Potential landslide hidden danger identification method based on rock mass deterioration characteristics

Abstract

The invention relates to the technical field of geological survey, in particular to a potential landslide hidden danger identification method based on rock mass deterioration characteristics. The technical problems that in the prior art, in landslide monitoring, the monitoring range is limited, data collection is not comprehensive, and early warning timeliness is insufficient are solved; according to the technical scheme, the potential landslide hidden danger identification method based on the rock mass degradation characteristics comprises the steps that feature extraction is conducted on a preprocessed image through a convolutional neural network, the rock mass degradation characteristics are identified, and the identified degradation characteristics are converted into numerical value characteristics; compared with the problems of limited monitoring range, incomplete data collection and insufficient early warning timeliness in landslide monitoring in the prior art, the intelligent machine vision technology is used for identifying the rock mass degradation characteristics, and the sensor network is deployed in combination with the Internet of Things technology to monitor the rock mass state in real time; the cloud computing platform is used for data analysis and processing, and finally intelligent early warning of the landslide hidden danger is achieved.

Description

Technical Field

[0001] The invention relates to the technical field of geological survey, and in particular to a method for identifying potential landslide hazards based on rock mass degradation characteristics. Background Art

[0002] Landslides are one of the most common geological disasters, posing a serious threat to people's lives and property. Traditional landslide monitoring methods mainly rely on manual inspections and simple instrument monitoring, which have problems such as limited monitoring range, incomplete data collection, and insufficient warning timeliness. With the advancement of science and technology, especially the development of artificial intelligence, the Internet of Things, and cloud computing technologies, new technical means have been provided for landslide monitoring. However, the existing technology still lacks a method that can comprehensively, real-timely, and accurately identify rock mass degradation characteristics and predict landslide hazards. Therefore, the existing technology in landslide monitoring has the problems of limited monitoring range, incomplete data collection, and insufficient warning timeliness. This paper proposes a potential landslide hazard identification method based on rock mass degradation characteristics. The rock mass degradation characteristics are identified by intelligent machine vision technology, and the sensor network is deployed in combination with the Internet of Things technology to monitor the rock mass status in real time. The cloud computing platform is used for data analysis and processing, and finally an intelligent early warning of landslide hazards is achieved. Summary of the invention

[0003] In order to overcome the problems of limited monitoring range, incomplete data collection and insufficient warning timeliness in existing landslide monitoring technologies.

[0004] The technical solution of the present invention is: a method for identifying potential landslide hazards based on rock mass degradation characteristics, comprising the following steps: S1: Data collection: Deploy high-definition cameras and drones for aerial photography, as well as infrared and radar sensors in the target areas of landslide hazard monitoring points to capture the surface morphology, color, texture, and temperature multi-dimensional information of the rock mass, and pre-process the collected image and video data, including denoising, contrast enhancement, and distortion correction; S2: Feature extraction: A convolutional neural network is used to extract features from the preprocessed image to identify rock mass degradation features, including but not limited to rock mass cracks, spalling and discoloration, and the identified degradation features are converted into numerical features, including but not limited to crack width, spalling area and color change degree; S3: Deployment of sensor network: Sensors are arranged at the landslide hazard monitoring points. The sensors include accelerometers, displacement sensors and moisture sensors, which are used to monitor the tiny vibrations, displacement changes and moisture content of the rock mass in real time. At the same time, the sensor data is transmitted to the cloud computing platform in real time using the Internet of Things technology; S4: Data storage on cloud computing platform: Sensor data and image data are stored in cloud database; S5: Data preparation for early warning algorithm: Collect information on historical landslide events, filter out characteristic variables related to landslides from real-time monitoring data, including displacement rate, crack extension speed and rainfall, and divide the data into training set and test set; S6: Model training: Use the training set data to build multiple decision trees. Each tree is split based on a randomly selected feature subset to form a forest. The parameters of the random forest, including the number of trees and the maximum depth, are adjusted through the cross-validation method to perform module optimization. S7: Early warning prediction: Input the real-time monitoring feature data into the trained random forest model to obtain the probability prediction of landslide occurrence; S8: Set a warning threshold based on historical landslide data and landslide risk assessment, and trigger a warning signal when the predicted landslide probability exceeds the threshold; S9: Warning release: Landslide warning information will be released to relevant departments and personnel through SMS, email and APP push. The landslide warning information includes warning level and recommended measures.

[0005] Preferably, in step S1, high-definition cameras and drones are deployed in the target area for aerial photography, as well as infrared and radar sensors to capture the surface morphology, color, texture, and temperature multi-dimensional information of the rock mass, and the collected image and video data are pre-processed, including denoising, contrast enhancement, and distortion correction; the specific steps are as follows: S101: Install a high-definition camera at the monitoring point in the target area of ​​the landslide hazard monitoring point, and configure the camera's shooting parameters, including resolution, frame rate, and exposure time; S102: Arrange the UAV to perform regular cruise photography according to the preset flight route and altitude to obtain large-scale, high-resolution images of the rock mass; S103: Install infrared sensors at the monitoring points in the target area of ​​landslide hazard monitoring points to detect the temperature distribution and abnormal hot spots of the rock mass, and deploy radar sensors to monitor the surface displacement and micro-vibration of the rock mass; S104: The high-definition camera and the drone automatically collect image and video data according to the set time interval and trigger conditions; S105: Infrared sensors and radar sensors collect rock temperature, displacement and vibration data in real time; S106: The sensor data is transmitted to the cloud computing platform via wireless means; S107: Using a Gaussian filtering algorithm to perform denoising on the collected image and video data to remove the clutter and noise caused by device noise and environmental interference; S108: using a histogram equalization method to make the bright and dark areas in the image data more distinct, performing contrast enhancement processing on the image data, and improving the visibility of the surface morphology and degradation characteristics of the rock mass; S109: using a barrel distortion correction algorithm to correct image distortion caused by camera lens distortion and drone shooting angle problems; S110: Output the pre-processed image and video data.

