A landslide susceptibility evaluation method based on federated learning and UNet

Abstract

The present application relates to the technical field of geological disaster risk analysis, and specifically relates to a landslide susceptibility evaluation method based on federated learning and UNet, comprising the following steps: S1, obtaining landslide influence factor data: collecting multiple landslide influence factor data; S2, establishing a federated learning framework: establishing a federated learning framework; S3, designing a UNet deep learning model: feature extraction and segmentation; S4, federated learning model training and parameter integration: updating the global UNet-based deep learning model; S5, federated learning iteration and model optimization: optimizing the UNet-based deep learning model parameters; S6, generating a landslide susceptibility feature map and probability regression: outputting landslide susceptibility evaluation results; S7, result verification and disaster prevention and mitigation suggestions: providing landslide risk early warning and disaster prevention and mitigation suggestions; the present application can realize more efficient and more accurate landslide susceptibility evaluation, and can update the model in time to cope with new data changes.

Description

Technical Field

[0001] This invention relates to the field of geological hazard risk analysis technology, and in particular to a landslide susceptibility assessment method based on federated learning and UNet. Background Technology

[0002] Landslide susceptibility assessment is a crucial step in geological disaster prevention and control. Traditional methods typically rely on field investigations and experience-based judgments, which are time-consuming and costly. With the development of computer technology and data science, data-driven methods have gradually become an important research direction in landslide susceptibility assessment. These methods predict the probability of future landslides by analyzing geological environmental factors, historical landslide data, and other factors, and provide decision support.

[0003] Traditional landslide susceptibility assessment methods, such as experimental models, analytic hierarchy process (AHP) and statistical models, can effectively predict landslide risk, but they have limitations in handling complex data and fusing multi-source information. In particular, when analyzing image data, traditional convolutional neural network (CNN) methods cannot perform pixel-level classification, resulting in the loss of detailed information and affecting the accurate identification of landslide areas. In addition, CNNs lack skip connections, cannot effectively recover high-resolution details of images, and perform poorly with small amounts of data, resulting in low efficiency.

[0004] To address the limitations of existing methods, this invention aims to propose a landslide susceptibility assessment method that combines federated learning and deep learning technologies. This method can effectively solve the problems of slow data processing speed, insufficient memory, and poor security in centralized training, and achieve a more efficient and secure landslide susceptibility assessment. Summary of the Invention

[0005] This invention provides a landslide susceptibility assessment method based on federated learning and UNet.

[0006] A landslide susceptibility assessment method based on federated learning and UNet includes the following steps:

[0007] S1, Obtain landslide impact factor data: Collect multiple landslide impact factor data and extract multiple impact factor layer data for landslide-prone areas;

[0008] S2, Establish a federated learning framework: Establish a federated learning framework among multiple data holders;

[0009] S3, Design UNet Deep Learning Model: Design a UNet-based deep learning model for landslide susceptibility assessment, and input the data from the influencing factor layer into the model for feature extraction and segmentation;

[0010] S4, Federated Learning Model Training and Parameter Integration: Under the federated learning framework, a UNet-based deep learning model is trained using a landslide sample dataset, and the model parameters are integrated using a federated averaging algorithm to update the global UNet-based deep learning model.

[0011] S5, Federated Learning Iteration and Model Optimization: Optimize the parameters of the UNet-based deep learning model through multiple rounds of federated learning iterations;

[0012] S6, Generate landslide susceptibility feature map and probabilistic regression: Apply the optimized UNet-based deep learning model to generate feature maps of landslide-prone areas, use the Sigmoid layer for probabilistic regression processing, and output the landslide susceptibility evaluation results.

[0013] S7, Results Verification and Disaster Prevention and Mitigation Recommendations: Verify and evaluate the landslide susceptibility assessment results, and provide landslide risk warnings and disaster prevention and mitigation recommendations based on the assessment results.

[0014] Optionally, the multiple landslide influencing factor data in S1 include topographic factors, geological factors, hydrological factors, land cover factors, and climatic factors, wherein;

[0015] The topographical factors include aspect, slope, elevation, and curvature;

[0016] The geological factors include lithology;

[0017] The hydrological factors include average annual precipitation and distance from the river system;

[0018] The land cover factors include enhanced vegetation index, normalized difference vegetation index, surface reflectance, and land use;

[0019] The climate factors include annual average temperature and surface temperature.

[0020] Optionally, the multiple influencing factor layer data in S1 include elevation, slope, aspect, curvature, lithology, distance from water system, and land use.

