Landslide susceptibility mapping method combined with feature fusion, medium and equipment
By combining features and fusion of Transformer network and PSInSAR method, the surface deformation rate is solved, and the problem of difficult to effectively utilize spatial position information and consider surface deformation rate in the existing landslide prone mapping method is achieved, and a higher landslide prone analysis is achieved. accuracy.
Patent Information
- Application Number
- CN202411468514.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-21
- Publication Date
- 2025-02-18
AI Technical Summary
The existing landslide prone mapping method is difficult to effectively utilize spatial position information and consider the surface deformation rate, resulting in insufficient accuracy of landslide prone analysis.
Using a landslide prone mapping method combining feature fusion, a feature fusion Transformer (FFTR) network is constructed by acquiring multi-source remote sensing data and historical landslide data, combining machine learning classifiers and permanent scatterer synthetic aperture radar interferometry (PSInSAR) method , fuse the surface deformation rate to improve the accuracy of landslide susceptibility analysis.
It improves the accuracy of landslide proneness analysis, can more comprehensively discover areas with landslide tendencies, and effectively supports the prevention and control of landslide geological disasters.
Smart Images
Figure CN119475213A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of digital mapping of geographic information systems, and more specifically to a method, medium and equipment for mapping landslide susceptibility combined with feature fusion. Background Art
[0002] Geological disasters refer to geological actions or phenomena that are caused by natural or human factors and cause loss of human life and property and damage to the environment. They mainly include landslides, collapses, mud-rock flows, etc. Landslides destroy mountain villages, railways, roads, houses, mines, and towns, causing serious casualties and property losses. They may also trigger secondary disasters such as dam breaches and floods. Mud-rock flows are easily formed by heavy rains or floods, causing more serious damage. The causes of landslide geological disasters are complex and highly harmful. It is urgent to analyze their disaster patterns regionally and explore areas with landslide tendencies to prevent and reduce the loss of life and property, which is conducive to disaster prevention and control and to ensure safe development.
[0003] In order to assist relevant departments in carrying out reasonable and effective disaster prevention and mitigation work, landslide geological disaster susceptibility mapping has become an important research method, which can predict the probability of disasters occurring in space at a certain location under the influence of complex factors. According to the development characteristics, disaster mechanisms, and disaster patterns of geological disasters in a certain area, the boundary conditions and slope anomalies that may constitute the structural surface of geological disasters are investigated and analyzed. The susceptibility zoning layer is output to determine the danger of geological disasters in the region and the possible impact range.
[0004] Therefore, more and more studies have begun to carry out landslide susceptibility mapping tasks based on traditional machine learning methods, CNN, RNN, etc. These classification models have higher accuracy than traditional statistical analysis-based methods, and the calculation methods of landslide susceptibility have been developed to a certain extent. Although these methods perform well in landslide susceptibility mapping, there are still some problems. On the one hand, affected by different geographical environments, landslide geological disasters are often the result of the combined action of multiple disaster-causing factors. Landslides have the characteristics of complex causes, and it is necessary to comprehensively consider different disaster-causing factors for landslide susceptibility mapping. Then how to pay attention to and capture spatial information globally and effectively use the contextual information of spatial location to obtain better accuracy is a problem that still needs to be solved. On the other hand, the surface deformation rate is an important indicator for monitoring landslide geological disasters. It can discover potential areas with landslide tendency. Therefore, the surface deformation rate needs to be considered in landslide susceptibility mapping. Therefore, how to effectively use spatial location information, consider the surface deformation rate and improve the accuracy of landslide susceptibility mapping is an urgent problem to be solved. Summary of the invention
[0005] The object of the present invention is to provide a landslide susceptibility mapping method, medium and equipment combined with feature fusion, which can improve the accuracy of landslide susceptibility analysis.
