Landslide and debris flow slope monitoring method based on microwave radar
By standardizing the microwave radar data and 2D heat map generation, combined with deep learning and decision tree model, the problem of false alarms in traditional microwave radar monitoring is solved, and high-accurate monitoring of landslide mudslide slopes is achieved.
Patent Information
- Application Number
- CN202510239908.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-03
- Publication Date
- 2025-05-30
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Traditional microwave radars are susceptible to environmental factors when monitoring the slopes of landslide mudslides, which has a higher probability of false alarms.
Microwave radar data with multi-point matrix speed measurement is standardized, and a two-dimensional thermal map is generated through Tableau tool. Combined with the Resnet-50 model and decision tree model, the landslide mudslide slope monitoring model is trained, and data processing is optimized to reduce the probability of false alarms.
The accuracy of landslide mudslide slope monitoring is improved, the probability of false alarms is reduced, the overall accuracy can reach more than 90%, and all-weather monitoring is supported.
Smart Images
Figure CN120065197A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of debris flow monitoring, and specifically to a slope monitoring method for landslides and debris flows based on microwave radar. Background Art
[0002] Debris flow is a natural disaster that usually occurs in mountainous or gully areas. It is a special flood triggered by water sources such as heavy rain and snowmelt, carrying a large amount of solid substances such as sediment and stones. It generally occurs in areas with steep terrain, abundant loose materials, and a large amount of water sources in a short time, and has a higher probability of occurring during extreme rainy seasons. Each occurrence will cause certain losses to personal and property, especially in some key areas, the losses are even greater; if rescue personnel arrive in time, they can carry out effective cleaning and rescue to ensure the personal safety of the disaster area. Therefore, how to identify and monitor the slope of landslides and debris flows has become the primary task of rescue work.
[0003] Traditional debris flow monitoring mostly uses cameras and sensors (rainfall sensors, displacement sensors, etc.) to cooperate to complete the monitoring of debris flows. Affected by harsh environments, the images obtained by cameras are unclear, affecting the monitoring results, and the cost of the entire system is also relatively high; in some areas, monitoring methods based on microwave radar are also used, but due to the movement of pedestrians, vehicles, leaves, and obstacles on the slope, the probability of false alarms is high. The present invention optimizes the data obtained by the microwave radar for multi-point monitoring based on a large model of image processing to improve the accuracy of slope monitoring for landslides and debris flows. Summary of the Invention
[0004] Aiming at the deficiencies of the prior art, the present invention provides a slope monitoring method for landslides and debris flows based on microwave radar, which solves the problem that the monitoring method of traditional microwave radar is affected by environmental factors and the probability of false alarms is high.
[0005] To achieve the above objectives, the present invention is realized through the following technical solutions:
[0006] A slope monitoring method for landslides and debris flows based on microwave radar includes the following steps:
[0007] S1. Perform multi-point matrix speed measurement on the target area based on the microwave radar for multi-point monitoring, and standardize the data;
[0008] S2. Input the data into the Tableau tool to generate a two-dimensional heat map;
[0009] S3. Input the two-dimensional heat map into the slope monitoring model for landslides and debris flows, and the slope monitoring model for landslides and debris flows outputs the state of the slope of landslides and debris flows.
[0010] Preferably, the training of the landslide debris flow slope monitoring model includes the following steps:
[0011] Standardize the historical data obtained by multi-point matrix velocity measurement of the target area by microwave radar based on multi-point monitoring to obtain a standard data set X 0 ;
[0012] Input the standard data set X 0 into the Tableau tool to generate a two-dimensional heat map and add labels to the two-dimensional heat map;
[0013] Based on the Resnet-50 model architecture, train the two-dimensional heat map to obtain a landslide debris flow slope monitoring model;
[0014] Among them, the cross-entropy loss function used in the landslide debris flow slope monitoring model is defined as follows:
[0015]
[0016] In formula ①, L represents the cross-entropy loss of the entire data set, N is the total number of samples, L i is the cross-entropy loss of the i-th sample, M is the total number of categories, y ic is the indicator variable that the i-th sample belongs to the c-th category, and p ic is the probability that the model predicts that the i-th sample belongs to the c-th category.
[0017] Preferably, the labels added to the two-dimensional heat map include: "Landslide debris flow warning", "Small-scale landslide debris flow", "Medium-scale landslide debris flow", "Large-scale landslide debris flow", "Normal state".
