Deep learning based unstable slope identification and detection method

Abstract

The application provides a kind of based on deep learning's unstable slope feature recognition and detection method.The method successively includes:S1.from the image data of the slope region in multi-source data is collected, and the standard data set is constructed after pre-processing;S2.SlopeNet model based on convolutional neural network is constructed and trained;S3.the scope and boundary of unstable slope region are identified by using the SlopeNet model trained in S2 step;S4.based on the scope and boundary of unstable slope region obtained in S3 step, combined with historical meteorological, hydrological, geological monitoring data, an intelligent analysis model of unstable slope evolution trend is established;S5.the unstable slope evolution trend, i.e., the instability probability, in a future period of time is predicted by using the intelligent analysis model of unstable slope evolution trend trained in S4 step.The application realizes the automatic identification and detection of unstable slope by the combination of multi-source data fusion, convolutional neural network and long short-term memory network, with high precision and efficiency.

Description

Technical Field

[0001] This invention belongs to the field of slope engineering, specifically a method for feature recognition and detection of unstable slopes based on deep learning SlopeNet. Background Technology

[0002] Slope stability is a critical issue in geotechnical engineering. The formation and evolution of unstable slopes not only affect the safety of engineering projects but can also pose significant threats to the environment and human life and property. Traditional slope stability assessment methods rely on manual surveys, geological investigations, and physical model tests. While these methods provide some reference, they suffer from low efficiency and the inability to monitor in real time.

[0003] With the rapid development of data acquisition technologies such as remote sensing, drones, and lidar, it has become possible to perform slope feature analysis by combining large-scale datasets and advanced computing methods. In recent years, deep learning, as a powerful data analysis tool, has achieved remarkable results in fields such as image processing, speech recognition, and natural language processing. However, the potential of deep learning has not yet been fully utilized for the automated identification and detection of unstable slope features, especially for unstable slopes under complex terrain conditions.

[0004] Therefore, the slope instability feature identification and detection method based on deep learning can automatically extract the unstable features of slopes from multi-source data such as remote sensing images and lidar point cloud data, and achieve rapid and accurate identification and detection through deep neural network models, and automatically generate a complete exploration plan. This is a technical problem that urgently needs to be solved in the field of geotechnical engineering exploration. Summary of the Invention

[0005] This invention addresses the problems existing in the prior art by providing a deep learning-based method for identifying and detecting unstable slope features. The method aims to automatically extract unstable slope features from multi-source data through deep learning technology, thereby achieving efficient, accurate, and intelligent slope stability assessment.

[0006] To achieve the above technical objectives, this invention provides a method for identifying and detecting unstable slope features based on deep learning, specifically including the following steps:

[0007] S1. Collect image data of the slope area from multi-source data, and preprocess the collected image data to construct a standard dataset for training, validation and testing of deep learning models;

[0008] S2. Construct and train a SlopeNet model based on a convolutional neural network. The SlopeNet model integrates multi-scale feature extraction, deep residual connections, adaptive attention mechanism, and slope feature enhancement module. By designing an adaptive loss function, it enhances the learning of features in unstable slope regions. The SlopeNet model is trained end-to-end using the standard dataset preprocessed in step S1 to achieve automated extraction of unstable slope features and image segmentation, thereby completing a pixel-level classification model.

[0009] S3. Using the SlopeNet model trained in step S2, reason about new or unlabeled remote sensing image data to achieve pixel-level classification and image segmentation of slope areas, thereby accurately identifying the range and boundaries of unstable slope areas.

[0010] S4. Based on the range and boundary of the unstable slope area obtained in step S3, and combined with historical meteorological, hydrological and geological monitoring data, a time series dataset is constructed. The Long Short-Term Memory (LSTM) time series deep learning model is used to train the time series dataset to establish an intelligent analysis model for the evolution trend of unstable slopes.

[0011] S5. Using the intelligent analysis model for the evolution trend of unstable slopes trained in step S4, input the current and recent spatial characteristics of unstable slopes and historical monitoring data sequences to predict the evolution trend of unstable slopes in the future, i.e., the probability of instability.

[0012] A further technical solution of the present invention: The method further includes the following steps: outputting the identification results in step S3 and the prediction results in step S5 in a visual manner to provide intuitive evaluation and early warning information.

[0013] A further technical solution of the present invention: The multi-source data in step S1 includes UAV image data, satellite remote sensing image data, geological exploration data, and meteorological and historical monitoring data; preprocessing includes data cleaning, format conversion, geometric correction, spatial registration, noise removal, image enhancement, data standardization, accurate labeling of unstable slope areas, data enhancement, multi-source data fusion, and dataset partitioning; wherein the accurate labeling of unstable slope areas is based on the slope, aspect, texture, and curvature characteristics of remote sensing images, lidar point clouds, and UAV aerial image data, observed and analyzed, and GIS software is used to draw polygons for possible landslides or unstable slope areas, marking the unstable slope areas.

