the landslide area identification results based on the region matching confidence set. Target areas with confidence levels below a preset backtracking threshold are considered misidentified and removed, thereby generating an optimized landslide area set. The preset backtracking threshold is a pre-set standard used to determine whether a target area is a true landslide area.

[0230] Specifically, the region matching confidence of each region in the region matching confidence set is compared with a preset backtracking threshold. For example, if the preset backtracking threshold is 0.7 and the region matching confidence of target region B is 0.6, then target region B is removed from the landslide region identification results. The remaining target regions form the optimized landslide region set.

[0231] Step S5110: performing spatial topological relationship analysis on adjacent areas in the optimized landslide area set, and extracting regional boundary overlap and geological continuity parameters.

[0232] Spatial topological relationship analysis optimizes the spatial relationship between adjacent regions within a landslide region set and extracts regional boundary overlap and geological continuity parameters. Regional boundary overlap reflects the overlap between adjacent region boundaries, while geological continuity parameters describe the degree of geological coherence between adjacent regions.

[0233] To perform this analysis, the spatial analysis capabilities of geographic information system (GIS) software can be used to analyze adjacent regions within the optimized landslide area set. For each pair of adjacent regions, the ratio of their boundary overlap area to their total area is calculated to determine the degree of boundary overlap. Simultaneously, geological information such as the geological deformation characteristics and slope correlation characteristics of the adjacent regions is analyzed to calculate the geological continuity parameter. For example, using GIS software to analyze adjacent regions C and D, the calculated boundary overlap is 0.2, and the geological continuity parameter is 0.8.

[0234] Step S5120: performing region merging processing on the optimized landslide region set according to the region boundary overlap and geological continuity parameters to generate a merged landslide region.

[0235] The region merging process is to merge adjacent regions with high overlap and geological continuity based on the regional boundary overlap and geological continuity parameters to generate the merged landslide region. This merging can make the landslide region division more reasonable and reduce unnecessary segmentation.

[0236] Specifically, for each pair of adjacent regions in the optimized landslide region set, the region boundary overlap and geological continuity parameter are compared with a preset merging threshold. For example, the preset merging threshold for region boundary overlap is 0.1, and the merging threshold for geological continuity parameter is 0.7. If the region boundary overlap of adjacent regions E and F is 0.2 and the geological continuity parameter is 0.8, which meets the merging threshold conditions, then regions E and F are merged into a single merged landslide region.

[0237] Step S5130: performing a coverage type cross-validation process based on the landslide identification feature set of the merged landslide area and the surface coverage features of the target remote sensing image dataset, eliminating coverage type conflict areas, and generating a final landslide area identification result.

[0238] The coverage type cross-validation process is to compare the landslide identification feature set of the merged landslide area with the surface coverage features of the target remote sensing image dataset, check whether the coverage type of the merged landslide area is consistent with the landslide surface type, exclude areas with conflicting coverage types, and generate the final landslide area identification result. Specifically, for each merged landslide area, the surface coverage features in its landslide identification feature set are compared with the surface coverage classification layer of the target remote sensing image dataset. If it is found that the coverage type does not match the landslide surface type, the area is excluded. For example, the landslide surface type is usually bare land or sparse vegetation, and the coverage type of a merged landslide area is water body, so the area is excluded. The remaining merged landslide areas constitute the final landslide area identification result.

[0239] In summary, the intelligent landslide identification method proposed in this paper, based on hierarchical frame selection and boundary feature fusion, accurately identifies landslide areas through a series of steps, including hierarchical frame selection processing, boundary feature extraction and fusion, feature comparison, and region labeling on the target remote sensing image dataset. This provides effective technical support for geological disaster prevention and control. Specific technical means and algorithms are employed in each step to ensure the feasibility and accuracy of the method. Furthermore, through multiple verification and revisions, the reliability of the landslide area identification results is further improved.