[0018] Preferably, in step S2, inputting the data into the Tableau tool to generate a two-dimensional heat map specifically includes:
[0019] Establish a first data set A(a 1 , a 2 , a 3 , a 4 ,..., a n ), where a x =(X x , Y x ), x∈(1, n);
[0020] Pre-input the first data set A into Tableau;
[0021] Input the second standardized data set X according to the arrangement positions of the multi-point monitoring microwave radars in sequence, and combine the first data set A and the second data set X to generate a two-dimensional heat map.
[0022] Preferably, it further includes: training of a decision tree model, specifically including the following steps:
[0023] Perform standardization processing on the historical data obtained by multi-point matrix speed measurement of the target area by the multi-point monitoring microwave radar to obtain a standard data set X 0 ;
[0024] Based on a Python program, extract the feature variable data set X of the decision tree model from the standard data set X 0 ; P ;
[0025] Based on the labels corresponding to the standard data set X 0 add classification tasks to the feature variable data set X P ;
[0026] Train the decision tree model;
[0027] Integrated learning diagnostic model, use integrated learning to fuse the landslide and debris flow slope monitoring model and the decision tree model at the decision level to improve the accuracy of landslide and debris flow slope monitoring;
[0028]
[0029] In Equation ②, P(x) is the finally predicted category, P 1 (x) is the category probability predicted by the landslide and debris flow slope monitoring model, P 2 (x) is the category probability of the decision tree model, a is the weight coefficient assigned to the landslide and debris flow slope monitoring model; b is the weight coefficient assigned to the decision tree model;
[0030] Output the landslide and debris flow slope state with the learning diagnostic model.
[0031] Preferably, the screening rule for extracting the feature variable data set X of the decision tree model from the standard data set X 0 is: based on the linear profile of the arrangement positions of the multi-point monitoring microwave radars. P ;
[0032] Preferably, in the process of extracting the feature variable data set X of the decision tree model from the standard data set X 0 optimize the correlation between the feature variable data set X P and the classification tasks by adjusting the screening rules; P ;
[0033] Evaluate the feature variable data set X P for its relevance to the classification task.
[0034] Preferably, in the fusion process of the landslide and debris flow slope monitoring model and the decision tree model, the genetic algorithm is used to optimize and determine the weight coefficient a assigned to the landslide and debris flow slope monitoring model and the weight coefficient b assigned to the decision tree model in the learning and diagnosis model.
[0035] The present invention provides a microwave radar-based one, having the following beneficial effects:
[0036] 1. In the present invention, by collecting the moving speed data of each point on the landslide and debris flow slope in matrix form, setting dimensions for these speed data, and inputting them into the Tableau tool to generate a two-dimensional heat map, a landslide and debris flow slope monitoring model is trained according to the two-dimensional heat map. This model can continuously optimize, reduce the influence of environmental factors, avoid the influence of multi-point speed jumps (the influence of environmental factors, local detection of speed), reduce the probability of false alarms, and the comprehensive accuracy rate can reach more than 90%, laying a foundation for all-weather monitoring of the landslide and debris flow slope. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] Figure 1 is a flowchart of a landslide and debris flow slope monitoring method based on microwave radar proposed by the present invention;
[0038] Figure 2 is the corresponding relationship between the arrangement positions of the microwave radars for multi-point monitoring in the landslide and debris flow slope monitoring method based on microwave radar proposed by the present invention and the data distribution in the two-dimensional heat map;
[0039] Figure 3 is a data corresponding diagram for screening the feature variable data set X based on the linear profile of the arrangement positions of the microwave radars for multi-point monitoring in the landslide and debris flow slope monitoring method based on microwave radar proposed by the present invention P of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0040] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.
[0041] Embodiment 1:
[0042] As Figure 1As shown in the figure, an embodiment of the present invention provides a method for monitoring the slope of landslides and debris flows based on microwave radar, which specifically includes the following steps:
[0043] S1. The microwave radar based on multi-point monitoring performs multi-point matrix speed measurement on the target area, and standardizes the data. The purpose of standardization is to unify the data groups obtained by the microwave radar based on multi-point monitoring for multi-point matrix speed measurement on the target area. For example, the minimum-maximum normalization process is used to ensure that the data in the data group is within the interval [0, 1].