[0014] A further technical solution of the present invention: the automated extraction process of unstable slope features in step S2 specifically includes:

[0015] S201. Multi-scale convolutional feature extraction: Multiple convolutional layers are used to extract multi-scale local features of the image across different receptive fields;

[0016] S202. Multi-scale feature fusion: Combine pooling layers to downsample the local feature maps extracted in step S201, use max pooling to reduce the image size, and extract high-level abstract information of the image;

[0017] S203. Deep Residual Connections: Borrowing from the idea of ​​ResNet, skip connections are introduced into the network; each residual block contains two convolutional layers and one skip connection;

[0018] S204. Design an adaptive attention mechanism, introduce a spatial attention module at different stages of the feature map, and dynamically adjust the importance weights of features at different spatial locations according to the saliency of the feature map, so that the model focuses more on the regions and features related to unstable slope discrimination.

[0019] S205. Slope Feature Enhancement Module: This module integrates prior geological information into the deep learning feature learning process to enhance the extraction of key local details of unstable slopes.

[0020] A further technical solution of the present invention: In step S2, during the automated extraction of unstable slope features, an adaptive loss function is used to focus on unstable slope regions, thereby improving the accuracy and robustness of feature extraction; this adaptive loss function comprises three parts: cross-entropy loss, regularization loss, and slope region loss, as follows:

[0021] L total = αL ce +βL reg +γL instability ;

[0022] Where: L ce This is the standard cross-entropy loss, used for classification;

[0023] L reg This is the regularization loss, used to prevent the model from overfitting;

[0024] L instability This is a special loss function for unstable slope regions. This loss function depends on the complexity of the slope region and can be adaptively adjusted according to the salience of the slope region.

[0025] In the loss function, α, β, and γ are hyperparameters used to adjust the weights of the three parts of the loss. Generally speaking, the value of γ should be large to enhance attention to unstable slope regions.

[0026] A further technical solution of the present invention: Step S2 uses the standard dataset preprocessed and divided in step S1 to perform end-to-end training of the SlopeNet model. The process employs an appropriate optimizer Adam and a learning rate scheduling strategy, with the goal of minimizing the adaptive weighted loss function. During training, a validation set is used to monitor model performance and adjust hyperparameters to prevent overfitting. Finally, the trained SlopeNet model weights are obtained.

[0027] A further technical solution of the present invention: The process of identifying the range and boundary of the unstable slope region based on the SlopeNet model in step S3 specifically includes:

[0028] S301. Data Input: Input slope area image data that has been preprocessed in step S1 but is not labeled, and is in the same format as the training data.

[0029] S302. Model Inference: The input data in S301 is fed into the trained SlopeNet model for forward propagation calculation. The model outputs a probability map of each pixel belonging to an unstable category.

[0030] S303. Post-processing: Post-process the output probability map by binarizing it with a threshold to generate a binary mask, where a pixel value of 1 indicates an unstable slope region and a pixel value of 0 indicates a stable region; use set operations to obtain a smooth and reasonable segmentation result.

[0031] S304. Output Results: Output a pixel-level binary segmentation map and vector polygon, representing the identification results of the range and boundary of the unstable slope region.

[0032] A further technical solution of the present invention: The process of establishing an intelligent analysis model of the evolution trend of an unstable slope based on LSTM in step S4 specifically includes:

[0033] S401. Historical Spatial Feature Extraction: For slope regions with time series data, repeat steps S1 and S3, and use the trained SlopeNet model to identify and label unstable slope regions at different historical time points; then, extract quantified spatial features from these time-varying region segmentation results.

[0034] S402. Historical monitoring data synchronization: Collect historical monitoring data synchronized with the time points of the above spatial feature extraction;

[0035] S403. Time series construction: The spatial feature vector extracted from each historical time point is concatenated with the corresponding historical monitoring data vector to form a comprehensive feature vector. These comprehensive feature vectors are arranged in chronological order to construct a multivariate time series dataset.

[0036] S404. Intelligent Analysis Model Design for Unstable Slope Evolution Trend: The intelligent analysis model for unstable slope evolution trend adopts an LSTM model architecture, including an input layer, LSTM layer, Dropout layer, fully connected layer, and output layer. The input layer receives multivariate time series data expanded step by step. The LSTM unit contains a gating mechanism, which can effectively learn and remember long-term dependency information. The Dropout layer is used to randomly deactivate some neurons between LSTM layers to prevent overfitting. The fully connected layer is responsible for mapping the output of the LSTM layer to the final prediction target. The output layer uses the sigmoid activation function to output probabilities to predict whether the slope will become unstable in the future.

[0037] A further technical solution of the present invention: the intelligent analysis model for the evolution trend of unstable slopes in step S5 outputs the prediction results and the identification results in step S3;

[0038] The prediction results are based on data from multiple past time steps, predicting the slope state at a future time point and the slope state at future time points in the sequence. The prediction result output is a time series, and the model outputs probability values ​​between 0 and 1 at multiple time points, representing the slope evolution trend and its probability of instability over a period of time, where 0 indicates no instability and 1 indicates instability.

[0039] The recognition results in step S3 include outputting a pixel-level binary mask and unstable region vector boundaries, representing the spatial extent of the unstable slope region, which can be superimposed on the original image or DEM for display.