[0006] The present invention provides a method for mapping landslide susceptibility combined with feature fusion, comprising the following steps: S1: acquiring multi-source remote sensing data and historical landslide data of a study area, and obtaining a hazard factor data set and a landslide sample data set according to the multi-source remote sensing and historical landslide data; S2: obtaining a preliminary landslide susceptibility map according to the hazard factor data set and the landslide sample data set by using a feature fusion network and a machine learning classifier; S3: obtaining a surface deformation rate according to SAR data of the study area by using a PSInSAR method; S4: obtaining a final landslide susceptibility map according to the preliminary landslide susceptibility map and the surface deformation rate.
[0007] Furthermore, the multi-source remote sensing data of the above-mentioned landslide susceptibility mapping method combined with feature fusion include multispectral remote sensing image data and multi-temporal SAR data, and the disaster factor data set includes lithology, elevation, curvature, plane curvature, profile curvature, slope, slope aspect, surface roughness, terrain moisture index, normalized vegetation index, precipitation, distance from faults, distance from rivers, distance from roads and surface cover data.
[0008] Furthermore, the feature fusion network of the above-mentioned landslide susceptibility mapping method combined with feature fusion includes a first feature fusion module, a second feature fusion module and a third feature fusion module; the first feature fusion module is used for discrete feature fusion to obtain discrete data features; the second feature fusion module is used for continuous feature fusion to obtain continuous data features; the third feature fusion module is used to combine discrete data features and continuous data features to obtain high-level spatial features; the first feature fusion module includes an Embedding layer, a Stack layer, a Transformer Block layer and a Flatten layer connected in sequence; the second feature fusion module includes an Expand Dimentions layer, a Concatenate layer and a LayerNormalization layer connected in sequence; the third feature fusion module includes a Concatenate layer, an MLP layer and a Dense layer connected in sequence; the output end of the Flatten layer and the output end of the Layer Normalization layer are connected to the input end of the third feature fusion module including the Concatenate layer connected in sequence.
[0009] Furthermore, step S2 of the above-mentioned landslide susceptibility mapping method combined with feature fusion specifically includes: S21: dividing the hazard factor dataset and the landslide sample dataset according to a preset ratio to obtain a training set and a test set; the training set and the test set both include discrete data and continuous data; S22: based on the discrete data, using the first feature fusion module to obtain discrete data features; based on the continuous data, using the second feature fusion module to obtain continuous data features; based on the discrete data features and the continuous data features, using the third feature fusion module to obtain high-level spatial features; S23: based on the high-level spatial features and the hazard factor dataset, obtaining disaster spatial features; based on the disaster spatial features, using a machine learning classifier to divide all data into landslides and non-landslides to obtain landslide susceptibility data; S24: based on the landslide susceptibility data, obtaining a preliminary landslide susceptibility map.
[0010] Furthermore, the discrete data of the above-mentioned landslide susceptibility mapping method combined with feature fusion include surface cover type, lithology and slope aspect; the continuous data include elevation, curvature, plan curvature, profile curvature, slope, surface roughness, terrain moisture index, normalized difference vegetation index, distance from faults, distance from rivers, distance from roads and precipitation.
[0011] Furthermore, step S23 of the above-mentioned landslide susceptibility mapping method combined with feature fusion also includes: performing accuracy evaluation on the machine learning classifier according to the landslide susceptibility data.
[0012] Furthermore, step S3 of the landslide susceptibility mapping method combined with feature fusion specifically includes: obtaining the surface deformation rate according to the SAR data of the study area by using the PSInSAR method, such as the formula: , in is the surface deformation rate, is the ground deformation rate along the line of sight, It is the angle between the satellite's line of sight and the vertical ground direction.
[0013] Furthermore, step S4 of the landslide susceptibility mapping method combined with feature fusion specifically includes: obtaining a final landslide susceptibility map according to the preliminary landslide susceptibility map and the surface deformation rate, such as the formula: , in, is the surface deformation rate, Moderate deformation. For a higher degree of deformation, For extremely high degree of deformation, is the mean, is the standard deviation.
[0014] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the landslide susceptibility mapping method combined with feature fusion are implemented.
[0015] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps of the above-mentioned landslide susceptibility mapping method combined with feature fusion are implemented.