[0006] Preferably, in step S2, a convolutional neural network is used to extract features from the preprocessed image to identify rock mass degradation features, which include but are not limited to rock mass cracks, spalling and discoloration, and the identified degradation features are converted into numerical features, which include but are not limited to crack width, spalling area and color change degree; wherein the specific steps of extracting features from the preprocessed image using a convolutional neural network are as follows: S201: constructing a convolutional neural network architecture of rock mass degradation features, including an input layer, a convolution layer, a pooling layer, a fully connected layer and an output layer; wherein the input layer receives preprocessed image data, and the output layer outputs the identified degradation feature category and location; S202: Divide the preprocessed image data into a training set, a validation set, and a test set; S203: training the convolutional neural network using the training set data, and optimizing the network parameters by a back propagation algorithm; S204: Select the cross entropy loss function and Adam optimizer to train the model, evaluate the model performance on the validation set regularly, and adjust the learning rate and batch size parameters to optimize the model; S205: Evaluate the trained model on the test set and select the model with the best performance as the feature extraction module; S206: inputting the preprocessed image into the trained convolutional neural network; S207: The image data passes through the convolution layer, the convolution kernel slides over the image and calculates the convolution operation to extract local features of the image, and multiple convolution layers gradually extract higher-level features; S208: The pooling layer downsamples the feature map output by the convolutional layer; S209: The fully connected layer flattens and connects the feature maps output by the pooling layer to form a one-dimensional feature vector; S210: The convolutional neural network outputs the identified rock mass degradation features, including but not limited to cracks, spalling, and discoloration.

[0007] Preferably, in step S2, a convolutional neural network is used to extract features from the preprocessed image to identify rock mass degradation features, which include but are not limited to rock mass cracks, spalling and discoloration, and the identified degradation features are converted into numerical features, which include but are not limited to crack width, spalling area and degree of color change; wherein the specific steps of converting the identified degradation features into numerical features are as follows: S211: measuring the width of the crack for the identified crack features by edge detection and contour extraction image processing methods, and recording the crack width as a numerical value; S212: For the identified spalling features, an image segmentation method is used to separate the spalling region from the background, and the area of ​​the spalling region is calculated and recorded as a numerical feature; S213: For the identified color change feature, a color space conversion and a color histogram analysis method are used to evaluate the degree of color change, and the color difference between the color change area and the surrounding normal area is calculated and recorded as a numerical feature; S214: Integrate the numerical features of crack width, spalling area and color change degree to form a complete numerical representation of rock mass degradation characteristics, and output the numerical features.

[0008] Preferably, in step S3, sensors are arranged at the landslide hazard monitoring points, and the sensors include accelerometers, displacement sensors and moisture sensors, which are used to monitor the tiny vibrations, displacement changes and moisture content of the rock mass in real time, and the sensor data are transmitted to the cloud computing platform in real time using the Internet of Things technology; the specific steps are as follows: S301: Arrange sensors at the landslide hazard monitoring points, including accelerometers, displacement sensors, and moisture sensors. The accelerometers are used to monitor tiny vibrations of the rock mass, the displacement sensors are used to monitor displacement changes of the rock mass, and the moisture sensors are used to monitor the water content of the rock mass. S302: Connect the sensor network to the Internet of Things platform and equip each sensor node with a unique identifier; S303: The data collected by the sensor node is transmitted to the cloud computing platform and the Internet of Things platform in real time through the Internet of Things communication method; S304: Real-time monitoring of sensor data through the IoT platform to display the vibration, displacement changes and water content information of the rock mass.

[0009] Preferably, in step S4, the sensor data and image data are stored in a cloud database; the specific steps are as follows: S401: The cloud computing platform receives sensor data and pre-processed image data from landslide hazard monitoring points in real time through the Internet of Things communication technology; S402: using a check code and data packet size check mechanism to mark and discard abnormal data, so as to perform integrity check and error detection on the received data; S403: Convert the received raw data into JSON format; S404: Selecting the relational database MySQL to construct a database table structure for the sensor data and the image data, including field names, data types, and indexes; S405: partition and store the data according to the timestamp and geographic location information of the data; S406: Write the preprocessed data into the database, and create an index for the database table through the B-tree index.

[0010] Preferably, in step S5, information on historical landslide events is collected, characteristic variables related to landslides are screened out from real-time monitoring data, the characteristic variables include displacement rate, crack propagation speed and rainfall, and the data are divided into a training set and a test set; the specific steps are as follows: S501: Collect historical landslide event information through geological survey reports, remote sensing monitoring data and ground observation records. The historical landslide event information should include the time, location, scale, impact range and triggering factors of the landslide. The historical landslide event information should be sorted out to form a standardized data set; S502: Determine characteristic variables related to landslides based on historical landslide event information, including displacement rate, crack extension speed, and rainfall; S503: Filtering data containing the above-mentioned characteristic variables from the real-time monitoring data; S504: Clean the screened data to remove abnormal values ​​and missing values; S505: Divide the cleaned data into a training set and a test set, wherein the training set is used to build and train the early warning algorithm model, and the test set is used to evaluate the performance of the model; S506: Standardizing the feature variables so that they have the same scale, and normalizing the feature variables so that the range of the feature variables is between [0, 1] or [-1, 1]; S507: using the principal component analysis method to select characteristic variables that contribute more to landslide early warning according to the correlation between historical landslide event information and real-time monitoring data; S508: Integrate the preprocessed and divided data into a final data set; S509: Save the data set in CSV format and store it in a database.