[0044] Each microwave radar for multi-point monitoring can monitor multiple groups of data. Multiple microwave radars for multi-point monitoring can be linearly installed on the slopes of landslides and debris flows in key areas, and a data collector with remote transceiver function is equipped. The data collector collects the data and sends it to the data processing center in the monitoring room.
[0045] S2. Input the data into the Tableau tool to generate a two-dimensional heat map. The data collected by the data collector is a data group obtained at the same moment, and the data in the data group has an order. Set two dimensions of the data group in the Tableau tool, and then generate a two-dimensional heat map.
[0046] S3. Input the two-dimensional heat map into the landslide and debris flow slope monitoring model, and the landslide and debris flow slope monitoring model outputs the state of the landslide and debris flow slope.
[0047] This method collects the moving speed data of each point on the slope of landslides and debris flows in matrix form, sets dimensions for these speed data, inputs them into the Tableau tool to generate a two-dimensional heat map, and trains the landslide and debris flow slope monitoring model based on the two-dimensional heat map. This model can be continuously optimized to reduce the influence of environmental factors, avoid the influence of multi-point speed jumps (the influence of environmental factors, local detection of speed), reduce the probability of false alarms, and the comprehensive accuracy rate can reach more than 90%, laying a foundation for all-weather monitoring of the slopes of landslides and debris flows.
[0048] Specifically, the training of the landslide and debris flow slope monitoring model includes the following steps:
[0049] Step 1. Standardize the historical data obtained by the microwave radar based on multi-point monitoring for multi-point matrix speed measurement on the target area to obtain a standard data group X 0 , for example, the obtained data is (0, 0, 2, 5, 12, 0, 0, 0, 0, 5, 2, 0, 0, 0). After the minimum-maximum normalization standard process, ensure that the data in the data group is within the interval [0, 1].
[0050] The formula for the minimum-maximum normalization standard process is:
[0051] Input the standard data set X 0 into the Tableau tool to generate a two-dimensional heat map, and add labels to the two-dimensional heat map. The labels added to the two-dimensional heat map include: "Landslide and debris flow warning", "Small-scale landslide and debris flow", "Medium-scale landslide and debris flow", "Large-scale landslide and debris flow", "Normal state". The labels reflect the landslide situation on the landslide and debris flow slope surface.
[0052] Based on the Resnet-50 model architecture, train the two-dimensional heat map to obtain a landslide and debris flow slope surface monitoring model. ResNet-50 is a deep convolutional neural network (CNN), which is a variant of the ResNet (Residual Network) architecture and contains 50 layers. Its main features are: solving the gradient disappearance problem: through skip connections, ResNet-50 can train deeper networks and avoid gradient disappearance; ResNet-50 can be used for various tasks, such as image classification and object detection. It trains on the two-dimensional heat map formed by the data obtained from the multi-point matrix speed measurement of the target area by the microwave radar for multi-point monitoring, finds the hidden relationship between the landslide and debris flow and the heat map, and proposes the influence of environmental factors to reduce the probability of false alarms.
[0053] Among them, the cross-entropy loss function used in the landslide and debris flow slope surface monitoring model is defined as follows:
[0054]
[0055] In Equation ①, L represents the cross-entropy loss of the entire data set, N is the total number of samples, L i is the cross-entropy loss of the i-th sample, M is the total number of categories, y ic is the indicator variable that the i-th sample belongs to the c-th category, and p ic is the probability that the model predicts that the i-th sample belongs to the c-th category.
[0056] In one embodiment, inputting the data into the Tableau tool in step S2 to generate a two-dimensional heat map specifically includes:
[0057] As Figure 2 shown, establish a first data set A(a 1 , a 2 , a 3 , a 4 ,..., a n ) for defining the rows and columns of the two-dimensional heat map according to the arrangement positions of the microwave radars for multi-point monitoring, where a x =(X x , Y x ), x ∈ (1, n). Each data a xEach corresponds to a coordinate point on two dimensions in the two-dimensional heat map and is also a position point for obtaining data by the microwave radar for multi-point monitoring. By pre-inputting the first data group A(a 1 ,a 2 ,a 3 ,a 4 ,...,a n ) into the Tableau tool, and then inputting the second data group X that has been standardized according to the arrangement positions of the microwave radars for multi-point monitoring in sequence later. The first data group A and the second data group X are combined to generate a two-dimensional heat map, and there is a strong correlation between this two-dimensional heat map and the velocity jump on the landslide and debris flow slope surface. Thus, the landslide and debris flow slope surface monitoring model trained in this way is also more accurate.