[0016] The landslide susceptibility mapping method, medium and device combined with feature fusion provided by the present invention have the following beneficial effects: The present invention prepares a landslide hazard factor dataset and a landslide sample dataset based on multi-source remote sensing data and landslide data, constructs a feature fusion Transformer (FFTR) network, and obtains global high-order features; specifically, the FFTR network can establish global long-distance dependencies and high-order spatial interactions for input hazard factors through a self-attention mechanism; the multi-head attention mechanism can capture richer feature information in landslide data and mine the relationship between global features; and it has a strong feature aggregation capability, which can establish long-distance feature links for landslide features and better understand global context information; by using Transformer to perform feature fusion work for hazard factors, the spatial characteristics of different hazard factors can be better and more comprehensively understood, so as to better analyze the susceptibility of landslides in the region; The present invention uses machine learning classifiers to further perform classification tasks on the high-order features obtained by the FFTR network, including random forest RF, XGBoost, CatBoost, and GBDT classifiers, among which RF is integrated based on decision trees, and has a better processing effect on missing values in disaster factors, and can evaluate the importance of each feature in the entire classification process; the ET algorithm classifies through some decorrelated decision trees, reduces the variance in the landslide classification process, and will not be seriously affected by certain disaster factors; GBDT uses CART regression trees as decision trees, and can automatically discover high-order relationships between disaster factor features; XGBoost can sort and store training data in advance, and use iteratively in subsequent calculations, thereby effectively improving the efficiency of landslide classification; CatBoost is calculated based on symmetric decision trees, and is efficient and reasonable in processing categorical features in disaster factors; at the same time, the model accuracy is evaluated, and a preliminary landslide susceptibility map is obtained; The present invention uses a method of Permanent Scatterers Interferometric Synthetic Aperture Radar (PSInSAR) based on multi-phase SAR data to calculate the distribution of surface deformation rate values in the study area. The method is based on multi-view SAR data of the same area, and can detect PS points that are not affected by the decorrelation of the temporal and spatial baselines. The time coverage of its deformation monitoring is large, and most of the atmospheric images can be removed. At the same time, the slope deformation of landslide geological disasters can be monitored, and the monitoring can reach the millimeter level. The surface deformation rate is integrated into the landslide susceptibility map to obtain the final landslide susceptibility mapping result, which can provide a reference for the surface deformation rate for landslide susceptibility mapping, making the landslide susceptibility analysis more accurate and comprehensive. The present invention can take into account the application of deep learning technology to landslide susceptibility mapping, comprehensively consider the cumulative deformation of the landslide slope, and more comprehensively discover areas with landslide tendency, effectively improving the accuracy of landslide susceptibility analysis and providing assistance for landslide geological disaster prevention and control work in the region. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] The present invention will be further described below with reference to the accompanying drawings and embodiments, in which: Figure 1 It is a flow chart of the landslide susceptibility mapping method combined with feature fusion provided by the present invention; Figure 2 It is a technical flow chart of the landslide susceptibility mapping method combined with feature fusion provided by the present invention; Figure 3 It is a FFTR network structure diagram of the landslide susceptibility mapping method combined with feature fusion provided by the present invention; Figure 4 It is a flow chart of FFTR network feature fusion and machine learning classifier of the landslide susceptibility mapping method combined with feature fusion provided by the present invention; Figure 5 It is a landslide susceptibility result map of the landslide susceptibility mapping method combined with feature fusion provided by the present invention; Figure 6 It is a surface deformation rate result map of the landslide susceptibility mapping method combined with feature fusion provided by the present invention; Figure 7 It is a landslide susceptibility result map fused with surface deformation rate of the landslide susceptibility mapping method combined with feature fusion provided by the present invention; Figure 8 It is a structural block diagram of the computer device provided by the present invention. DETAILED DESCRIPTION
[0018] In order to have a clearer understanding of the technical features, purposes and effects of the present invention, specific embodiments of the present invention are now described in detail with reference to the accompanying drawings.