[0040] The prediction results output by the intelligent analysis model for unstable slope evolution trends and the identification results in step S3 are both displayed through visualization. Specifically, this includes: overlaying the vector boundaries of the identified unstable areas onto the 3D terrain model for visualization; using heatmaps or hierarchical color schemes to overlay and display the pixel-level predicted probabilities from the SlopeNet identification results on the slope map; and using time series plots or bar charts to display the changes in historical unstable area areas, key monitoring data curves, and the future instability probability curve of the intelligent analysis model for unstable slope evolution trends over time, providing dynamic evolution trend information. The predicted probabilities are combined with preset warning thresholds, and different colors or markers are used to represent different warning levels.

[0041] The preferred technical solution of the present invention is as follows: the spatial features extracted in S401 include: the area of ​​each unstable region, the shape features of the unstable region (i.e., the aspect ratio), the topographic feature statistics within the unstable region, and the change of the unstable region over time; the topographic feature statistics within the unstable region include the average slope, the maximum slope, the surface roughness, and the undulation; the change of the unstable region over time includes the area change and the boundary change; the historical monitoring data in S402 includes accumulated precipitation, groundwater level, and the displacement of measuring points.

[0042] Compared with the prior art, the present invention has the following significant advantages:

[0043] (1) The present invention has high precision and robustness: through deep fusion of multi-source data and a specially designed SlopeNet model (including multi-scale feature extraction, attention mechanism, targeted feature enhancement module and adaptive weighted loss), it can capture the subtle features and complex patterns of unstable slopes more comprehensively and accurately, significantly improving the pixel-level recognition accuracy of unstable slope areas and the robustness in complex geological environments; the adaptive loss function particularly enhances the model's learning ability for high-risk and complex areas.

[0044] (2) The present invention is automated and efficient: the entire identification and detection process is highly automated, reducing reliance on human experience and heavy labor, greatly improving the efficiency of slope survey and monitoring, and is particularly suitable for rapid screening of large areas.

[0045] (3) The present invention has pixel-level fine recognition: SlopeNet achieves pixel-level classification based on the idea of ​​image segmentation, which can accurately depict the boundary and range of unstable slope areas, providing accurate spatial information for subsequent detailed exploration, monitoring and management.

[0046] (4) This invention can realize intelligent evolution trend prediction: It creatively combines the quantitative features extracted from the historical spatial identification results with traditional monitoring data as the input of the time series model LSTM, realizing intelligent analysis and probability prediction of the future evolution trend of unstable slopes, providing the possibility of early warning, and moving from "identifying the current situation" to "predicting the future".

[0047] (5) The present invention has strong adaptability: the deep learning model has strong generalization ability and after training with diverse data, it has good adaptability to slopes of different types and in different environments. Detailed Implementation

[0048] The present invention will be further described below with reference to embodiments. The technical solutions presented in the embodiments are specific solutions of the embodiments of the present invention and are not intended to limit the scope of the claimed invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] The embodiment provides a method for identifying and detecting unstable slope features based on deep learning, which specifically includes the following steps:

[0050] S1. Collect image data of the slope area from multiple sources (UAV image data, satellite remote sensing image data, geological exploration data), and perform comprehensive preprocessing on the collected data, including data cleaning, format conversion, geometric correction, spatial registration, noise removal, image enhancement, data standardization, accurate manual annotation of unstable slope areas (as training labels for deep learning models), data augmentation, and multi-source data fusion, and finally construct a standard dataset for training, validating and testing deep learning models.

[0051] Satellite remote sensing image data acquisition: Use high-resolution satellite imagery (such as Google Earth Engine, Sentinel-2, etc.) to acquire two-dimensional ground images of the slope area; the acquired images must cover the entire range of the slope, including potential landslide areas, crack zones, and adjacent topographic features;

[0052] Drone Image Data Acquisition: Aerial photography of slope areas is conducted using drones equipped with high-resolution cameras. Drones can provide low-altitude, high-resolution images, making them particularly suitable for inaccessible areas. Drones can also acquire multi-dimensional data such as 3D point cloud data and vegetation cover of slope areas by carrying LiDAR or multispectral cameras.

[0053] Geological exploration data acquisition: Combining traditional geological exploration data, such as soil type, groundwater level, and rock and soil samples, and through methods such as surface drilling and geological profile analysis, we can assist in assessing slope stability.

[0054] Meteorological and historical monitoring data: Collect meteorological data (such as precipitation, temperature, and wind speed) and monitoring data on historical landslide events. This data is of great value for understanding the dynamic changes of slopes and the triggering conditions for landslides.

[0055] After data collection in step S1, the raw data needs to be preprocessed so that the deep learning model can efficiently extract useful features from it. The preprocessing steps specifically include the following steps:

[0056] Data cleaning: Remove irrelevant data and outliers to ensure the quality of input data; for example, remove cloud cover, haze, and other elements that significantly affect remote sensing images.

[0057] Format conversion: Converting data from different sources into a unified format to facilitate processing by deep learning models; for example, converting remote sensing images from different formats (such as .tif, .jpg) into standard input formats such as PNG or JPEG.