[0011] Preferably, in step S6, multiple decision trees are constructed using the training set data, each tree is split based on a randomly selected feature subset to form a forest, and the parameters of the random forest, including the number of trees and the maximum depth, are adjusted by a cross-validation method to perform module optimization; the specific steps are as follows: S601: Randomly extract multiple sample subsets from the preprocessed training set data, and each sample subset will be used to train a decision tree; S602: During the training process of each decision tree, a feature subset is randomly selected as a split feature of the current node; S603: For each node, according to the selected feature subset, select an optimal feature and threshold for splitting, so that the purity of the subset after splitting reaches the maximum. This process is recursively performed until the stopping condition is met. The stopping condition includes that the number of samples contained in the node is less than a certain threshold and the purity of the node reaches a certain standard; S604: generating a complete decision tree by continuously splitting nodes, and the decision tree is used to output corresponding landslide hazard prediction results according to the input feature values; S605: Repeat steps S601-S604 to construct multiple decision trees to form a random forest model; S606: Divide the training set data into K subsets in K-fold cross validation, each subset will be used as a validation set in turn, and the remaining subsets will be used as training sets; S607: Setting the initial parameters of the random forest model, including the number of trees, maximum depth, and minimum number of samples; S608: For each round of cross-validation, the random forest model is trained using the training set data of the current round, and the trained model is evaluated using the validation set data to calculate the prediction accuracy, recall rate, and F1 score; S609: Based on the evaluation results on the validation set, analyze the impact of different parameters on the model performance, use the Bayesian optimization method to adjust the parameters of the random forest, and optimize the model.

[0012] Preferably, in step S7, the feature data of real-time monitoring is input into the trained random forest model to obtain the probability prediction of landslide occurrence; the specific steps are as follows: S701: inputting characteristic data for real-time monitoring, including displacement rate, crack extension speed, rainfall, accelerometer data, displacement sensor data, and moisture sensor data; S702: Loading a trained random forest model from a cloud computing platform or local storage; S703: The feature data monitored in real time is used as input and passed to the input layer of the random forest model; S704: For each decision tree in the random forest, starting from the root node, according to the input feature data, the node splitting condition and the threshold, the decision tree is traversed layer by layer, wherein at each node, according to the splitting feature of the current node, it is determined which child node the input data belongs to, and the traversal is continued downward until a leaf node is reached, and the leaf node stores the prediction result of the decision tree for the input data; S705: Calculate the average of all decision tree prediction results as the final prediction result of the random forest model for the input data. For all decision trees in the random forest, collect the prediction results of each tree, that is, the probability of landslide occurrence. The calculation steps are as follows: The random forest model is set to contain N decision trees, and the prediction result of each tree is , then the final prediction result P of the random forest model is: ; in, is the prediction result of the i-th decision tree for the input data, i.e., the probability of landslide occurrence; S706: Output the predicted probability value of landslide occurrence, indicating the possibility of landslide occurrence.

[0013] Preferably, in step S8, a warning threshold is set according to historical landslide data and landslide risk assessment, and when the predicted landslide probability exceeds the threshold, a warning signal is triggered; the specific steps are as follows: S801: Collect landslide events from geological survey reports, historical records, and monitoring data sources, record the time, location, scale, triggering factors, and actual damage caused by each landslide, and correspond to each landslide event, find and record the predicted probability at that time; S802: Based on the relationship between historical landslide data and predicted probability, determine the landslide occurrence frequency corresponding to different predicted probability intervals, set risk levels for different hazard levels based on the potential hazards of landslides, including casualties, property losses, and environmental impacts, and develop a landslide risk assessment curve based on the predicted probability and hazard level; S803: According to the risk assessment curve, the prediction probability thresholds corresponding to different risk levels are determined, and multiple thresholds are set corresponding to different warning levels, including low-level warning, medium-level warning, and high-level warning; S804: Acquire the real-time monitoring characteristic data and the probability prediction value of landslide occurrence from the cloud computing platform; S805: Compare the predicted probability with the set warning threshold. If the predicted probability exceeds the threshold of a certain warning level, a warning signal of the corresponding level is triggered.

[0014] Beneficial effects of the present invention: 1. Compared with the existing problems of limited monitoring range and incomplete data collection in landslide monitoring, the present invention realizes comprehensive and real-time monitoring of rock mass degradation characteristics through intelligent machine vision and Internet of Things technology, greatly improving monitoring efficiency and accuracy; 2. Compared with the problem of insufficient warning timeliness in landslide monitoring in the prior art, the present invention uses cloud computing platform for data analysis and processing, combined with intelligent warning algorithm, to predict the possibility of landslide in real time and issue warning signals in time, thus winning precious time for disaster prevention and mitigation and enhancing the timeliness of warning; 3. Compared with manual inspections and simple instrument monitoring under the existing technology, the present invention can cover a wider monitoring area, while reducing the investment in manpower and material resources and lowering the monitoring cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 What is shown is a simplified flow chart of the steps of the potential landslide hazard identification method based on rock mass degradation characteristics of the present invention; Figure 2 What is shown is a schematic flow chart of detailed steps of the method for identifying potential landslide hazards based on rock mass degradation characteristics of the present invention. DETAILED DESCRIPTION

[0016] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0017] See also Figure 1-2 The present invention provides an embodiment: a method for identifying potential landslide hazards based on rock mass degradation characteristics, comprising the following steps: S1: Data collection: Deploy high-definition cameras and drones for aerial photography, as well as infrared and radar sensors in the target areas of landslide hazard monitoring points to capture the surface morphology, color, texture, and temperature multi-dimensional information of the rock mass, and pre-process the collected image and video data, including denoising, contrast enhancement, and distortion correction; S2: Feature extraction: A convolutional neural network is used to extract features from the preprocessed image to identify rock mass degradation features, including but not limited to rock mass cracks, spalling and discoloration, and the identified degradation features are converted into numerical features, including but not limited to crack width, spalling area and color change degree; S3: Deployment of sensor network: Sensors are arranged at the landslide hazard monitoring points. The sensors include accelerometers, displacement sensors and moisture sensors, which are used to monitor the tiny vibrations, displacement changes and moisture content of the rock mass in real time. At the same time, the sensor data is transmitted to the cloud computing platform in real time using the Internet of Things technology; S4: Data storage on cloud computing platform: Sensor data and image data are stored in cloud database; S5: Data preparation for early warning algorithm: Collect information on historical landslide events, filter out characteristic variables related to landslides from real-time monitoring data, including displacement rate, crack extension speed and rainfall, and divide the data into training set and test set; S6: Model training: Use the training set data to build multiple decision trees. Each tree is split based on a randomly selected feature subset to form a forest. The parameters of the random forest, including the number of trees and the maximum depth, are adjusted through the cross-validation method to perform module optimization. S7: Early warning prediction: Input the real-time monitoring feature data into the trained random forest model to obtain the probability prediction of landslide occurrence; S8: Set a warning threshold based on historical landslide data and landslide risk assessment, and trigger a warning signal when the predicted landslide probability exceeds the threshold; S9: Warning release: Landslide warning information will be released to relevant departments and personnel through SMS, email and APP push. The landslide warning information includes warning level and recommended measures.