[0019] The present invention discloses a landslide susceptibility mapping method combined with feature fusion, which combines Feature Fusion Transformer (FFTR) and Permanent Scatterers Interferometric Synthetic Aperture Radar (PSInSAR); Figure 1 A schematic diagram of a landslide susceptibility mapping method combined with feature fusion in this embodiment is shown. In this embodiment, the landslide susceptibility mapping method combined with feature fusion includes the following steps: S1: Obtain multi-source remote sensing data and historical landslide data, and obtain the hazard factor dataset and landslide sample dataset based on the multi-source remote sensing and historical landslide data; Specifically, the multi-source remote sensing data include multi-spectral remote sensing image data and multi-temporal SAR data, and the disaster factor data set includes lithology, elevation, curvature, plane curvature, profile curvature, slope, aspect, surface roughness, terrain wetness index, normalized difference vegetation index, precipitation, distance from faults, distance from rivers, distance from roads and surface cover data; It should be noted that the landslide sample dataset is derived from the historical landslide data accumulated in the study area, as well as the landslide data and non-landslide data annotated by Google Earth based on multispectral remote sensing data, ensuring that the ratio of landslide samples to non-landslide samples is balanced and close to 1:1; It should be noted that land cover data include land cover types; S2: Based on the hazard factor dataset and landslide sample dataset, a preliminary landslide susceptibility map is obtained using feature fusion network and machine learning classifier; Specifically, the feature fusion network includes a first feature fusion module, a second feature fusion module and a third feature fusion module; the first feature fusion module is used for discrete feature fusion to obtain discrete data features; the second feature fusion module is used for continuous feature fusion to obtain continuous data features; the third feature fusion module is used to combine discrete data features and continuous data features to obtain high-level spatial features; Specifically, the first feature fusion module includes an Embedding layer, a Stack layer, a TransformerBlock layer, and a Flatten layer connected in sequence; the second feature fusion module includes an Expand Dimensions layer, a Concatenate layer, and a Layer Normalization layer connected in sequence; the third feature fusion module includes a Concatenate layer, an MLP layer, and a Dense layer connected in sequence; the output end of the Flatten layer and the output end of the Layer Normalization layer are connected to the input end of the Concatenate layer connected in sequence in the third feature fusion module; Specifically, the machine learning classifier is random forest RF; Specifically, step S2 includes: S21: dividing the hazard factor dataset and the landslide sample dataset according to a preset ratio to obtain a training set and a test set; both the training set and the test set include discrete data and continuous data; Specifically, the preset ratio is 7:3; Specifically, discrete data include land cover type, lithology and slope aspect; continuous data include elevation, curvature, plan curvature, profile curvature, slope, surface roughness, terrain wetness index, normalized difference vegetation index, distance to faults, distance to rivers, distance to roads and precipitation; S22: according to the discrete data, using the first feature fusion module, obtain discrete data features; according to the continuous data, using the second feature fusion module, obtain continuous data features; according to the discrete data features and the continuous data features, using the third feature fusion module, obtain high-level spatial features; It should be noted that in step S22, each type of discrete data is processed by the Embedding layer, the Stack layer, the Transformer Block and the Flatten layer, and is initially fused to obtain discrete data features; each type of continuous data is expanded in dimension, and then is input into the Layer Normalization layer for regularization operation after the dimension is combined, and is initially fused to obtain continuous data features; the discrete data features and the continuous data features are input into the Concatenate layer for dimension combination, the MLP layer, and the fully connected layer Dense to obtain high-level spatial features, that is, global spatial features; S23: Based on the high-level spatial features and the hazard factor dataset, the disaster spatial features are obtained; based on the disaster spatial features, all data are classified into landslide and non-landslide using a machine learning classifier to obtain landslide susceptibility data; It should be noted that the FFTR network feature fusion and machine learning classifier process is as follows Figure 4 As shown, the fused spatial features output by FFTR are combined with the hazard factors to obtain the disaster spatial features, that is, the spatial features of landslides; all data are divided into landslides and non-landslides using a machine learning classifier; It should be noted that the machine learning classifier can be any of the random forest RF, XGboost, GBDT, CatBoost and ET classifiers; Specifically, the feature selection methods of ET and RF classifiers are both Gini coefficients: , is the probability that a landslide sample belongs to the K category, K is the landslide