[0058] Image geometric correction: For remote sensing images, geometric correction is required to ensure that the slope areas in the image correspond to the actual coordinates on the ground; this step includes image registration, geographic coordinate transformation and image distortion correction.

[0059] Data registration: Multi-source data (such as remote sensing images and UAV images) need to be spatially registered, that is, the different data sources are converted to the same coordinate system in order to facilitate subsequent fusion and analysis.

[0060] Noise Removal: Remote sensing images are often affected by factors such as weather, sensor accuracy, and changes in lighting, resulting in noise in the data; filtering algorithms (such as median filtering, Gaussian filtering, etc.) can be used to remove noise from the data.

[0061] Image enhancement: To improve image quality, image enhancement methods such as histogram equalization and contrast stretching are used. By enhancing the contrast of slope features (such as cracks and landslide signs), deep learning models can better identify these key features.

[0062] Data standardization: Data from different sources may have different dimensions and scales, and must be standardized. For example, the mean-variance standardization method can be used to convert all input features into a uniform scale range (such as between 0 and 1) to ensure that the training process of deep learning models is not affected by differences in data scale.

[0063] Unstable slope labeling: Manual labeling is used. Based on the slope, aspect, texture, curvature, and other characteristics of data such as remote sensing imagery, lidar point clouds, and UAV aerial images, the slope is observed and analyzed. Using GIS software such as ArcGIS, QGIS, and ENVI, polygons are drawn for possible landslides or unstable slope areas, and unstable slope areas are manually labeled.

[0064] Data augmentation: To enhance the robustness of the model, data augmentation techniques are used to perform transformations such as rotation, translation, cropping, and scaling on the original image; these transformations can generate diverse training samples, helping the model to generalize better.

[0065] Multi-source data fusion: Data from different sources such as remote sensing images and drone aerial photography are fused to form a comprehensive dataset that includes spatial information (such as slope surface topography), temporal information (such as meteorological data and precipitation), and three-dimensional information (slope, aspect, etc.), while ensuring that the coordinates of all input data are aligned.

[0066] Dataset partitioning: The preprocessed dataset is divided into a training set, a validation set, and a test set, with a common partition ratio of 70%:15%:15%. The model is trained using the training set, hyperparameters are tuned using the validation set, and the model performance is finally evaluated using the test set.

[0067] S2. Construct and train a SlopeNet model based on a convolutional neural network (CNN). This model integrates multi-scale feature extraction, deep residual connections, adaptive attention mechanisms, and feature enhancement modules. An adaptive loss function is designed to enhance the learning of features in unstable slope regions. The SlopeNet model is trained end-to-end using the standard dataset preprocessed in step S1, resulting in a model capable of automatically extracting unstable slope features, performing image segmentation, and ultimately achieving pixel-level classification. This invention proposes a "SlopeNet" deep learning algorithm for the automated extraction of unstable slope features. This algorithm combines convolutional neural networks (CNN) and multi-scale fusion techniques, and uses an adaptive loss function to focus on unstable slope regions, thereby improving the accuracy and robustness of feature extraction. The automated extraction process of unstable slope features specifically includes:

[0068] S201. Multi-scale Convolutional Feature Extraction Block: Employs multiple convolutional layers to extract multi-scale local features of the image across different receptive fields; by using convolutional kernels of different sizes (such as 3x3, 5x5, 7x7, etc.) to perform convolution operations at different scales, it can capture subtle changes in slope terrain (such as features like soil landslides and rock debris accumulation).

[0069] Input: Feature map from the previous layer (size H′×W′×C′).

[0070] Output: Feature maps output by the convolutional layer (size H′×W′×C) out C out (Number of output channels for the convolutional layer)

[0071] S202. Multi-scale feature fusion module: Combines pooling layer to downsample the local feature map extracted in step S201, uses max pooling to reduce image size, and extracts high-level abstract information of the image;

[0072] Input: Feature maps output by the convolutional layer (size H′×W′×C) out );

[0073] Output: Pooled feature map (size is...) × ×C out Furthermore, a feature fusion layer uses 1×1 convolutions to fuse information from different scales, achieving multi-scale information fusion to preserve a wider range of contextual information.

[0074] Input: Feature maps at multiple scales (sizes H′×W′×C1, H′×W′×C1, ..., H′×W′×C1) 2， …); Output: The fused feature map (size H′×W′×C) fused ).

[0075] S203. Deep Residual Connections: Borrowing from ResNet, skip connections are introduced into the network. Residual connections allow information to be directly passed across layers, which helps in training deeper networks, alleviates the vanishing gradient problem, accelerates convergence, and improves model performance. Each residual block contains two convolutional layers and one skip connection;

[0076] Input: Input feature map for each residual block (size H′×W′×Cresidual);

[0077] Output: Feature map after residual learning (size H′×W′×Cresidual).

[0078] S204. Adaptive Attention Mechanism: This mechanism introduces spatial attention modules at different stages of the feature map. Based on the saliency of the feature map (regions associated with unstable slopes typically have higher saliency, such as obvious morphological changes or texture anomalies), the attention mechanism dynamically adjusts the importance weights of features at different spatial locations, allowing the model to focus more on regions and features related to unstable slope discrimination.