[0240] It is understandable that the various algorithms involved in the above-mentioned introductions of the embodiments of the present invention, such as the Euclidean distance algorithm, the terrain analysis algorithm, the DBSCAN algorithm, etc., can all be learned from the relevant content in the prior art. In order to save space, they will not be expanded too much in the embodiments of the present invention. In addition, when implementing the scheme of the present invention, those skilled in the art can supplement the details according to the common knowledge in this field. For example, according to the common knowledge in this field, normalization can be used to eliminate dimensional conflicts before feature fusion, interpolation can be used to eliminate dimensional differences, and thresholds can be reasonably set based on historical data, experience or business scenario requirements. The model can be trained based on a general model training method, and the number of layers in the model structure can be set based on actual needs, the activation function can be selected, etc. The present invention will no longer provide redundant introductions to the overly detailed implementation process.

[0241] See also Figure 2 , Figure 2 This is a schematic diagram of the structure of a computer system provided in an embodiment of the present invention. The computer system includes at least a processor 101, a communication interface 102, and a memory 103. The processor 101, communication interface 102, and memory 103 may be connected via a bus or other means. The processor 101 (also known as the Central Processing Unit (CPU)) is the computing and control core of the computer system, capable of parsing various instructions within the computer system and processing various data within the computer system. The communication interface 102 may optionally include a standard wired interface or a wireless interface (such as Wi-Fi, a mobile communication interface, etc.), which can be used to send and receive data under the control of the processor 101. The communication interface 102 may also be used for data transmission and interaction within the computer system. The memory 103 is a storage device in the computer system for storing programs and data. It is understood that the memory 103 herein may include both the built-in memory of the computer system and, of course, the extended memory supported by the computer system. The memory 103 provides storage space, which stores the computer system's operating system, but this is not limited to this in the present invention. In one embodiment, the processor 101 executes the landslide intelligent identification method based on hierarchical frame selection and boundary feature fusion provided in the above embodiment of the present invention by running the computer program in the memory 103 .

Claims

1. A landslide intelligent identification method based on hierarchical frame selection and boundary feature fusion, characterized by: The following steps are involved: Acquire a target remote sensing image dataset, and perform hierarchical frame selection processing on the target remote sensing image dataset to obtain a hierarchical frame selection area set; each hierarchical frame selection area corresponds to a frame selection level identifier; Performing boundary feature extraction processing on the hierarchical frame selection area set to obtain a boundary feature set of each hierarchical frame selection area; the boundary feature set includes spatial distribution features, geometric morphology features, and texture association features; Inputting the boundary feature set into a pre-trained boundary feature fusion model for feature fusion processing to generate a fused boundary feature set; the fused boundary feature set is used to characterize the correlation between boundary features between different frame selection levels; Based on the fused boundary feature set and a preset landslide morphological feature library, feature comparison processing is performed to generate a landslide identification feature set; the landslide identification feature set includes geological deformation features, slope correlation features and surface cover features of the target area; Performing region annotation processing on the target remote sensing image dataset according to the landslide identification feature set to generate a landslide region identification result; The step of inputting the boundary feature set into a pre-trained boundary feature fusion model for feature fusion processing to generate a fused boundary feature set includes: Performing a multi-level feature association analysis on the spatial distribution features in the boundary feature set to generate a first association feature sequence; the first association feature sequence is used to characterize the spatial distribution association strength between different frame selection levels; Performing morphological similarity measurement on the geometric features in the boundary feature set to generate a geometric morphological similarity map; the geometric morphological similarity map is used to indicate the geometric morphological differences of different frame selection levels; Performing cross-feature matching on the first associated feature sequence and the geometric morphology similarity map to generate a dynamic weight adjustment coefficient; the dynamic weight adjustment coefficient is used to balance the influence of spatial distribution and geometric morphology of different frame selection levels on the fusion result; Performing adaptive weighted aggregation processing on the texture-related features in the boundary feature set according to the dynamic weight adjustment coefficient to generate an initial fused texture feature; Performing feature splicing processing on the initial fused texture feature and the first associated feature sequence to generate a cross-level splicing feature; Nonlinear mapping processing is performed on the cross-level splicing features to eliminate feature redundancy and enhance the consistency of feature expression across levels, thereby generating the fused boundary feature set.