[0021] Optionally, the construction of the landslide sample dataset includes:

[0022] Data collection: Acquire historical landslide data and data on potential landslide sites;

[0023] Constructing a landslide sample dataset: Based on historical landslide data and landslide hazard point data, a landslide sample dataset is constructed. By judging the number of landslide grids in a rectangular area, if the number of landslide grids in the rectangular area exceeds 50%, the rectangular area is marked as a landslide point; if it is less than 50%, it is marked as a non-landslide point. The marked landslide sample dataset is defined as the experimental dataset.

[0024] Divide the dataset into training and validation sets: Randomly divide the experimental dataset into training and validation sets in an 8:2 ratio, with landslide samples labeled as 1 and non-landslide samples labeled as 0.

[0025] Data augmentation: Data augmentation is performed on the constructed landslide sample data.

[0026] Optionally, the data enhancement includes:

[0027] Identify landslide feature grids: Identify the grids for landslide points in the landslide sample dataset;

[0028] Applying Mixup data augmentation technology: Mixup data augmentation technology is applied to randomly stitch together the grids of the identified landslide points to generate augmented samples, represented as follows:

[0029] ;

[0030] in, For the generated enhanced samples, It is in the interval Weighting coefficients selected randomly within the range, and For landslide samples;

[0031] Add augmented samples to the experimental dataset: Add the generated augmented samples to the experimental dataset.

[0032] Optionally, the parameter update of the global UNet-based deep learning model is represented as:

[0033] ;

[0034] in, Indicates the first The amount of data held by each data holder This represents the total amount of all data. For learning rate, Indicates the first UNet-based deep learning models in federated learning rounds This refers to the UNet-based deep learning model from the previous round of federated learning. This represents the global gradient function.

[0035] Optionally, the training of the UNet-based deep learning model includes:

[0036] Setting the activation function: LeakyReLU is selected as the activation function for each convolutional layer in the UNet-based deep learning model, expressed as:

[0037] ;

[0038] in, It is 0.01;

[0039] Model training: Applying weight-based learning in the convolutional layers, represented as:

[0040] ;

[0041] in, A convolution kernel with a stride of 2. For bias terms, For activation functions;

[0042] Normalization: Multiple landslide influencing factor data are used as input data, and a normalization layer is introduced to normalize the input data. Normalization is performed, and the normalized values ​​are adjusted using scaling and translation parameters, as follows:

[0043] ;

[0044] in, and These are the mean and variance of the current batch, respectively. It is a constant;

[0045] ;

[0046] in, and These are learnable parameters;

[0047] Dropout regularization: Uses the Dropout technique to prevent overfitting, expressed as:

[0048] ;

[0049] Reflection padding: Reflection padding is used to handle convolution boundaries, represented as:

[0050] ;

[0051] in, For the input feature map, This is the feature map after reflection filling. It is the index calculation function for reflection fill;

[0052] Model validation: The binary cross-entropy loss function is used to measure the difference between the model's predicted probability distribution and the true distribution, expressed as:

[0053] ;

[0054] in, It is the number of landslide samples. It is the first The true label (0 or 1) for each landslide sample. It is the model's prediction of the first The probability value of each sample.

[0055] Optionally, the generation of the landslide susceptibility feature map and probability regression in S6 includes:

[0056] S61, Optimization and Application of UNet Model: The optimized UNet-based deep learning model is applied to predict landslide-prone areas, and features of landslide areas are extracted through multi-layer convolution and operations.

[0057] S62, Generate Feature Map: Based on the input data, generate feature maps of landslide-prone areas. ,in Represents spatial coordinates;

[0058] S63, Probabilistic Regression: The generated feature map is processed through a Sigmoid layer to calculate the probability of each pixel belonging to each level of landslide susceptibility, and the landslide susceptibility assessment result is output. , is represented as:

[0059] ;

[0060] in, Representing coordinates The probability of a landslide occurring at a certain location. The value of the feature map at that location;

[0061] S64, Calculate landslide susceptibility: Using the probability results calculated by the Sigmoid layer, predict the probability value of landslide occurrence for each pixel in the image, thereby generating a landslide susceptibility map.