[0018] Preferably, in step S1, high-definition cameras and drones are deployed in the target area for aerial photography, as well as infrared and radar sensors to capture the surface morphology, color, texture, and temperature multi-dimensional information of the rock mass, and the collected image and video data are pre-processed, including denoising, contrast enhancement, and distortion correction; the specific steps are as follows: S101: Install a high-definition camera at the monitoring point in the target area of ​​the landslide hazard monitoring point, and configure the camera's shooting parameters, including resolution, frame rate, and exposure time; S102: Arrange the UAV to perform regular cruise photography according to the preset flight route and altitude to obtain large-scale, high-resolution images of the rock mass; S103: Install infrared sensors at the monitoring points in the target area of ​​landslide hazard monitoring points to detect the temperature distribution and abnormal hot spots of the rock mass, and deploy radar sensors to monitor the surface displacement and micro-vibration of the rock mass; S104: The high-definition camera and the drone automatically collect image and video data according to the set time interval and trigger conditions; S105: Infrared sensors and radar sensors collect rock temperature, displacement and vibration data in real time; S106: The sensor data is transmitted to the cloud computing platform via wireless means; S107: Using a Gaussian filtering algorithm to perform denoising on the collected image and video data to remove the clutter and noise caused by device noise and environmental interference; S108: using a histogram equalization method to make the bright and dark areas in the image data more distinct, performing contrast enhancement processing on the image data, and improving the visibility of the surface morphology and degradation characteristics of the rock mass; S109: using a barrel distortion correction algorithm to correct image distortion caused by camera lens distortion and drone shooting angle problems; S110: Output the pre-processed image and video data.

[0019] Preferably, in step S2, a convolutional neural network is used to extract features from the preprocessed image to identify rock mass degradation features, which include but are not limited to rock mass cracks, spalling and discoloration, and the identified degradation features are converted into numerical features, which include but are not limited to crack width, spalling area and color change degree; wherein the specific steps of extracting features from the preprocessed image using a convolutional neural network are as follows: S201: constructing a convolutional neural network architecture of rock mass degradation features, including an input layer, a convolution layer, a pooling layer, a fully connected layer and an output layer; wherein the input layer receives preprocessed image data, and the output layer outputs the identified degradation feature category and location; S202: Divide the preprocessed image data into a training set, a validation set, and a test set; S203: training the convolutional neural network using the training set data, and optimizing the network parameters by a back propagation algorithm; S204: Select the cross entropy loss function and Adam optimizer to train the model, evaluate the model performance on the validation set regularly, and adjust the learning rate and batch size parameters to optimize the model; S205: Evaluate the trained model on the test set and select the model with the best performance as the feature extraction module; S206: inputting the preprocessed image into the trained convolutional neural network; S207: The image data passes through the convolution layer, the convolution kernel slides over the image and calculates the convolution operation to extract local features of the image, and multiple convolution layers gradually extract higher-level features; S208: The pooling layer downsamples the feature map output by the convolutional layer; S209: The fully connected layer flattens and connects the feature maps output by the pooling layer to form a one-dimensional feature vector; S210: The convolutional neural network outputs the identified rock mass degradation features, including but not limited to cracks, spalling, and discoloration.

[0020] Preferably, in step S2, a convolutional neural network is used to extract features from the preprocessed image to identify rock mass degradation features, which include but are not limited to rock mass cracks, spalling and discoloration, and the identified degradation features are converted into numerical features, which include but are not limited to crack width, spalling area and degree of color change; wherein the specific steps of converting the identified degradation features into numerical features are as follows: S211: measuring the width of the crack for the identified crack features by edge detection and contour extraction image processing methods, and recording the crack width as a numerical value; S212: For the identified spalling features, an image segmentation method is used to separate the spalling region from the background, and the area of ​​the spalling region is calculated and recorded as a numerical feature; S213: For the identified color change feature, a color space conversion and a color histogram analysis method are used to evaluate the degree of color change, and the color difference between the color change area and the surrounding normal area is calculated and recorded as a numerical feature; S214: Integrate the numerical features of crack width, spalling area and color change degree to form a complete numerical representation of rock mass degradation characteristics, and output the numerical features.

[0021] Preferably, in step S3, sensors are arranged at the landslide hazard monitoring points, and the sensors include accelerometers, displacement sensors and moisture sensors, which are used to monitor the tiny vibrations, displacement changes and moisture content of the rock mass in real time, and the sensor data are transmitted to the cloud computing platform in real time using the Internet of Things technology; the specific steps are as follows: S301: Arranging sensors at landslide hazard monitoring points, including accelerometers, displacement sensors, and moisture sensors. The accelerometers are used to monitor tiny vibrations of the rock mass, the displacement sensors are used to monitor displacement changes of the rock mass, and the moisture sensors are used to monitor the water content of the rock mass. S302: Connect the sensor network to the Internet of Things platform and equip each sensor node with a unique identifier; S303: The data collected by the sensor node is transmitted to the cloud computing platform and the Internet of Things platform in real time through the Internet of Things communication method; S304: Real-time monitoring of sensor data through the IoT platform to display the vibration, displacement changes and water content information of the rock mass.