2. The method according to claim 1, characterized in that The acquiring of the target remote sensing image dataset and performing hierarchical frame selection processing on the target remote sensing image dataset to obtain a hierarchical frame selection area set includes: Performing multi-scale segmentation processing on the target remote sensing image data set to obtain a plurality of initial segmentation regions; each initial segmentation region corresponds to a segmentation scale identifier; Performing region merging processing on the multiple initial segmented regions according to the segmentation scale identifier to generate a merged region set; each merged region in the merged region set includes at least two initial segmented regions; Performing a regional screening process on the merged region set to obtain a candidate region set; each candidate region in the candidate region set meets a preset landslide candidate condition; the landslide candidate condition includes a region area threshold, a region terrain continuity, and a region texture consistency; According to the distribution density and area overlap of the candidate area set, the candidate area set is hierarchically divided to generate the hierarchical framed area set; each hierarchical framed area corresponds to a framed level identifier, and the framed level identifier is used to indicate the spatial division priority of the hierarchical framed area at different segmentation scales.

3. The method according to claim 2, characterized in that The multi-scale segmentation processing is performed on the target remote sensing image data set to obtain multiple initial segmentation regions, including: Obtaining image resolution and land cover type distribution information of the target remote sensing image dataset; Determining a segmentation scale range according to the image resolution, and selecting a plurality of candidate segmentation scales within the segmentation scale range based on the surface cover type distribution information; Performing image segmentation processing on the target remote sensing image dataset at each candidate segmentation scale to generate an initial segmentation area; the image segmentation processing includes superpixel segmentation processing and edge detection processing; The initial segmentation regions generated under different candidate segmentation scales are subjected to scale alignment processing to obtain the multiple initial segmentation regions; each initial segmentation region corresponds to a segmentation scale identifier and a corresponding segmentation parameter set.

4. The method according to claim 1, wherein The boundary feature fusion model is trained by the following steps: Acquire a historical remote sensing image sample set and a corresponding annotated landslide area set; perform hierarchical frame selection processing on the historical remote sensing image sample set to obtain a sample hierarchical frame selection area set; perform boundary feature extraction processing on the sample hierarchical frame selection area set to obtain a sample boundary feature set; Initializing a feature fusion network structure; the feature fusion network structure includes a multi-level feature association module, a dynamic weight generation module and a nonlinear mapping module; Inputting the sample boundary feature set into the multi-level feature association module in the feature fusion network structure to generate a first associated feature sequence of the sample; Inputting the first associated feature sequence of the sample into a dynamic weight generation module, and generating a sample dynamic weight adjustment coefficient by combining the sample geometric features and texture associated features; Performing weighted aggregation processing on the sample texture-related features based on the sample dynamic weight adjustment coefficient to generate the sample initial fusion texture features; Inputting the sample initial fusion texture features and the sample first correlation feature sequence into a nonlinear mapping module to generate a predicted fusion feature set; Extracting standard spatial distribution features, standard geometric features and standard texture association features from the set of annotated landslide areas, and constructing a standard fusion feature template; Calculating the distribution difference, morphological matching and texture consistency index between the predicted fusion feature set and the standard fusion feature template; Generating a model optimization loss value according to a weighted combination result of the distribution difference, morphological matching and texture consistency indicators; Based on the model optimization loss value, the network parameters of the multi-level feature association module, the dynamic weight generation module and the nonlinear mapping module are jointly iteratively optimized until the feature deviation between the predicted fusion feature set and the standard fusion feature template is less than the preset optimization threshold, thereby obtaining a trained boundary feature fusion model.

5. The method according to claim 1, wherein The step of performing feature comparison processing based on the fused boundary feature set and a preset landslide morphological feature library to generate a landslide identification feature set includes: Obtaining a standard landslide morphological feature set from the landslide morphological feature library; the standard landslide morphological feature set includes geological deformation characteristics, slope correlation characteristics, and surface cover characteristics of a typical landslide area; Performing feature analysis on each fused boundary feature in the fused boundary feature set to extract target geological deformation parameters, target slope parameters, and target surface cover parameters; Calculating similarities between the target geological deformation parameter, target slope parameter, and target surface cover parameter and the corresponding parameters in the standard landslide morphological feature set to generate geological deformation similarity, slope correlation similarity, and surface cover similarity; Determining the landslide identification confidence corresponding to the fused boundary feature according to a weighted summation result of the geological deformation similarity, the slope correlation similarity, and the surface cover similarity; The feature set of the target area corresponding to the fused boundary features whose landslide identification confidence is greater than a preset confidence threshold is determined as the landslide identification feature set.