class or non-landslide class; Specifically, the loss function of the GBDT classifier is the logarithmic loss function: , is the probability that a landslide sample belongs to the K category, K is the landslide class or non-landslide class; It should be noted that, in step S23, the disaster spatial characteristics are input into the classifier and evaluated, and finally the landslide susceptibility data in the study area is calculated; Optionally, step S23 further includes: performing accuracy evaluation on the machine learning classifier according to the landslide susceptibility data; It should be noted that in order to comprehensively measure the reliability of each method in landslide susceptibility mapping so as to select the optimal method for accurate mapping, this embodiment uses five evaluation methods to evaluate each model, namely Accuracy, Precision, Recall, F1 score and AUC value; S24: obtaining a preliminary landslide susceptibility map based on the landslide susceptibility data; It should be noted that the probability of landslide occurrence is finally output to obtain a preliminary landslide susceptibility map; S3: Based on the SAR data of the study area, the surface deformation rate is obtained using the PSInSAR method; Specifically, step S3 includes: according to the SAR data of the study area, using the PSInSAR method, obtaining the surface deformation rate, such as the formula: , in is the surface deformation rate, is the ground deformation rate along the line of sight, It is the angle between the satellite line of sight and the vertical ground direction; It should be noted that the PSInSAR algorithm is based on multi-phase SAR data of the same area, based on the echo signals of permanent scatterers, and uses long-term observation sequences to measure millimeter-level ground deformation, which is not affected by spatiotemporal coherence; S4: Obtain the final landslide susceptibility map based on the preliminary landslide susceptibility map and the surface deformation rate; Specifically, step S4 includes: obtaining a final landslide susceptibility map according to the preliminary landslide susceptibility map and the surface deformation rate, such as the formula: , in, is the surface deformation rate, Moderate deformation. For a higher degree of deformation, For extremely high degree of deformation, is the mean, is the standard deviation; It should be noted that the three degrees of deformation are respectively integrated with the medium susceptibility, high susceptibility and extremely high susceptibility in the landslide susceptibility level.
[0020] In some embodiments, the above-mentioned landslide susceptibility mapping method combined with feature fusion can also be implemented in the following manner.
[0021] This embodiment aims to solve the technical problems of how to effectively use spatial location information and how to consider the surface deformation rate. Please refer to Figure 2 In this embodiment, the landslide susceptibility mapping method combined with feature fusion includes the following steps: Step 1: Obtain multi-source remote sensing data and historical landslide data to create a hazard factor dataset and a landslide sample dataset; Specifically, multi-source remote sensing data include but are not limited to multi-spectral data and multi-temporal SAR data; disaster factor data include but are not limited to the following according to the selected study area: lithology, elevation, curvature, plane curvature, profile curvature, slope, aspect, surface roughness, terrain moisture index, normalized difference vegetation index, precipitation, distance from faults, distance from rivers, distance from roads and surface cover data; landslide sample data set comes from the accumulated landslide historical data in the study area, as well as the landslide data and non-landslide data marked by Google Earth in combination with multi-spectral remote sensing data, to ensure that the ratio of landslide samples to non-landslide samples is balanced and close to 1:1; Step 2: Divide all data sets into training sets and test sets, input them into the FFTR network and machine learning classifier, evaluate the model accuracy, and obtain a preliminary landslide susceptibility map; Specifically, all landslide sample sets are divided into training set and test set in a ratio of 7:3; the FFTR network structure is as follows: Figure 3 As shown in the figure, all data are divided into discrete data, including lithology, aspect and surface cover data, and continuous data, including elevation, curvature, plane curvature, profile curvature, slope, surface roughness, terrain moisture index, normalized vegetation index, precipitation, distance from faults, distance from rivers and distance from roads; discrete feature fusion first inputs the data into their respective Embedding layers, and then inputs various discrete features into the Stack layer, and then into TransformerBlock for the next step of feature fusion; for the fusion of continuous features, the continuous data is expanded in dimension and then input into the Concatenate layer, and then the dimension is combined and then input into the Layer Normalization for regularization operation; the output of the discrete feature fusion and the output of the continuous feature fusion are input into the Concatenate layer, and the dimension is combined and then input into the MLP, until the last Dense fully connected layer, and the fused high-level spatial features are output; Specifically, the FFTR network feature fusion and machine learning classifier process is as follows: Figure 4 As shown in the figure, the spatial characteristics of the landslide are obtained by combining the fused spatial features output by FFTR with the hazard factors. All data are divided into landslide and non-landslide using random forest (RF), ET, XGBoost, GBDT and CatBoost classifiers. Finally, the possibility of landslide occurrence is output to obtain a preliminary landslide susceptibility map. The feature selection method of ET and RF classifiers is the Gini coefficient: , is the probability that a landslide sample belongs to the K category, K is the landslide class or non-landslide class; The loss function of the GBDT classifier is the logarithmic loss function: , is the probability that a landslide sample belongs to the K category, K is the landslide class or non-landslide class; In this example, in order to test the performance of the proposed landslide susceptibility mapping model, the surrounding area of the Karakoram Highway is selected as the object to illustrate the model performance and mapping result performance; in order to comprehensively measure the reliability of each method in landslide susceptibility mapping so as to select the optimal method for accurate mapping, this embodiment adopts five evaluation methods to evaluate each model, namely Accuracy, Precision, Recall, F1 score and AUC value; first, the positive sample (Positive, P) is recorded as 1, and the negative sample (Negative, N) is recorded as 0; TP (True Positive) means that the predicted value and the actual value are the same, both are 1, that is, the classification is correct; FP (False Positive) means that the predicted value is 1 and the actual value is 0, that is, the prediction is wrong; FN (False Negative) means that the predicted value is 0 and the actual value is 1, that is, the prediction is wrong; TN (True Negative) means that the predicted value and the actual value are both 0, that is, the prediction is correct; The probability of correct classification is: , The probability of correct classification among all samples classified as positive is: , is the probability of correct classification among all positive samples; , Score: Consider both and , let both reach the highest level at the same time and find a balance; , The AUC value represents the distinguishability of all samples. The stronger the distinguishability, the better the accuracy. is the horizontal axis, The ROC curve is drawn as the ordinate, and the area enclosed by the curve is AUC; this value will not be affected by changes in the ratio of positive and negative samples; , , The performance comparison of the method and model of this embodiment is shown in Table 1, and the ablation experiment results are shown in Table 2. The spatial features fused by FFTR are input into the machine learning classifier, and the best performance is the combination of FFTR and RF; compared with using FFTR alone, the AUC is improved by 2.23%; compared with using RF alone, the AUC is improved by 4.82%; the evaluation indicators of the other models are also improved; this shows that after adding FFTR for feature fusion, the performance of the model has been further improved; Table 1 Comparison of performance of the models of the method of the present invention
[0022] Table 2 Ablation experiment results
[0023] The preliminary mapping results of this example using FFTR and machine learning classifiers are as follows Figure 5 As shown, all landslide susceptibility is classified into very low (0 to 0.2), low (0.2 to 0.4), medium (0.4 to 0.6), high (0.6 to 0.8) and very high (0.8 to 1); Step 3: Use the PSInSAR method to calculate the surface deformation rate in the study area; Specifically, this embodiment takes the vicinity of the Karakoram Highway as the object, uses the PSInSAR algorithm, and uses a long-term observation sequence based on multi-phase SAR data and permanent scatterer echo signals of the same area to measure millimeter-level ground deformation without being affected by spatiotemporal coherence; the SAR data used are shown in Table 3, with a total of 100 scenes of SAR data, including 50 scenes of ascending orbit data and 50 scenes of descending orbit data; Table 3 SAR data
[0024] The calculation method of the surface deformation rate of the landslide slope is: , in is the surface deformation rate, is the ground deformation rate along the line of sight, is the angle between the satellite line of sight and the vertical ground direction; the result of PSInSAR in the method of this embodiment is as follows Figure 6 As shown; Step 4: Integrate the surface deformation rate into the landslide susceptibility map calculated in S2 to obtain the final landslide susceptibility mapping result; Specifically, the surface deformation rate is integrated into the landslide susceptibility map calculated in S2: , in is the surface deformation rate, which is divided into medium deformation , a higher degree of deformation and extremely high deformation , is the mean, is the standard deviation; the three degrees of deformation are respectively integrated with the medium susceptibility, high susceptibility and extremely high susceptibility in the landslide susceptibility level; In the method of this embodiment, the surface deformation rate is integrated into the landslide susceptibility map calculated in S2, such as Figure 7 As shown in the figure, it can be found that after adding the surface deformation rate of the landslide slope, some high-susceptibility areas missed by the classification model are supplemented, making the landslide susceptibility mapping more comprehensive and accurate.