[0022] Preferably, in step S4, the sensor data and image data are stored in a cloud database; the specific steps are as follows: S401: The cloud computing platform receives sensor data and pre-processed image data from landslide hazard monitoring points in real time through the Internet of Things communication technology; S402: using a check code and data packet size check mechanism to mark and discard abnormal data, so as to perform integrity check and error detection on the received data; S403: Convert the received raw data into JSON format; S404: Selecting the relational database MySQL to construct a database table structure for the sensor data and the image data, including field names, data types, and indexes; S405: partition and store the data according to the timestamp and geographic location information of the data; S406: Write the preprocessed data into the database, and create an index for the database table through the B-tree index.

[0023] Preferably, in step S5, information on historical landslide events is collected, characteristic variables related to landslides are screened out from real-time monitoring data, the characteristic variables include displacement rate, crack propagation speed and rainfall, and the data are divided into a training set and a test set; the specific steps are as follows: S501: Collect historical landslide event information through geological survey reports, remote sensing monitoring data and ground observation records. The historical landslide event information should include the time, location, scale, impact range and triggering factors of the landslide. The historical landslide event information should be sorted out to form a standardized data set; S502: Determine characteristic variables related to landslides based on historical landslide event information, including displacement rate, crack extension speed, and rainfall; S503: Filtering data containing the above-mentioned characteristic variables from the real-time monitoring data; S504: Clean the screened data to remove abnormal values ​​and missing values; S505: Divide the cleaned data into a training set and a test set, wherein the training set is used to build and train the early warning algorithm model, and the test set is used to evaluate the performance of the model; S506: Standardizing the feature variables so that they have the same scale, and normalizing the feature variables so that the range of the feature variables is between [0, 1] or [-1, 1]; S507: using the principal component analysis method to select characteristic variables that contribute more to landslide early warning according to the correlation between historical landslide event information and real-time monitoring data; S508: Integrate the preprocessed and divided data into a final data set; S509: Save the data set in CSV format and store it in a database.

[0024] Preferably, in step S6, multiple decision trees are constructed using the training set data, each tree is split based on a randomly selected feature subset to form a forest, and the parameters of the random forest, including the number of trees and the maximum depth, are adjusted by a cross-validation method to perform module optimization; the specific steps are as follows: S601: Randomly extract multiple sample subsets from the preprocessed training set data, and each sample subset will be used to train a decision tree; S602: During the training process of each decision tree, a feature subset is randomly selected as a split feature of the current node; S603: For each node, according to the selected feature subset, select an optimal feature and threshold for splitting, so that the purity of the subset after splitting reaches the maximum. This process is recursively performed until the stopping condition is met. The stopping condition includes that the number of samples contained in the node is less than a certain threshold and the purity of the node reaches a certain standard; S604: generating a complete decision tree by continuously splitting nodes, and the decision tree is used to output corresponding landslide hazard prediction results according to the input feature values; S605: Repeat steps S601-S604 to construct multiple decision trees to form a random forest model; S606: Divide the training set data into K subsets in K-fold cross validation, each subset will be used as a validation set in turn, and the remaining subsets will be used as training sets; S607: Setting the initial parameters of the random forest model, including the number of trees, maximum depth, and minimum number of samples; S608: For each round of cross-validation, the random forest model is trained using the training set data of the current round, and the trained model is evaluated using the validation set data to calculate the prediction accuracy, recall rate, and F1 score; S609: Based on the evaluation results on the validation set, analyze the impact of different parameters on the model performance, use the Bayesian optimization method to adjust the parameters of the random forest, and optimize the model.

[0025] Preferably, in step S7, the feature data of real-time monitoring is input into the trained random forest model to obtain the probability prediction of landslide occurrence; the specific steps are as follows: S701: inputting characteristic data for real-time monitoring, including displacement rate, crack extension speed, rainfall, accelerometer data, displacement sensor data, and moisture sensor data; S702: Loading a trained random forest model from a cloud computing platform or local storage; S703: The feature data monitored in real time is used as input and passed to the input layer of the random forest model; S704: For each decision tree in the random forest, starting from the root node, according to the input feature data, the node splitting condition and the threshold, the decision tree is traversed layer by layer, wherein at each node, according to the splitting feature of the current node, it is determined which child node the input data belongs to, and the traversal is continued downward until a leaf node is reached, and the leaf node stores the prediction result of the decision tree for the input data; S705: Calculate the average of all decision tree prediction results as the final prediction result of the random forest model for the input data. For all decision trees in the random forest, collect the prediction results of each tree, that is, the probability of landslide occurrence. The calculation steps are as follows: The random forest model is set to contain N decision trees, and the prediction result of each tree is , then the final prediction result P of the random forest model is: ; in, is the prediction result of the i-th decision tree for the input data, i.e., the probability of landslide occurrence; S706: Output the predicted probability value of landslide occurrence, indicating the possibility of landslide occurrence.

[0026] Preferably, in step S8, a warning threshold is set according to historical landslide data and landslide risk assessment, and when the predicted landslide probability exceeds the threshold, a warning signal is triggered; the specific steps are as follows: S801: Collect landslide events from geological survey reports, historical records, and monitoring data sources, record the time, location, scale, triggering factors, and actual damage caused by each landslide, and correspond to each landslide event, find and record the predicted probability at that time; S802: Based on the relationship between historical landslide data and predicted probability, determine the landslide occurrence frequency corresponding to different predicted probability intervals, set risk levels for different hazard levels based on the potential hazards of landslides, including casualties, property losses, and environmental impacts, and develop a landslide risk assessment curve based on the predicted probability and hazard level; S803: According to the risk assessment curve, the prediction probability thresholds corresponding to different risk levels are determined, and multiple thresholds are set corresponding to different warning levels, including low-level warning, medium-level warning, and high-level warning; S804: Acquire the real-time monitoring characteristic data and the probability prediction value of landslide occurrence from the cloud computing platform; S805: Compare the predicted probability with the set warning threshold. If the predicted probability exceeds the threshold of a certain warning level, a warning signal of the corresponding level is triggered.