[0062] Optionally, the landslide risk warning and disaster prevention and mitigation recommendations include:

[0063] Early warning tiers: Based on the landslide susceptibility assessment results, a tiered risk early warning mechanism is established. When the probability of a landslide in a certain area... Greater than the preset high-risk threshold A red alert is issued when the probability of a landslide in a certain area is high. Between the medium risk threshold and high risk threshold A yellow alert is issued when the probability of a landslide in a certain area is between [a certain point in time]. Below the medium risk threshold A green alert will be issued at that time;

[0064] Disaster prevention and mitigation recommendations: For different risk warning areas, corresponding disaster prevention and mitigation recommendations are formulated. For red warning areas, it is recommended to strengthen monitoring, evacuate people, and restrict development activities. For yellow warning areas, it is recommended to implement engineering control measures and strengthen patrols. For green warning areas, it is recommended to carry out routine monitoring and periodic assessments.

[0065] The beneficial effects of this invention are:

[0066] This invention employs federated learning technology, enabling participating parties to jointly learn from locally trained models without sharing the original data. This ensures data privacy and security. This technology overcomes the bottlenecks of traditional centralized data storage and computation, allowing multiple data holders to share knowledge in a distributed environment without exposing sensitive data. The introduction of federated learning gives this method strong privacy protection capabilities, making it suitable for fields with high data privacy requirements, such as government agencies and research institutions.

[0067] This invention utilizes the UNet-based deep learning model, which possesses excellent feature extraction capabilities, to effectively improve the efficiency and accuracy of identifying landslide-prone areas. The UNet model extracts features from various influencing factors such as topography and geology through multi-layer convolution and pooling operations, thereby more accurately assessing the likelihood of landslides. By combining these advanced technologies, more efficient and accurate landslide susceptibility assessments can be achieved, and the model can be updated in a timely manner to cope with new data changes, demonstrating high application value and social benefits. Attached Figure Description

[0068] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only for this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0069] Figure 1 This is a schematic diagram of the evaluation method flow according to an embodiment of the present invention;

[0070] Figure 2 This is a schematic diagram showing the geographical location and landslide distribution of the study area in County B, according to an embodiment of the present invention.

[0071] Figure 3 This is a schematic diagram showing the geographical location and landslide distribution of the study area in County A, according to an embodiment of the present invention.

[0072] Figure 4 This is a schematic diagram of a distributed training landslide susceptibility map according to an embodiment of the present invention;

[0073] Figure 5This is a schematic diagram of a non-distributed training landslide susceptibility map according to an embodiment of the present invention;

[0074] Figure 6 This is a schematic diagram of a distributed training landslide susceptibility map according to an embodiment of the present invention;

[0075] Figure 7 This is a schematic diagram of the non-distributed training landslide susceptibility map according to an embodiment of the present invention. Detailed Implementation

[0076] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. It should also be noted that, to make the embodiments more comprehensive, the following embodiments are the best and preferred embodiments, and those skilled in the art can use other alternative methods to implement some well-known technologies; moreover, the accompanying drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.

[0077] It should be noted that the use of terms such as "an embodiment," "an embodiment," "an exemplary embodiment," and "some embodiments" in the specification indicates that the described embodiment may include a specific feature, structure, or characteristic, but not every embodiment necessarily includes that specific feature, structure, or characteristic. Furthermore, when a specific feature, structure, or characteristic is described in connection with an embodiment, implementing such a feature, structure, or characteristic in conjunction with other embodiments (whether explicitly described or not) should be within the knowledge of those skilled in the art.

[0078] Generally, terms can be understood at least partly from their use in context. For example, depending at least partly on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in a singular sense, or a combination of features, structures, or characteristics in a plural sense. Additionally, the term "based on" can be understood not necessarily to convey an exclusive set of factors, but rather, alternatively, depending at least partly on the context, to allow for the presence of other factors that are not necessarily explicitly described.

[0079] like Figures 1-7 As shown, a landslide susceptibility assessment method based on federated learning and UNet includes the following steps:

[0080] Step 1: Extract landslide spatial data of the study area:

[0081] In this embodiment, a landslide-prone area was selected as the research object, with counties A and B as the focus for method validation. These areas are known for their complex topography and geological conditions, with frequent seismic activity and heavy rainfall events leading to frequent landslide disasters. Since the earthquake in county A, changes in the geological structure, combined with rainfall, have further exacerbated the landslide risk.

[0082] like Figure 2 Figure 3 As shown, landslide sample data were collected from counties B and A.

[0083] Step 2: Extract impact factor information:

[0084] Influence factor information is extracted from the multi-source data in Step 1 using the ArcGIS platform.