[0025] This embodiment provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the landslide susceptibility mapping method combined with feature fusion are implemented. The storage medium may be a magnetic disk, an optical disk, a read-only memory (ROM), a random access memory (RAM), a flash memory, a hard disk drive (HDD) or a solid-state drive (SSD), etc.; the storage medium may also include a combination of the above-mentioned types of memory.
[0026] This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps of the above-mentioned landslide susceptibility mapping method combined with feature fusion are implemented.
[0027] like Figure 8As shown, the computer device may include: at least one processor 121, such as a CPU (Central Processing Unit), at least one communication interface 123, a memory 124, and at least one communication bus 122. Among them, the communication bus 122 is used to realize the connection and communication between these components. Among them, the communication interface 123 may include a display screen (Display), a keyboard (Keyboard), and the optional communication interface 123 may also include a standard wired interface and a wireless interface. The memory 124 may be a high-speed RAM memory (Random Access Memory, volatile random access memory), or a non-volatile memory (non-volatile memory), such as at least one disk memory. The memory 124 may also be at least one storage device located away from the aforementioned processor 121. Among them, the memory 124 stores an application program, and the processor 121 calls the program code stored in the memory 124 to perform any of the above method steps. Among them, the communication bus 122 may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The communication bus 122 can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 8Only one line is used to represent it, but it does not mean that there is only one bus or one type of bus. Among them, the memory 124 may include a volatile memory (English: volatile memory), such as a random access memory (English: random-access memory, abbreviated: RAM); the memory may also include a non-volatile memory (English: non-volatile memory), such as a flash memory (English: flash memory), a hard disk drive (English: hard disk drive, abbreviated: HDD) or a solid-state drive (English: solid-state drive, abbreviated: SSD); the memory 124 may also include a combination of the above-mentioned types of memory. Among them, the processor 121 may be a central processing unit (English: central processing unit, abbreviated: CPU), a network processor (English: network processor, abbreviated: NP) or a combination of CPU and NP. Among them, the processor 121 may further include a hardware chip. The above-mentioned hardware chip may be an application-specific integrated circuit (English: application-specific integrated circuit, abbreviated: ASIC), a programmable logic device (English: programmable logic device, abbreviated: PLD) or a combination thereof. The PLD may be a complex programmable logic device (CPLD), a field-programmable gate array (FPGA), a generic array logic (GAL), or any combination thereof. Optionally, the memory 124 is also used to store program instructions. The processor 121 may call the program instructions to implement the landslide susceptibility mapping method combined with feature fusion as in this embodiment.
[0028] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation modes, which are merely illustrative rather than restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the present invention and the claims, all of which are within the protection of the present invention.