[0058] Embodiment 2
[0059] A landslide and debris flow slope surface monitoring method based on microwave radar disclosed in this embodiment further trains a decision tree model on the basis of Embodiment 1, and uses an integrated learning diagnosis model to fuse the decision tree model with the landslide and debris flow slope surface monitoring model, and finally obtains a learning diagnosis model with higher monitoring accuracy.
[0060] The training of the decision tree model specifically includes the following steps:
[0061] Standardize the historical data obtained by multi-point matrix speed measurement of the target area by the microwave radar for multi-point monitoring to obtain the standard data group X 0 .
[0062] Based on the Python program, extract the feature variable data group X 0 from the standard data group X P , and the feature variable data group X P ∈ the standard data group X 0 .
[0063] For example: input
[0064] data: The original data, which can be a list.
[0065] indices: The list of serial numbers, indicating the indices of the elements that need to be taken out from the data.
[0066] Output
[0067] new_data: The new data group taken out from the data according to the serial numbers in the indices.
[0068] Based on the label corresponding to the standard data group X 0 , for the feature variable data group X PAdd a classification task, that is, each feature variable data group X P corresponds to a classification task label, and the classification task label is the same as the label for training the landslide and debris flow slope monitoring model.
[0069] Train a decision tree model;
[0070] Integrated learning diagnostic model, using integrated learning to fuse the landslide and debris flow slope monitoring model and the decision tree model at the decision level to improve the accuracy of landslide and debris flow slope monitoring.
[0071]
[0072] In Equation ②, P(x) is the finally predicted class, P 1 (x) is the class probability predicted by the landslide and debris flow slope monitoring model, P 2 (x) is the class probability of the decision tree model, a is the weight coefficient assigned to the landslide and debris flow slope monitoring model; b is the weight coefficient assigned to the decision tree model.
[0073] Finally, use the learning diagnostic model to monitor the landslide and debris flow slope, and take the output landslide and debris flow slope state of the learning diagnostic model as the final monitoring result.
[0074] In one embodiment, as Figure 3 shown in, from the standard data group X 0 extract the feature variable data group X P of the decision tree model, and the screening rule is: based on the linear profile of the microwave radar layout positions for multi-point monitoring, the data group obtained by this method contains the edge velocity data of the rectangular array. Generally, when a landslide and debris flow occur, there must be a jump in the edge velocity data (the area of the landslide and debris flow is larger than the monitoring area, so the velocity data can definitely be detected at the edge). Therefore, the edge velocity data of the rectangular array is highly correlated with the occurrence of the landslide and debris flow.
[0075] In one embodiment, in order to find a screening rule that can make the selected feature variable data group X P have the greatest correlation with the occurrence of the landslide and debris flow and finally train a more stable decision tree model, this embodiment designs an optimization scheme: during the process of extracting the feature variable data group X 0 of the decision tree model from the standard data group X P , optimize the correlation between the feature variable data group X P and the classification task by adjusting the screening rule, which can be randomly set based on a Python program.
[0076] In the above optimization scheme, evaluate the feature variable data group X PThe relevance to the classification task, and the corresponding screening rule with the greatest relevance is the optimal one.
[0077] In one embodiment, in the process of fusing the landslide debris flow slope monitoring model and the decision tree model, the genetic algorithm is used to optimize and determine the weight coefficient a assigned to the landslide debris flow slope monitoring model and the weight coefficient b assigned to the decision tree model in the learning diagnosis model.
[0078] It should be noted that in the above solution, the landslide debris flow slope monitoring model, the decision tree model, and the finally fused learning diagnosis model are given. Each of these three models can be independently used to monitor the landslide debris flow slope. Generally, the performance of the learning diagnosis model is greater than that of the landslide debris flow slope monitoring model, which is greater than that of the decision tree model.
[0079] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principle and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.