[0027] Example Optionally, it is assumed that a certain mountainous area has complex terrain, active geological structures, and frequent landslide disasters; in order to effectively monitor and prevent landslide disasters, the local government decides to adopt the potential landslide hazard identification method based on rock mass degradation characteristics provided by the present invention; the method realizes real-time monitoring and early warning of landslide hazards by comprehensively applying high-definition cameras, drones, infrared and radar sensors, combined with convolutional neural networks and random forest algorithms; the specific implementation steps are as follows: A1: Install 5 high-definition cameras in the target area of ​​landslide hazard monitoring points, with a resolution of 4K, a frame rate of 30fps, and automatic exposure time adjustment to adapt to different lighting conditions; A2: Arrange drones to conduct patrol photography twice a week, with a flight altitude of 200 meters, covering the entire mountain area, and obtaining high-resolution images; A3: Install 10 infrared sensors and 5 radar sensors at key monitoring points to monitor rock temperature and surface displacement in real time; A4: Sensor data is transmitted to the cloud computing platform in real time via 4G network; A5: Preprocess the collected image and video data, including denoising, contrast enhancement and distortion correction, to ensure data quality; A6: Build a convolutional neural network architecture, including 5 convolutional layers, 3 pooling layers and 2 fully connected layers; A7: The convolutional neural network was trained using a training set of 1,000 images of rock mass degradation, a validation set of 200 images, and a test set of 300 images. The model was trained using the cross entropy loss function and the Adam optimizer, and the accuracy of the final model on the test set reached 90%; A8: Input the preprocessed image into the trained convolutional neural network to identify rock degradation features such as cracks, spalling and discoloration; A9: Convert the identified degradation features into numerical features, such as the average crack width of 2.5 mm, the maximum peeling area of ​​1.2 square meters, and the degree of color change represented by the difference in RGB values; A10: 20 accelerometers, 15 displacement sensors and 10 moisture sensors are deployed at the landslide hazard monitoring points; A11: The sensor network is connected to the Internet of Things platform to transmit data to the cloud computing platform in real time to display the vibration, displacement change and water content information of the rock mass; A12: The cloud computing platform receives sensor data and image data in real time, performs integrity verification and error detection, and converts them into JSON format and stores them in the MySQL database. The data is partitioned and stored by timestamp and geographic location to improve query efficiency. A13: Collect historical landslide event information from the past 10 years, and select displacement rate, crack expansion speed and rainfall as characteristic variables; select data containing these characteristic variables from real-time monitoring data, and divide them into training set (70%) and test set (30%) after cleaning; standardize and normalize the characteristic variables, and use principal component analysis method to select characteristic variables that contribute more to landslide warning; the final data set is saved in CSV format and stored in the database; A14: Use the training set data to build a random forest model, including 100 decision trees, with a maximum depth of 10 and a minimum number of samples of 5. Use the 5-fold cross-validation method to adjust the parameters and optimize the model. The final model has a prediction accuracy of 85%, a recall rate of 80%, and an F1 score of 82% on the test set. A15: Input the real-time monitoring feature data into the trained random forest model to obtain the probability prediction value of landslide occurrence; for example, the probability of landslide occurrence corresponding to the feature data monitored in real time is 75%; A16: Based on historical landslide data and landslide risk assessment, the warning thresholds are set as follows: low-level warning (50%), medium-level warning (70%), and high-level warning (90%); A17: Obtain the real-time monitoring feature data and the predicted probability value of landslide occurrence, and compare them with the warning threshold; when the predicted probability is 75%, trigger the intermediate warning signal; A18: Issue intermediate landslide warning information to relevant departments and personnel via SMS, email and APP push, and recommend taking corresponding measures for prevention and response.

[0028] By implementing the potential landslide hazard identification method based on rock degradation characteristics provided by the present invention, the mountainous area successfully predicted and prevented multiple landslide disasters, effectively reducing casualties and property losses; the application of this method not only improved the accuracy and timeliness of landslide monitoring, but also provided a scientific basis for the prevention and control of landslide disasters.

[0029] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge scope of those skilled in the art without departing from the spirit of the present invention.

Claims

1. A method for identifying potential landslide hazards based on rock mass degradation characteristics; characterized by: The following steps are included: S1: Data collection: Deploy high-definition cameras and drones for aerial photography, as well as infrared and radar sensors in the target areas of landslide hazard monitoring points to capture the surface morphology, color, texture, and temperature multi-dimensional information of the rock mass, and pre-process the collected image and video data, including denoising, contrast enhancement, and distortion correction; S2: Feature extraction: A convolutional neural network is used to extract features from the preprocessed image to identify rock mass degradation features, including but not limited to rock mass cracks, spalling and discoloration, and the identified degradation features are converted into numerical features, including but not limited to crack width, spalling area and color change degree; S3: Deployment of sensor network: Sensors are arranged at the landslide hazard monitoring points. The sensors include accelerometers, displacement sensors and moisture sensors, which are used to monitor the tiny vibrations, displacement changes and moisture content of the rock mass in real time. At the same time, the sensor data is transmitted to the cloud computing platform in real time using the Internet of Things technology; S4: Data storage on cloud computing platform: Sensor data and image data are stored in cloud database; S5: Data preparation for early warning algorithm: Collect information on historical landslide events, filter out characteristic variables related to landslides from real-time monitoring data, including displacement rate, crack extension speed and rainfall, and divide the data into training set and test set; S6: Model training: Use the training set data to build multiple decision trees. Each tree is split based on a randomly selected feature subset to form a forest. The parameters of the random forest, including the number of trees and the maximum depth, are adjusted through the cross-validation method to perform module optimization. S7: Early warning prediction: Input the real-time monitoring feature data into the trained random forest model to obtain the probability prediction of landslide occurrence; S8: Set a warning threshold based on historical landslide data and landslide risk assessment, and trigger a warning signal when the predicted landslide probability exceeds the threshold; S9: Warning release: Landslide warning information will be released to relevant departments and personnel through SMS, email and APP push. The landslide warning information includes warning level and recommended measures.