[0079] S205. Feature Enhancement Modules integrate prior geological information into the deep learning feature learning process. This module enhances the extraction of key local details of unstable slopes, such as potential cracks, terraces, vegetation anomalies, and water confluence.

[0080] In step S2, during the automated extraction of unstable slope features, an adaptive loss function is used to focus on unstable slope regions, thereby improving the accuracy and robustness of feature extraction. This adaptive loss function consists of three parts: cross-entropy loss, regularization loss, and slope region loss.

[0081] Loss function: L total = αL ce +βLreg +γL instability

[0082] Where: L ce : Conventional cross-entropy loss, used for classification.

[0083] L reg Regularization loss is used to prevent the model from overfitting.

[0084] L instability This is a special loss function for unstable slope regions. This loss function depends on the complexity of the slope region and can be adaptively adjusted according to the saliency of the slope region. For example, saliency can be calculated based on the slope variation, texture complexity, etc.

[0085] In the loss function, α, β, and γ are hyperparameters used to adjust the weights of the three parts of the loss. Generally, the value of γ should be relatively large to enhance attention to unstable slope regions.

[0086] Input: The input data consists of three parts: (1) Predicted output: The output of the network is a predicted probability distribution map (size H′×W′×2), which represents the probability of each pixel belonging to an unstable slope and a stable slope; (2) Ground Truth: The ground truth is a label map with size H′×W′, where each pixel value represents the true category of the pixel; (3) Instability Map: The instability map dynamically adjusts the loss function by calculating the complexity of the slope region in the image (such as slope, aspect change, etc.); the size of the instability map is the same as that of the input image (H′×W′).

[0087] Output: The final total loss value, reflecting the model's training error, which serves as the optimization target.

[0088] S3. Using the SlopeNet model trained in step S2, perform inference (forward propagation) on new or unlabeled remote sensing image data to achieve pixel-level classification and image segmentation of slope areas, thereby accurately identifying the range and boundaries of unstable slope areas.

[0089] The training of the model specifically includes:

[0090] (1) Data input:

[0091] Training data: A set of labeled images (size H×W×C) and their corresponding labels (size H×W);

[0092] Training batch: The images are divided into small batches, each containing multiple samples.

[0093] (2) Training process: Forward Propagation:

[0094] Input: Image data for each batch (size B×H×W×C, where B is the batch size);

[0095] Output: Prediction results (size B×H′×W′×N).

[0096] (3) Loss calculation:

[0097] Input: Predicted output and true labels;

[0098] Output: The calculated loss value (in bytes).

[0099] (4) Backward Propagation: By calculating the gradient of the loss function, the error is backpropagated and the network parameters are updated.

[0100] (5) Optimization algorithm: The Adam optimizer is used for training. The gradient is calculated based on the loss value and the network weights are updated. Input: The calculated gradient and the current network parameters. Output: The optimized network parameters.

[0101] (6) Model evaluation: Use evaluation metrics such as accuracy, precision, recall, and F1 score to evaluate the model’s performance on the test set.

[0102] Optimized network model: After multiple rounds of training, the final optimized model is obtained, which can accurately extract slope features from remote sensing images and identify unstable slope areas, specifically including:

[0103] The input consists of unlabeled slope region images preprocessed in step S1, consistent with the training data format. The input data is fed into a trained SlopeNet model for forward propagation, and the model outputs a probability map indicating the unstable category for each pixel. The output probability map is post-processed by binarizing it using a threshold, generating a binary mask where a pixel value of 1 indicates an unstable slope region and a pixel value of 0 indicates a stable region. Set operations are then used to obtain a smooth and reasonable segmentation result. Finally, a pixel-level binary segmentation map and a vector polygon are output, representing the extent and boundary identification of the unstable slope region.

[0104] S4. Based on the range and boundaries of the unstable slope area obtained in step S3, and combined with historical meteorological, hydrological, and geological monitoring data, a time series dataset is constructed. The Long Short-Term Memory (LSTM) time series deep learning model is used to train the time series dataset to establish an intelligent analysis model for the evolution trend of unstable slopes. This process will be divided into data preparation, model design, and training.

[0105] (1) Data Preparation: In previous work, the Slopenet network was used to segment the slope images, obtaining label images of stable and unstable regions of the slope. We need to extract the spatial features of unstable slope regions from these labels. The segmentation result obtained by Slopenet is usually a binary image, where each pixel represents whether a certain area of ​​the slope is stable or not. Stable regions are labeled with 1, and unstable regions are labeled with 0. The spatial features of these unstable slope regions can be further analyzed by extracting morphological features, slope changes, crack distribution, soil composition changes, etc., to construct spatial feature vectors, which will serve as the input for the subsequent LSTM model.

[0106] Historical monitoring data related to unstable slopes will be collected. This data will be used to train an LSTM network, primarily including:

[0107] Slope displacement data: Displacement data that can be measured by devices such as GPS and inclinometers.

[0108] Precipitation: Historical precipitation data, reflecting the impact of precipitation on slope stability.

[0109] Seismic activity data: earthquake events and their intensity, reflecting the impact of external geological factors on slope stability.