Claims
1. A landslide susceptibility mapping method combined with feature fusion, characterized in that: The following steps are involved: S1: Acquire multi-source remote sensing data and historical landslide data of the study area, and obtain a disaster factor dataset and a landslide sample dataset based on the multi-source remote sensing and historical landslide data; S2: Based on the hazard factor dataset and the landslide sample dataset, a preliminary landslide susceptibility map is obtained using a feature fusion network and a machine learning classifier; S3: Based on the SAR data of the study area, the surface deformation rate is obtained using the PSInSAR method; S4: Obtaining a final landslide susceptibility map according to the preliminary landslide susceptibility map and the surface deformation rate.
2. The method for mapping landslide susceptibility combined with feature fusion according to claim 1, characterized in that: The multi-source remote sensing data includes multi-spectral remote sensing image data and multi-phase SAR data, and the disaster factor data set includes lithology, elevation, curvature, plane curvature, profile curvature, slope, aspect, surface roughness, terrain moisture index, normalized vegetation index, precipitation, distance from faults, distance from rivers, distance from roads and surface cover data.
3. The method for mapping landslide susceptibility combined with feature fusion according to claim 1, characterized in that: The feature fusion network includes a first feature fusion module, a second feature fusion module and a third feature fusion module; the first feature fusion module is used for discrete feature fusion to obtain discrete data features; the second feature fusion module is used for continuous feature fusion to obtain continuous data features; the third feature fusion module is used to combine the discrete data features and the continuous data features to obtain high-level spatial features; the first feature fusion module includes an Embedding layer, a Stack layer, a Transformer Block layer and a Flatten layer connected in sequence; the second feature fusion module includes an Expand Dimentions layer, a Concatenate layer and a Layer Normalization layer connected in sequence; the third feature fusion module includes a Concatenate layer, an MLP layer and a Dense layer connected in sequence; the output end of the Flatten layer and the output end of the Layer Normalization layer are connected to the input end of the Concatenate layer connected in sequence in the third feature fusion module.
4. The method for mapping landslide susceptibility combined with feature fusion according to claim 3, characterized in that: Step S2 specifically includes: S21: dividing the hazard factor dataset and the landslide sample dataset according to a preset ratio to obtain a training set and a test set; the training set and the test set both include discrete data and continuous data; S22: according to the discrete data, using the first feature fusion module to obtain discrete data features; according to the continuous data, using the second feature fusion module to obtain continuous data features; according to the discrete data features and the continuous data features, using the third feature fusion module to obtain high-level spatial features; S23: obtaining disaster spatial features according to the high-level spatial features and the disaster factor data set; and using a machine learning classifier to classify all data into landslide and non-landslide according to the disaster spatial features to obtain landslide susceptibility data; S24: Obtaining a preliminary landslide susceptibility map based on the landslide susceptibility data.
5. The method for mapping landslide susceptibility combined with feature fusion according to claim 4, characterized in that: The discrete data include land cover type, lithology and slope aspect; the continuous data include elevation, curvature, plan curvature, profile curvature, slope, surface roughness, terrain wetness index, normalized difference vegetation index, distance from faults, distance from rivers, distance from roads and precipitation.
6. The method for mapping landslide susceptibility combined with feature fusion according to claim 4, characterized in that: Step S23 also includes: performing accuracy evaluation on the machine learning classifier according to the landslide susceptibility data.
7. The method for mapping landslide susceptibility combined with feature fusion according to claim 1, characterized in that: Step S3 specifically includes: according to the SAR data of the study area, using the PSInSAR method, obtaining the surface deformation rate, such as the formula: ， in is the surface deformation rate, is the ground deformation rate along the line of sight, It is the angle between the satellite's line of sight and the vertical ground direction.
8. The method for mapping landslide susceptibility combined with feature fusion according to claim 1, characterized in that: Step S4 specifically includes: obtaining a final landslide susceptibility map according to the preliminary landslide susceptibility map and the surface deformation rate, such as the formula: ， in, is the surface deformation rate, Moderate deformation. For a higher degree of deformation, For extremely high degree of deformation, is the mean, is the standard deviation.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the landslide susceptibility mapping method combined with feature fusion as described in any one of claims 1 to 8 are implemented.
10. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the landslide susceptibility mapping method combined with feature fusion as described in any one of claims 1 to 8 are implemented.