Claims
1. A method for monitoring landslide and debris flow slope based on microwave radar, characterized in that: The following steps are involved: S1. Microwave radar based on multi-point monitoring performs multi-point matrix speed measurement on the target area and standardizes the data; S2. Input the data into the Tableau tool to generate a two-dimensional heat map; S3. Input the two-dimensional thermal map into the landslide and debris flow slope monitoring model, and the landslide and debris flow slope monitoring model outputs the landslide and debris flow slope status.
2. The method for monitoring landslide and debris flow slope based on microwave radar according to claim 1, characterized in that: The training of the landslide and debris flow slope monitoring model includes the following steps: The historical data obtained by multi-point matrix speed measurement in the target area based on multi-point monitoring microwave radar is standardized to obtain a standard data set X0; Input the standard data set X0 into the Tableau tool to generate a two-dimensional heat map and add labels to the two-dimensional heat map; Based on the Resnet-50 model architecture, the two-dimensional thermal map was trained to obtain the landslide and debris flow slope monitoring model; Among them, the cross entropy loss function used in the landslide and debris flow slope monitoring model is defined as follows: In formula ①, L represents the cross entropy loss of the entire data set, N is the total number of samples, and L i is the cross entropy loss of the ith sample, M is the total number of categories, and y ic is the indicator variable that the i-th sample belongs to the c-th class, p ic is the probability that the model predicts that the i-th sample belongs to the c-th class.
3. The method for monitoring landslide and debris flow slope based on microwave radar according to claim 2, characterized in that: The labels added to the two-dimensional heat map include: "land slide and debris flow warning", "small-scale land slide and debris flow", "medium-scale land slide and debris flow", "large-scale land slide and debris flow", and "normal state".
4. A method for monitoring landslide and debris flow slope based on microwave radar according to claim 2 or 3, characterized in that: In the step S2, the data is input into the Tableau tool to generate a two-dimensional heat map, which specifically includes: A first data set A (a1, a2, a3, a4, ..., a5) for defining rows and columns of a two-dimensional thermal map is established according to the arrangement positions of the microwave radars for multi-point monitoring. n ), where a x =(X x ,Y x ),x∈(1,n); Input the first data set A into Tableau in advance; The standardized second data group X is input in sequence according to the arrangement positions of the microwave radars for multi-point monitoring, and the first data group A and the second data group X are combined to generate a two-dimensional thermal map.
5. The method for monitoring landslide and debris flow slope based on microwave radar according to claim 2, characterized in that: Also includes: The training of the decision tree model includes the following steps: The historical data obtained by multi-point matrix speed measurement in the target area based on multi-point monitoring microwave radar is standardized to obtain a standard data set X0; Based on Python program, extract the characteristic variable data set X of decision tree model from the standard data set X0 P ; Based on the label corresponding to the standard data set X0, it is the characteristic variable data set X P Add classification tasks; Train a decision tree model; Ensemble learning diagnostic model, which uses ensemble learning to fuse the landslide and debris flow slope monitoring model and decision tree model at the decision level to improve the accuracy of landslide and debris flow slope monitoring; In formula ②, P(x) is the final predicted category, P1(x) is the category probability predicted by the landslide and debris flow slope monitoring model, P2(x) is the category probability of the decision tree model, a is the weight coefficient assigned to the landslide and debris flow slope monitoring model; b is the weight coefficient assigned to the decision tree model; The learning diagnosis model is used to output the landslide and debris flow slope status.
6. The method for monitoring landslide and debris flow slope based on microwave radar according to claim 5, characterized in that: Extract the characteristic variable data set X of the decision tree model from the standard data set X0 P The screening rule is: linear outline of the microwave radar layout location based on multi-point monitoring.
7. The method for monitoring landslide and debris flow slope based on microwave radar according to claim 5, characterized in that: Extract the characteristic variable data set X of the decision tree model from the standard data set X0 P In the process, the characteristic variable data set X is optimized by adjusting the screening rules. P Relevance to classification tasks; Evaluate the feature variable data set X with the accuracy of the decision tree model P Relevance to classification tasks.
8. The method for monitoring landslide and debris flow slope based on microwave radar according to claim 5, characterized in that: The fusion process of the landslide and debris flow slope monitoring model and the decision tree model is optimized by genetic algorithm to determine the weight coefficient a assigned to the landslide and debris flow slope monitoring model and the weight coefficient b assigned to the decision tree model in the learning diagnosis model.