[0085] Step 3: Construct a landslide sample dataset:

[0086] Acquire historical landslide data and landslide hazard point data;

[0087] Based on historical landslide data and landslide hazard point data, a landslide sample dataset is constructed. Specifically, by determining the number of landslide grid cells within a rectangular area, if the number of landslide grid cells in that area exceeds 50%, the area is marked as a landslide point; if the number is less than 50%, it is marked as a non-landslide point.

[0088] The experimental dataset was randomly divided into a training set and a validation set in an 8:2 ratio, with landslide samples labeled as 1 and non-landslide samples labeled as 0.

[0089] Step 4: Perform data augmentation on the landslide sample dataset:

[0090] The specific steps for data augmentation of the landslide sample dataset are as follows:

[0091] In the landslide sample dataset, identify rasters that contain landslide features;

[0092] Mixup data augmentation technology was applied to randomly stitch together selected landslide grids. Specifically, two landslide samples were selected. and New samples are generated using the following formula. :

[0093] ;

[0094] in, It is in the interval The weighting coefficients are randomly selected within the sample and are used to control the mixing ratio of the two samples.

[0095] The generated augmented samples were added to the experimental dataset to improve the model's generalization ability and robustness.

[0096] Step 5: Build and train the UNet deep learning model.

[0097] A deep learning model based on UNet was designed to ensure better extraction of landslide image features. Specifically, the LeakyRelU activation function was adopted, whose mathematical expression is:

[0098] ;

[0099] in Setting it to 0.01 ensures the continuous propagation of gradients in deep networks, improving the model's learning ability. It replaces max-pooling with convolutional downsampling, expressed as: ,in A convolution kernel with a stride of 2. For bias terms, As the activation function, this design reduces feature resolution while preserving more spatial information. In the decoder, it employs nearest-neighbor interpolation upsampling, combined with 1x1 convolutions to adjust the number of channels. ,in Using an upsampling factor, this design avoids the checkerboard effect that transposed convolution might introduce, ensuring the quality of the output feature map. A normalization layer is introduced into the UNet network for the input... The normalization formula is: ,in, and These are the mean and variance of the current batch, respectively. It is a small constant; the normalized value is then adjusted using scaling and translation parameters. ,in and These are learnable parameters, and Dropout regularization is used:

[0100] .

[0101] Reflection padding is used to handle the convolution boundary. The formula for reflection padding is:

[0102] ,in For the input feature map, This is the feature map after reflection filling. It is the index calculation function for reflection-filled data.

[0103] A binary cross-entropy loss function is used to measure the difference between the probability distribution predicted by the model and the true distribution. Specifically, the loss function is defined as:

[0104] ;

[0105] in, It is the number of landslide samples. It is the first The true label (0 or 1) for each landslide sample. It is the model's prediction of the first The probability value of a sample (between 0 and 1).

[0106] The UNet model was trained for 5 epochs on landslide datasets from counties A and B, yielding gradient parameters for the two locations. and .

[0107] Step Six: Establish a secure federated learning framework to ensure data privacy and security.

[0108] Specifically, a secure communication protocol is established between the two data holders, and federated learning is performed using the landslide data to obtain updated local Unet gradient parameters. During the training process, the gradient parameter update formula for the global Unet model is as follows:

[0109] ;

[0110] in, Indicates the first The amount of data held by each data holder This represents the total amount of all data. For learning rate, Indicates the first Global Unet model in federated learning This represents the global Unet model from the previous round of federated learning. This represents the global gradient function, which performs multiple gradient descent operations based on the computing power of different data holder devices.

[0111] Throughout the process, the servers in County A and County B retain the original data and only transmit encrypted model parameters and gradient information to ensure data privacy and security.

[0112] Step 7: Model Application and Result Output

[0113] The optimized Unet model was applied to data from multiple influencing factor layers in landslide-prone areas. High-dimensional features were extracted through multi-layer convolution operations to generate feature maps. ,in The feature map is represented by spatial coordinates. Then, a sigmoid layer is used to perform probabilistic regression on the feature map, calculating the probability of each pixel belonging to each landslide susceptibility level, and outputting the landslide susceptibility assessment result. Specifically, the Sigmod function is defined as: ,in Representing coordinates The probability of a landslide occurring at a certain location. This represents the value of the feature map at that location. Finally, a landslide susceptibility assessment map is generated based on these probability values, where the grayscale value of each pixel corresponds to the landslide probability at that location.