2. The method for identifying potential landslide hazards based on rock mass degradation characteristics according to claim 1 is characterized in that: In step S1, high-definition cameras and drones are deployed in the target area for aerial photography, as well as infrared and radar sensors to capture the surface morphology, color, texture, and temperature multi-dimensional information of the rock mass, and the collected image and video data are pre-processed, including denoising, contrast enhancement, and distortion correction. The specific steps are as follows: S101: Install a high-definition camera at the monitoring point in the target area of ​​the landslide hazard monitoring point, and configure the camera's shooting parameters, including resolution, frame rate, and exposure time; S102: Arrange the UAV to perform regular cruise photography according to the preset flight route and altitude to obtain large-scale, high-resolution images of the rock mass; S103: Install infrared sensors at the monitoring points in the target area of ​​landslide hazard monitoring points to detect the temperature distribution and abnormal hot spots of the rock mass, and deploy radar sensors to monitor the surface displacement and micro-vibration of the rock mass; S104: The high-definition camera and the drone automatically collect image and video data according to the set time interval and trigger conditions; S105: Infrared sensors and radar sensors collect rock temperature, displacement and vibration data in real time; S106: The sensor data is transmitted to the cloud computing platform via wireless means; S107: Using a Gaussian filtering algorithm to perform denoising on the collected image and video data to remove the clutter and noise caused by device noise and environmental interference; S108: using a histogram equalization method to make the bright and dark areas in the image data more distinct, performing contrast enhancement processing on the image data, and improving the visibility of the surface morphology and degradation characteristics of the rock mass; S109: using a barrel distortion correction algorithm to correct image distortion caused by camera lens distortion and drone shooting angle problems; S110: Output the pre-processed image and video data.

3. The method for identifying potential landslide hazards based on rock mass degradation characteristics according to claim 2 is characterized in that: In step S2, a convolutional neural network is used to extract features from the preprocessed image to identify rock mass degradation features, which include but are not limited to rock mass cracks, spalling and discoloration, and the identified degradation features are converted into numerical features, which include but are not limited to crack width, spalling area and color change degree; wherein the specific steps of extracting features from the preprocessed image using a convolutional neural network are as follows: S201: constructing a convolutional neural network architecture of rock mass degradation features, including an input layer, a convolution layer, a pooling layer, a fully connected layer and an output layer; wherein the input layer receives preprocessed image data, and the output layer outputs the identified degradation feature category and location; S202: Divide the preprocessed image data into a training set, a validation set, and a test set; S203: training the convolutional neural network using the training set data, and optimizing the network parameters by a back propagation algorithm; S204: Select the cross entropy loss function and Adam optimizer to train the model, evaluate the model performance on the validation set regularly, and adjust the learning rate and batch size parameters to optimize the model; S205: Evaluate the trained model on the test set and select the model with the best performance as the feature extraction module; S206: inputting the preprocessed image into the trained convolutional neural network; S207: The image data passes through the convolution layer, the convolution kernel slides over the image and calculates the convolution operation to extract local features of the image, and multiple convolution layers gradually extract higher-level features; S208: The pooling layer downsamples the feature map output by the convolutional layer; S209: The fully connected layer flattens and connects the feature maps output by the pooling layer to form a one-dimensional feature vector; S210: The convolutional neural network outputs the identified rock mass degradation features, including but not limited to cracks, spalling, and discoloration.

4. The method for identifying potential landslide hazards based on rock mass degradation characteristics according to claim 3 is characterized in that: In step S2, a convolutional neural network is used to extract features from the preprocessed image to identify rock mass degradation features, which include but are not limited to rock mass cracks, spalling and discoloration, and the identified degradation features are converted into numerical features, which include but are not limited to crack width, spalling area and color change degree; the specific steps of converting the identified degradation features into numerical features are as follows: S211: measuring the width of the crack for the identified crack features by edge detection and contour extraction image processing methods, and recording the crack width as a numerical value; S212: For the identified spalling features, an image segmentation method is used to separate the spalling region from the background, and the area of ​​the spalling region is calculated and recorded as a numerical feature; S213: For the identified color change feature, a color space conversion and a color histogram analysis method are used to evaluate the degree of color change, and the color difference between the color change area and the surrounding normal area is calculated and recorded as a numerical feature; S214: Integrate the numerical features of crack width, spalling area and color change degree to form a complete numerical representation of rock mass degradation characteristics, and output the numerical features.

5. The method for identifying potential landslide hazards based on rock mass degradation characteristics according to claim 4 is characterized in that: In step S3, sensors are arranged at the landslide hazard monitoring points. The sensors include accelerometers, displacement sensors and moisture sensors, which are used to monitor the tiny vibrations, displacement changes and moisture content of the rock mass in real time. At the same time, the sensor data are transmitted to the cloud computing platform in real time using the Internet of Things technology. The specific steps are as follows: S301: Arrange sensors at the landslide hazard monitoring points, including accelerometers, displacement sensors, and moisture sensors. The accelerometers are used to monitor tiny vibrations of the rock mass, the displacement sensors are used to monitor displacement changes of the rock mass, and the moisture sensors are used to monitor the water content of the rock mass. S302: Connect the sensor network to the Internet of Things platform and equip each sensor node with a unique identifier; S303: The data collected by the sensor node is transmitted to the cloud computing platform and the Internet of Things platform in real time through the Internet of Things communication method; S304: Real-time monitoring of sensor data through the IoT platform to display the vibration, displacement changes and water content information of the rock mass.

6. The method for identifying potential landslide hazards based on rock mass degradation characteristics according to claim 5 is characterized in that: In step S4, the sensor data and image data are stored in a cloud database; the specific steps are as follows: S401: The cloud computing platform receives sensor data and pre-processed image data from landslide hazard monitoring points in real time through the Internet of Things communication technology; S402: using a check code and data packet size check mechanism to mark and discard abnormal data, so as to perform integrity check and error detection on the received data; S403: Convert the received raw data into JSON format; S404: Selecting the relational database MySQL to construct a database table structure for the sensor data and the image data, including field names, data types, and indexes; S405: partition and store the data according to the timestamp and geographic location information of the data; S406: Write the preprocessed data into the database, and create an index for the database table through the B-tree index.