[0110] Temperature data: If applicable, changes in air temperature may also affect slope stability.

[0111] Other geological data, such as groundwater level and ground vibration.

[0112] These monitoring data are arranged in chronological order to form a time series, which serves as the input to the LSTM network. The spatial characteristics of the unstable slope area are combined with historical monitoring data to ensure that the timestamps of these two types of data are aligned. In particular, in the input of the time series data, it is ensured that the monitoring data at each time point is consistent with the spatial characteristics of the corresponding time period.

[0113] (2) Design of intelligent analysis model for the evolution trend of unstable slope: The prediction model adopts the LSTM model architecture, and the input data should include time series data and spatial features.

[0114] For each time step t, the input data can be a vector containing the following: a. Historical monitoring data: such as displacement, precipitation, seismic activity, etc. b. Spatial features: spatial feature vectors of unstable regions segmented from SlopeNet. The time-series data and spatial features are concatenated to form an input vector containing multiple features.

[0115] The LSTM model architecture includes an input layer, an LSTM layer, a Dropout layer, a fully connected layer, and an output layer.

[0116] The size of the input layer depends on the dimensionality of the historical data. For example, if prediction is made using data from 10 time steps, each containing several features (such as displacement, precipitation, etc.), then the dimension of the input layer is (10, n), where n is the number of features (e.g., 4 features).

[0117] LSTM layer 1: Contains 64 units and uses the tanh activation function. The goal of the LSTM layer is to capture long-term dependencies in time series data.

[0118] LSTM Layer 2: Stack another LSTM layer to enhance the model's representational power. It can be set to 32 units.

[0119] Dropout layer: Add a Dropout layer (e.g., 0.2 or 0.3) after each LSTM layer to prevent overfitting.

[0120] Fully connected layer (Dense Layer): The output of the LSTM layer is passed to a fully connected layer to further extract high-level features; the size of the output layer is consistent with the prediction target. For regression problems, the output layer has one node; for classification problems, the output layer has multiple nodes (e.g., unstable / stable).

[0121] The output layer uses either the sigmoid or softmax activation function (depending on whether the problem is regression or classification); for regression problems, the output is the predicted value of the slope displacement; for classification problems, the output is the probability of whether an instability event will occur in the future.

[0122] Model hyperparameters

[0123] Number of LSTM layers: You can choose 1-2 LSTM layers, depending on the complexity of the data.

[0124] LSTM cell count: 64, 128, etc.; can be adjusted according to the task.

[0125] Learning rate: The learning rate can be set to 0.001 to use the Adam optimizer.

[0126] Batch Size: Generally 32 or 64, the specific value is adjusted according to the amount of training data.

[0127] Training epochs: For example, train for 50-100 epochs, depending on the model's performance on the validation set.

[0128] (3) Model training process: The data is divided into training and validation sets, with a typical ratio of 80% for training and 20% for validation. The validation set is used to monitor whether the model is overfitting and to adjust hyperparameters. Historical monitoring data and spatial features of unstable slopes are used as inputs for model training. Model parameters are optimized by minimizing prediction errors (such as mean squared error (MSE) or cross-entropy loss). Grid search or random search is used to adjust hyperparameters such as learning rate, number of LSTM units, and Dropout ratio to find the optimal model.

[0129] S5. Using the intelligent analysis model for the evolution trend of unstable slopes trained in step S4, input the current and recent spatial characteristics of unstable slopes and historical monitoring data sequences to predict the evolution trend of unstable slopes in the future, i.e., the probability of instability.

[0130] Model Prediction and Application: Using a pre-trained intelligent analysis model for unstable slope evolution trends, new historical monitoring data and spatial characteristics are input, and the model will output the slope stability evolution trend for a future period. For example:

[0131] Predict future slope displacement values;

[0132] Predict whether a slope instability event may occur in the future.

[0133] Based on the prediction results output by the model, a threshold can be set to determine whether the slope is about to become unstable. If the predicted value exceeds the threshold (such as excessive slope displacement or excessive rainfall), the corresponding early warning system can be activated.

[0134] Use mean squared error (MSE) or accuracy (if it's a classification problem) to evaluate the performance of the predictions. The model can be further optimized through cross-validation.

[0135] The model's final output includes the identification results from step S3 and the prediction results from step S5, both of which are displayed visually to provide intuitive evaluation and early warning information.

[0136] The prediction results are based on data from multiple past time steps, predicting the slope state at a future time point and the slope state at future time points in the sequence. The prediction result output is a time series, and the model outputs probability values ​​between 0 and 1 at multiple time points, representing the slope evolution trend and its probability of instability over a period of time, where 0 indicates no instability and 1 indicates instability.

[0137] The recognition results in step S3 include outputting a pixel-level binary mask and unstable region vector boundaries, representing the spatial extent of the unstable slope region, which can be superimposed on the original image or DEM for display.