[0114] Table 1 shows the distribution of each level in the landslide susceptibility zoning maps predicted by the two training methods. In the distribution prediction results for counties A and B, distribution training demonstrated certain advantages in landslide susceptibility assessment, especially in county B, where it significantly improved the prediction accuracy for low-susceptibility areas and reduced misclassifications for medium- and high-susceptibility areas. This is because distribution training can better utilize the data characteristics of each sub-region, enhancing the model's generalization ability and thus improving the overall accuracy and reliability of the predictions.

[0115] Table 1 Statistical Analysis of Landslide Susceptibility Zones

[0116]

[0117] Step 8, Susceptibility Mapping:

[0118] The specific criteria for forming the initial landslide susceptibility distribution map based on probability values ​​are as follows:

[0119] ① A probability value of <50% indicates a low-risk area;

[0120] ②A probability value of 50% ≤ 70% indicates a moderately prone area;

[0121] ③ A probability value > 70% indicates a high-risk area;

[0122] Based on the landslide susceptibility assessment results, a tiered early warning mechanism is established: when the landslide probability in a certain area... Greater than the preset high-risk threshold At that time, a red alert was issued; this falls within the medium risk threshold. and high risk threshold A yellow alert is issued when the risk level is between [a certain threshold] and below the medium risk threshold. When a green alert is issued, corresponding disaster prevention and mitigation recommendations are formulated for areas with different risk levels: for red alert areas, it is recommended to strengthen monitoring, evacuate personnel, and restrict development activities; for yellow alert areas, it is recommended to implement engineering control measures and strengthen patrols; for green alert areas, it is recommended to carry out routine monitoring and periodic assessments. At the same time, the evaluation results will be integrated into the geographic information system to provide a scientific basis for regional disaster prevention and mitigation planning and decision-making.

[0123] This invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To provide the public with a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments; however, those skilled in the art will fully understand the invention even without these details. Furthermore, to avoid unnecessary misunderstanding of the essence of this invention, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0124] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A landslide susceptibility assessment method based on federated learning and UNet, characterized in that, Includes the following steps: S1, Obtain landslide impact factor data: Collect multiple landslide impact factor data and extract multiple impact factor layer data for landslide-prone areas; S2, Establish a federated learning framework: Establish a federated learning framework among multiple data holders; S3, Design UNet Deep Learning Model: Design a UNet-based deep learning model for landslide susceptibility assessment, and input the data from the influencing factor layer into the model for feature extraction and segmentation; S4, Federated Learning Model Training and Parameter Integration: Under the federated learning framework, a UNet-based deep learning model is trained using a landslide sample dataset, and the model parameters are integrated using a federated averaging algorithm to update the global UNet-based deep learning model. S5, Federated Learning Iteration and Model Optimization: Optimize the parameters of the UNet-based deep learning model through multiple rounds of federated learning iterations; S6, Generating Landslide Susceptibility Feature Maps and Probabilistic Regression: An optimized UNet-based deep learning model is applied to generate feature maps of landslide-prone areas. A Sigmoid layer is used for probabilistic regression processing, outputting the landslide susceptibility evaluation results. Specifically, this includes: S61, Optimization and Application of UNet Model: The optimized UNet-based deep learning model is applied to predict landslide-prone areas, and features of landslide areas are extracted through multi-layer convolution and operations. S62, Generate Feature Map: Based on the input data, generate feature maps of landslide-prone areas. ,in Represents spatial coordinates; S63, Probabilistic Regression: The generated feature map is processed through a Sigmoid layer to calculate the probability of each pixel belonging to each level of landslide susceptibility, and the landslide susceptibility assessment result is output. , is represented as: ； in, Representing coordinates The probability of a landslide occurring at a certain location. The value of the feature map at that location; S64, Calculate landslide susceptibility: Using the probability results calculated by the Sigmoid layer, predict the probability value of landslide occurrence for each pixel in the image, thereby generating a landslide susceptibility map; S7, Results Verification and Disaster Prevention and Mitigation Recommendations: Verify and evaluate the landslide susceptibility assessment results, and provide landslide risk warnings and disaster prevention and mitigation recommendations based on the assessment results.

2. The landslide susceptibility assessment method based on federated learning and UNet according to claim 1, characterized in that, The multiple landslide influencing factor data in S1 include topographic factors, geological factors, hydrological factors, land cover factors, and climatic factors, among which; The topographical factors include aspect, slope, elevation, and curvature; The geological factors include lithology; The hydrological factors include average annual precipitation and distance from the river system; The land cover factors include enhanced vegetation index, normalized difference vegetation index, surface reflectance, and land use; The climate factors include annual average temperature and surface temperature.