6. The method according to claim 5, characterized in that The step of determining the feature set of the target area corresponding to the fused boundary features whose landslide identification confidence is greater than a preset confidence threshold as the landslide identification feature set includes: Obtaining the frame selection level identifier and spatial position information corresponding to the fused boundary feature set; Determining the spatial division priority of the target area according to the frame selection level identifier, and performing spatial clustering processing on the target area according to the spatial position information to generate a set of candidate landslide areas; Performing a regional merging verification process on each candidate landslide region in the candidate landslide region set to obtain a verified merged region; the regional merging verification process includes regional overlap verification and geological continuity verification; Performing a secondary comparison process on the feature set corresponding to the verification merged area and the standard landslide morphological feature set to generate a secondary comparison similarity; If the secondary comparison similarity is greater than a preset secondary threshold, the feature set corresponding to the verification merged area is added to the landslide identification feature set.

7. The method according to claim 1, characterized in that The performing region labeling processing on the target remote sensing image dataset according to the landslide identification feature set to generate a landslide region identification result includes: Performing spatial continuity analysis on the geological deformation features in the landslide identification feature set to generate a deformation continuous area distribution map; the deformation continuous area distribution map is used to identify target areas with consistent surface deformation trends and coherent spatial distribution; Performing a superposition analysis on the deformation continuous region distribution map and the slope correlation feature to extract candidate deformation regions where the slope change overlaps with the deformation continuous region; Performing texture pattern matching based on the texture association features of the candidate deformation region to generate a texture consistency region set; the regions in the texture consistency region set satisfy a preset landslide texture evolution law; Performing boundary closure verification processing on the texture consistency region set, screening regions where the density of boundary turning points is lower than a preset closure threshold as closed candidate regions; Performing coverage type matching processing on the spatial range of the closed candidate area and the surface coverage features in the target remote sensing image dataset, excluding interference areas whose coverage types do not match the landslide surface types, and generating an initial landslide annotation area; Performing multi-scale context consistency correction processing on the initial landslide annotation area, fusing the boundary feature correlations of different frame selection levels, and generating a corrected landslide annotation area; The landslide area recognition result is generated according to the spatial coordinate mapping relationship between the corrected landslide marked area and the target remote sensing image data set.

8. The method according to claim 7, characterized in that The multi-scale context consistency correction processing is performed on the initial landslide annotation area, and the boundary feature correlation of different frame selection levels is integrated to generate a corrected landslide annotation area, including: Extracting the frame selection level identification and boundary feature correlation parameters corresponding to the initial landslide marked area; Acquire a set of boundary features of adjacent levels in the set of level selection regions according to the level selection level identifier; Performing context association analysis on the boundary feature sets of the adjacent levels to generate cross-level boundary constraints; the cross-level boundary constraints are used to limit the spatial offset amplitude of boundary features of different levels; Dynamically adjusting the boundary of the initial landslide marked area based on the cross-level boundary constraint condition to generate an adjusted boundary area; Performing geological continuity verification on the adjusted boundary area to detect the distribution of topographic mutation points and fault zones within the area; Performing a region segmentation process on the adjusted boundary region according to the distribution of the terrain mutation points and the fault zones to generate a plurality of sub-regions; Perform a secondary matching process of independent surface cover features and texture correlation features on each sub-region, and select sub-regions with matching degrees higher than the preset consistency sub-threshold as valid landslide sub-regions; The spatial range of the effective landslide sub-area is topologically merged to eliminate overlaps and gaps, thereby generating the corrected landslide marked area.

9. A computer system, characterized in that: include: a memory, wherein the computer program is stored in the memory; A processor is configured to load the computer program to implement the landslide intelligent identification method based on hierarchical frame selection and boundary feature fusion according to any one of claims 1 to 8.

Images