[0138] The prediction results output by the intelligent analysis model for unstable slope evolution trends and the identification results in step S3 are both displayed through visualization. Specifically, this includes: overlaying the vector boundaries of the identified unstable areas onto the 3D terrain model for visualization; using heatmaps or hierarchical color schemes to overlay and display the pixel-level predicted probabilities from the SlopeNet identification results on the slope map; and using time series plots or bar charts to display the changes in historical unstable area areas, key monitoring data curves, and the future instability probability curve of the intelligent analysis model for unstable slope evolution trends over time, providing dynamic evolution trend information. The predicted probabilities are combined with preset warning thresholds, and different colors or markers are used to represent different warning levels.

[0139] For example: Using a bar chart to display the classification results at each time step, the table below shows the prediction results for each future week:

[0140]

[0141] The predicted probability values ​​can also be displayed as a bar chart, where each bar at a given time point represents the probability of slope instability at that time point. The predicted probability values ​​can be directly compared to a threshold (e.g., 0.7). If the probability is greater than the threshold, it can be displayed as "instability"; if it is less than the threshold, it can be displayed as "stable".

[0142] This invention proposes a deep learning-based method for identifying and detecting unstable slope features. Through the fusion of multi-source data and the combination of convolutional neural networks and long short-term memory networks, it achieves automated identification and detection of unstable slopes. This method boasts high accuracy and efficiency, providing effective support for geotechnical engineering investigation, landslide disaster prediction, environmental protection, and other fields, and has broad application prospects and practical value.

Claims

1. A method for feature recognition and detection of unstable slopes based on deep learning, characterized in that, Specifically, the following steps are included: S1. Collect image data of the slope area from multi-source data, and preprocess the collected image data to construct a standard dataset for training, validation and testing of deep learning models; S2. Construct and train a SlopeNet model based on a convolutional neural network. The SlopeNet model integrates multi-scale feature extraction, deep residual connections, adaptive attention mechanism and slope feature enhancement module, and strengthens the learning of features of unstable slope regions by designing an adaptive loss function. The SlopeNet model is trained end-to-end using the standard dataset preprocessed in step S1 to automate the extraction of unstable slope features and perform image segmentation, thereby completing a pixel-level classification model. The automated extraction process of unstable slope features in step S2 specifically includes: S201. Multi-scale convolutional feature extraction: Multiple convolutional layers are used to extract multi-scale local features of the image across different receptive fields; S202. Multi-scale feature fusion: Combine pooling layers to downsample the local feature maps extracted in step S201, use max pooling to reduce the image size, and extract high-level abstract information of the image; S203. Deep Residual Connections: Borrowing from the idea of ​​ResNet, skip connections are introduced into the network; each residual block contains two convolutional layers and one skip connection; S204. Design an adaptive attention mechanism, introduce a spatial attention module at different stages of the feature map, and dynamically adjust the importance weights of features at different spatial locations according to the saliency of the feature map, so that the model focuses more on the regions and features related to unstable slope discrimination. S205. Slope Feature Enhancement Module: This module integrates prior geological information into the deep learning feature learning process to enhance the extraction of key local details of unstable slopes. S3. Using the SlopeNet model trained in step S2, reason about new or unlabeled remote sensing image data to achieve pixel-level classification and image segmentation of slope areas, thereby accurately identifying the range and boundaries of unstable slope areas. S4. Based on the range and boundary of the unstable slope area obtained in step S3, and combined with historical meteorological, hydrological and geological monitoring data, a time series dataset is constructed. The Long Short-Term Memory (LSTM) time series deep learning model is used to train the time series dataset to establish an intelligent analysis model for the evolution trend of unstable slopes. S5. Using the intelligent analysis model for the evolution trend of unstable slopes trained in step S4, input the current and recent spatial characteristics of unstable slopes and historical monitoring data sequences to predict the evolution trend of unstable slopes in the future, i.e., the probability of instability.

2. The method for identifying and detecting unstable slope features based on deep learning according to claim 1, characterized in that, The method further includes the following steps: outputting the identification results in step S3 and the prediction results in step S5 in a visual manner to provide intuitive evaluation and early warning information.

3. A method for identifying and detecting unstable slope features based on deep learning according to claim 1 or 2, characterized in that: The multi-source data in step S1 includes UAV image data, satellite remote sensing image data, geological exploration data, and meteorological and historical monitoring data; preprocessing includes data cleaning, format conversion, geometric correction, spatial registration, noise removal, image enhancement, data standardization, accurate labeling of unstable slope areas, data enhancement, multi-source data fusion, and dataset partitioning; wherein the accurate labeling of unstable slope areas is based on the slope, aspect, texture, and curvature characteristics of remote sensing images, lidar point clouds, and UAV aerial image data, observed and analyzed, and GIS software is used to draw polygons for possible landslides or unstable slope areas, and mark the unstable slope areas.

4. A method for identifying and detecting unstable slope features based on deep learning according to claim 1 or 2, characterized in that: In step S2, during the automated extraction of unstable slope features, an adaptive loss function is used to focus on unstable slope regions, thereby improving the accuracy and robustness of feature extraction. This adaptive loss function consists of three parts: cross-entropy loss, regularization loss, and slope region loss. The adaptive loss function is as follows: L total = αL ce +βL reg +γL instability ; Where: L ce This is the standard cross-entropy loss, used for classification; L reg This is the regularization loss, used to prevent the model from overfitting; L instability This is a special loss function for unstable slope regions. This loss function depends on the complexity of the slope region and can be adaptively adjusted according to the salience of the slope region. In the loss function, α, β, and γ are hyperparameters used to adjust the weights of the three parts of the loss. The value of γ should be relatively large to enhance attention to unstable slope regions.