3. The landslide susceptibility assessment method based on federated learning and UNet according to claim 2, characterized in that, The multiple influencing factor layer data in S1 include elevation, slope, aspect, curvature, lithology, distance from water system, and land use.

4. The landslide susceptibility assessment method based on federated learning and UNet according to claim 3, characterized in that, The construction of the landslide sample dataset includes: Data collection: Acquire historical landslide data and data on potential landslide sites; Constructing a landslide sample dataset: Based on historical landslide data and landslide hazard point data, a landslide sample dataset is constructed. By judging the number of landslide grids in a rectangular area, if the number of landslide grids in the rectangular area exceeds 50%, the rectangular area is marked as a landslide point; if it is less than 50%, it is marked as a non-landslide point. The marked landslide sample dataset is defined as the experimental dataset. Divide the dataset into training and validation sets: Randomly divide the experimental dataset into training and validation sets in an 8:2 ratio, with landslide samples labeled as 1 and non-landslide samples labeled as 0. Data augmentation: Data augmentation is performed on the constructed landslide sample data.

5. The landslide susceptibility assessment method based on federated learning and UNet according to claim 4, characterized in that, The data enhancements include: Identify landslide feature grids: Identify the grids for landslide points in the landslide sample dataset; Applying Mixup data augmentation technology: Mixup data augmentation technology is applied to randomly stitch together the grids of the identified landslide points to generate augmented samples, represented as follows: ； in, For the generated enhanced samples, It is in the interval Weighting coefficients selected randomly within the range, and For landslide samples; Add augmented samples to the experimental dataset: Add the generated augmented samples to the experimental dataset.

6. The landslide susceptibility assessment method based on federated learning and UNet according to claim 5, characterized in that, The parameter update of the global UNet-based deep learning model is represented as follows: ； in, Indicates the first The amount of data held by each data holder This represents the total amount of all data. For learning rate, Indicates the first UNet-based deep learning models in federated learning rounds This refers to the UNet-based deep learning model from the previous round of federated learning. This represents the global gradient function.

7. The landslide susceptibility assessment method based on federated learning and UNet according to claim 6, characterized in that, The training of the UNet-based deep learning model includes: Setting the activation function: LeakyReLU is selected as the activation function for each convolutional layer in the UNet-based deep learning model, expressed as: ； in, It is 0.01; Model training: Applying weight-based learning in the convolutional layers, represented as: ； in, A convolution kernel with a stride of 2. For bias terms, For activation functions; Normalization: Multiple landslide influencing factor data are used as input data, and a normalization layer is introduced to normalize the input data. Normalization is performed, and the normalized values ​​are adjusted using scaling and translation parameters, as follows: ； in, and These are the mean and variance of the current batch, respectively. It is a constant; ； in, and These are learnable parameters; Dropout regularization: Uses the Dropout technique to prevent overfitting, expressed as: ； Reflection padding: Reflection padding is used to handle convolution boundaries, represented as: ； in, For the input feature map, This is the feature map after reflection filling. It is the index calculation function for reflection fill; Model validation: The binary cross-entropy loss function is used to measure the difference between the model's predicted probability distribution and the true distribution, expressed as: ； in, It is the number of landslide samples. It is the first The true label of a landslide sample It is the model's prediction of the first The probability value of each sample.

8. The landslide susceptibility assessment method based on federated learning and UNet according to claim 1, characterized in that, The landslide risk warning and disaster prevention and mitigation recommendations include: Early warning tiers: Based on the landslide susceptibility assessment results, a tiered risk early warning mechanism is established. When the probability of a landslide in a certain area... Greater than the preset high-risk threshold A red alert is issued when the probability of a landslide in a certain area is high. Between medium risk threshold and high risk threshold A yellow alert is issued when the probability of a landslide in a certain area is between [a certain point in time]. Below the medium risk threshold A green alert will be issued at that time; Disaster prevention and mitigation recommendations: For different risk warning areas, corresponding disaster prevention and mitigation recommendations are formulated. For red warning areas, it is recommended to strengthen monitoring, evacuate people, and restrict development activities. For yellow warning areas, it is recommended to implement engineering control measures and strengthen patrols. For green warning areas, it is recommended to carry out routine monitoring and periodic assessments.

Images