7. The method for identifying potential landslide hazards based on rock mass degradation characteristics according to claim 6 is characterized in that: In step S5, information on historical landslide events is collected, characteristic variables related to landslides are screened out from real-time monitoring data, the characteristic variables include displacement rate, crack extension speed and rainfall, and the data is divided into a training set and a test set; the specific steps are as follows: S501: Collect historical landslide event information through geological survey reports, remote sensing monitoring data and ground observation records. The historical landslide event information should include the time, location, scale, impact range and triggering factors of the landslide. The historical landslide event information should be sorted out to form a standardized data set; S502: Determine characteristic variables related to landslides based on historical landslide event information, including displacement rate, crack extension speed, and rainfall; S503: Filtering data containing the above-mentioned characteristic variables from the real-time monitoring data; S504: Clean the screened data to remove abnormal values ​​and missing values; S505: Divide the cleaned data into a training set and a test set, wherein the training set is used to build and train the early warning algorithm model, and the test set is used to evaluate the performance of the model; S506: Standardizing the feature variables so that they have the same scale, and normalizing the feature variables so that the range of the feature variables is between [0, 1] or [-1, 1]; S507: using the principal component analysis method to select characteristic variables that contribute more to landslide early warning according to the correlation between historical landslide event information and real-time monitoring data; S508: Integrate the preprocessed and divided data into a final data set; S509: Save the data set in CSV format and store it in a database.

8. The method for identifying potential landslide hazards based on rock mass degradation characteristics according to claim 7 is characterized in that: In step S6, multiple decision trees are constructed using the training set data, each tree is split based on a randomly selected feature subset to form a forest, and the parameters of the random forest, including the number of trees and the maximum depth, are adjusted by a cross-validation method to perform module optimization. The specific steps are as follows: S601: Randomly extract multiple sample subsets from the preprocessed training set data, and each sample subset will be used to train a decision tree; S602: During the training process of each decision tree, a feature subset is randomly selected as a split feature of the current node; S603: For each node, according to the selected feature subset, an optimal feature and threshold are selected for splitting, so that the purity of the subset after splitting reaches the maximum. This process is recursively performed until the stopping condition is met. The stopping condition includes that the number of samples contained in the node is less than a certain threshold and the purity of the node reaches a certain standard; S604: generating a complete decision tree by continuously splitting nodes, and the decision tree is used to output corresponding landslide hazard prediction results according to the input feature values; S605: Repeat steps S601-S604 to construct multiple decision trees to form a random forest model; S606: Divide the training set data into K subsets in K-fold cross validation, each subset will be used as a validation set in turn, and the remaining subsets will be used as training sets; S607: Setting the initial parameters of the random forest model, including the number of trees, maximum depth, and minimum number of samples; S608: For each round of cross-validation, the random forest model is trained using the training set data of the current round, and the trained model is evaluated using the validation set data to calculate the prediction accuracy, recall rate, and F1 score; S609: Based on the evaluation results on the validation set, analyze the impact of different parameters on the model performance, use the Bayesian optimization method to adjust the parameters of the random forest, and optimize the model.

9. The method for identifying potential landslide hazards based on rock mass degradation characteristics according to claim 8, characterized in that: In step S7, the real-time monitored feature data is input into the trained random forest model to obtain a probability prediction of landslide occurrence; the specific steps are as follows: S701: inputting characteristic data for real-time monitoring, including displacement rate, crack extension speed, rainfall, accelerometer data, displacement sensor data, and moisture sensor data; S702: Loading a trained random forest model from a cloud computing platform or local storage; S703: The feature data monitored in real time is used as input and passed to the input layer of the random forest model; S704: For each decision tree in the random forest, starting from the root node, according to the input feature data, the node splitting condition and the threshold, the decision tree is traversed layer by layer, wherein at each node, according to the splitting feature of the current node, it is determined which child node the input data belongs to, and the traversal is continued downward until a leaf node is reached, and the leaf node stores the prediction result of the decision tree for the input data; S705: Calculate the average of all decision tree prediction results as the final prediction result of the random forest model for the input data. For all decision trees in the random forest, collect the prediction results of each tree, that is, the probability of landslide occurrence. The calculation steps are as follows: The random forest model is set to contain N decision trees, and the prediction result of each tree is , then the final prediction result P of the random forest model is: ； in, is the prediction result of the i-th decision tree for the input data, i.e., the probability of landslide occurrence; S706: Output the predicted probability value of landslide occurrence, indicating the possibility of landslide occurrence.

10. The method for identifying potential landslide hazards based on rock mass degradation characteristics according to claim 9, characterized in that: In step S8, a warning threshold is set according to historical landslide data and landslide risk assessment, and a warning signal is triggered when the predicted landslide probability exceeds the threshold. The specific steps are as follows: S801: Collect landslide events from geological survey reports, historical records, and monitoring data sources, record the time, location, scale, triggering factors, and actual damage caused by each landslide, and correspond to each landslide event, find and record the predicted probability at that time; S802: Based on the relationship between historical landslide data and predicted probability, determine the landslide occurrence frequency corresponding to different predicted probability intervals, set risk levels for different hazard levels based on the potential hazards of landslides, including casualties, property losses, and environmental impacts, and develop a landslide risk assessment curve based on the predicted probability and hazard level; S803: According to the risk assessment curve, the prediction probability thresholds corresponding to different risk levels are determined, and multiple thresholds are set corresponding to different warning levels, including low-level warning, medium-level warning, and high-level warning; S804: Acquire the real-time monitoring characteristic data and the probability prediction value of landslide occurrence from the cloud computing platform; S805: Compare the predicted probability with the set warning threshold. If the predicted probability exceeds the threshold of a certain warning level, a warning signal of the corresponding level is triggered.

Images