5. A method for identifying and detecting unstable slope features based on deep learning according to claim 1 or 2, characterized in that: Step S2 uses the standard dataset preprocessed and partitioned in step S1 to perform end-to-end training of the SlopeNet model. The process employs an appropriate optimizer, Adam, and a learning rate scheduling strategy, with the goal of minimizing the adaptive weighted loss function. During training, a validation set is used to monitor model performance and adjust hyperparameters to prevent overfitting. Finally, the trained SlopeNet model weights are obtained.

6. A method for identifying and detecting unstable slope features based on deep learning according to claim 1 or 2, characterized in that, The process of identifying the extent and boundaries of unstable slope regions based on the SlopeNet model in step S3 specifically includes: S301. Data Input: Input slope area image data that has been preprocessed in step S1 but is not labeled, and is in the same format as the training data. S302. Model Inference: The input data in S301 is fed into the trained SlopeNet model for forward propagation calculation. The model outputs a probability map of each pixel belonging to an unstable category. S303. Post-processing: Post-process the output probability map by binarizing it with a threshold to generate a binary mask, where a pixel value of 1 indicates an unstable slope region and a pixel value of 0 indicates a stable region; use set operations to obtain a smooth and reasonable segmentation result. S304. Output Results: Output a pixel-level binary segmentation map and vector polygon, representing the identification results of the range and boundary of the unstable slope region.

7. A method for identifying and detecting unstable slope features based on deep learning according to claim 1 or 2, characterized in that, The process of establishing an intelligent analysis model of the evolution trend of unstable slopes based on LSTM in step S4 specifically includes: S401. Historical Spatial Feature Extraction: For slope regions with time series data, repeat steps S1 and S3, and use the trained SlopeNet model to identify and label unstable slope regions at different historical time points; then, extract quantified spatial features from these time-varying region segmentation results. S402. Historical monitoring data synchronization: Collect historical monitoring data synchronized with the time points of the above spatial feature extraction; S403. Time series construction: The spatial feature vector extracted from each historical time point is concatenated with the corresponding historical monitoring data vector to form a comprehensive feature vector. These comprehensive feature vectors are arranged in chronological order to construct a multivariate time series dataset. S404. Intelligent Analysis Model Design for Unstable Slope Evolution Trend: The intelligent analysis model for unstable slope evolution trend adopts an LSTM model architecture, including an input layer, LSTM layer, Dropout layer, fully connected layer, and output layer. The input layer receives multivariate time series data expanded step by step. The LSTM unit contains a gating mechanism, which can effectively learn and remember long-term dependency information. The Dropout layer is used to randomly deactivate some neurons between LSTM layers to prevent overfitting. The fully connected layer is responsible for mapping the output of the LSTM layer to the final prediction target. The output layer uses the sigmoid activation function to output probabilities to predict whether the slope will become unstable in the future.

8. The method for identifying and detecting unstable slope features based on deep learning according to claim 2, characterized in that, In step S5, the intelligent analysis model for the evolution trend of unstable slopes outputs prediction results and identification results from step S3. The prediction results are based on data from multiple past time steps, predicting the slope state at a future time point, as well as the slope state at future time points in the sequence. The prediction result is a time series. The model outputs probability values ​​between 0 and 1 at multiple time points, representing the slope evolution trend and its instability probability in the future, where 0 indicates no instability and 1 indicates instability. The recognition results in step S3 include outputting a pixel-level binary mask and unstable region vector boundaries, representing the spatial extent of the unstable slope region, which can be superimposed on the original image or DEM for display. The prediction results output by the intelligent analysis model for unstable slope evolution trends and the identification results in step S3 are both displayed through visualization. Specifically, this includes: overlaying the vector boundaries of the identified unstable areas onto the 3D terrain model for visualization; using heatmaps or hierarchical color schemes to overlay and display the pixel-level predicted probabilities from the SlopeNet identification results on the slope map; and using time series plots or bar charts to display the changes in historical unstable area areas, key monitoring data curves, and the future instability probability curve of the intelligent analysis model for unstable slope evolution trends over time, providing dynamic evolution trend information. The predicted probabilities are combined with preset warning thresholds, and different colors or markers are used to represent different warning levels.

9. The method for identifying and detecting unstable slope features based on deep learning according to claim 7, characterized in that: The spatial features extracted in S401 include: the area of ​​each unstable region, the shape features of the unstable region (i.e., aspect ratio), the topographic feature statistics within the unstable region, and the changes in the unstable region over time; the topographic feature statistics within the unstable region include average slope, maximum slope, surface roughness, and undulation; the changes in the unstable region over time include area changes and boundary changes; the historical monitoring data in S402 include accumulated precipitation, groundwater level, and measuring point displacement.

Images