Dangerous rock body risk assessment method based on fracture density analysis

Abstract

The invention relates to a dangerous rock body risk assessment method based on fracture density analysis. The method comprises the following steps: firstly, acquiring an aerial image of a dangerous rock mass, and preprocessing the aerial image; then, carrying out target detection on the preprocessed dangerous rock mass aerial image by adopting a YOLOv8 neural network to obtain a dangerous rock mass detection frame; on the basis of the dangerous rock mass detection frame, a Fast-SCNN convolutional neural network is adopted to carry out dangerous rock mass crack segmentation and density calculation, and dangerous rock mass crack density is obtained; and finally, determining the risk grade of the dangerous rock mass based on the crack density of the dangerous rock mass in combination with the crack direction and the cross point. That is to say, a dangerous rock mass risk assessment method based on deep learning is provided, the dangerous rock mass crack density, namely the proportion of the effective area of cracks in a dangerous rock mass target frame in the whole frame, is used as a main assessment index, and an intersection point generated due to different crack directions is used as an auxiliary judgment basis. The risk of the dangerous rock mass can be efficiently evaluated, so that the disaster risk of the rock mass is judged.

Description

Technical Field

[0001] The present application relates to the field of deep learning image processing technology in geological disaster prevention and control, and in particular to a dangerous rock mass risk assessment method based on crack density analysis. Background Art

[0002] On steep slopes, unstable rock blocks formed by multiple groups of rock structure surfaces that cut each other are called dangerous rocks. Collapse of dangerous rock masses is a common and serious phenomenon in geological disasters. Crack density is an important indicator for evaluating the stability of dangerous rock masses, and has a high correlation with the occurrence of geological disasters such as collapse. Crack density refers to the ratio of the effective area of ​​the remaining cracks to the entire frame area on the exposed surface of dangerous rocks after removing small cracks that have little impact on the overall stability of the rock mass. When the crack density is low, the internal connectivity and integrity of the rock mass are strong, the structure is relatively stable, and the risk of collapse is low. At this time, cracks may only appear on the surface of the rock mass or in a few areas, and the distance between the cracks is far, and no coherent fracture zone is formed between each other, so the rock mass can still maintain a certain stability under the action of external forces. When the crack density is high, the high density of cracks means that the interior of the rock mass has been divided into many small pieces by a large number of cracks, and the integrity is greatly weakened. The distance between the cracks becomes very close, and a fracture network that runs through the entire rock mass may be formed. In this case, the structure of the rock mass becomes very fragile, and any external force (such as earthquakes, rainfall, weathering, etc.) may cause the cracks to further expand and connect, leading to the overall instability of the rock mass, and may eventually cause geological disasters such as collapse.

[0003] Traditional dangerous rock mass monitoring and early warning methods mainly rely on manual inspections and ground monitoring equipment, which have problems such as limited monitoring range, poor real-time performance, high cost, and high risk. Therefore, in the relevant technology, there is an urgent need for a method that can improve the accuracy and efficiency of dangerous rock mass monitoring and early warning and reduce missed reports and false reports. Summary of the invention

[0004] Based on this, it is necessary to provide a dangerous rock mass risk assessment method based on crack density analysis that can improve the accuracy and efficiency of dangerous rock mass monitoring and early warning and reduce missed reports and false alarms in response to the above technical problems.

[0005] In a first aspect, the present application provides a dangerous rock mass risk assessment method based on crack density analysis. The method comprises:

[0006] Obtain aerial images of dangerous rock masses and perform preprocessing;

[0007] The YOLOv8 neural network is used to detect the preprocessed dangerous rock mass aerial images and obtain the dangerous rock mass detection frame;

[0008] Based on the dangerous rock mass detection frame, a Fast-SCNN convolutional neural network is used to segment and calculate the density of dangerous rock mass cracks to obtain the density of dangerous rock mass cracks;

[0009] The risk level of the dangerous rock mass is determined based on the crack density of the dangerous rock mass combined with the crack direction and intersection points.

[0010] Optionally, in one embodiment of the present application, the preprocessing includes:

[0011] Standardizing the aerial image of the dangerous rock mass;

[0012] The standardized dangerous rock mass aerial images are classified and labeled.

[0013] Optionally, in one embodiment of the present application, the preprocessing further includes:

[0014] The standardized aerial image of the dangerous rock mass is affine transformed and then classified and labeled.

[0015] Optionally, in one embodiment of the present application, the feature extraction and detection head part of the YOLOv8 neural network includes a rotation invariance module and a deformation perception module.

[0016] Optionally, in one embodiment of the present application, the method of using the YOLOv8 neural network for target detection includes:

[0017] A YOLOv8 neural network model is constructed and trained, and parameters of the YOLOv8 neural network model are adjusted in combination with a composite loss function, wherein the composite loss function includes a bounding box center coordinate error, a bounding box width and height error, a category probability error, and a confidence score error.

[0018] Optionally, in one embodiment of the present application, the formula of the composite loss function is:

[0019]

[0020] in, Indicates that the jth bounding box of the i-th grid contains the object, Indicates that the jth bounding box of the i-th grid does not contain the object, (x i ,y i ,w i ,h i ) represents the predicted bounding box parameters, represents the ground-truth bounding box parameters, C i represents the confidence of the prediction, Indicates the true confidence, p i (c) represents the predicted category probability, represents the true category probability, λ coord and λ noobj Represents the weight parameter of the loss function.

[0021] Optionally, in one embodiment of the present application, the use of Fast-SCNN convolutional neural network to perform dangerous rock mass crack segmentation and density calculation includes:

[0022] Selecting sample data based on the dangerous rock body detection frame, and convolving and downsampling the sample data;

[0023] Four-layer bilinear interpolation and deconvolution are used to obtain deep feature maps;

[0024] Classification is performed based on the deep-level feature map to obtain a dangerous rock mass crack segmentation image;

[0025] The crack density is calculated based on the dangerous rock mass crack segmentation image.

[0026] In a second aspect, the present application also provides a dangerous rock mass risk assessment device based on crack density analysis. The device comprises:

[0027] A data acquisition module is used to acquire aerial images of dangerous rock masses and perform preprocessing;

[0028] The dangerous rock mass positioning module is used to detect the pre-processed dangerous rock mass aerial images using the YOLOv8 neural network to obtain the dangerous rock mass detection frame;

[0029] A dangerous rock mass crack density calculation module is used to perform dangerous rock mass crack segmentation and density calculation based on the dangerous rock mass detection frame using a Fast-SCNN convolutional neural network to obtain the dangerous rock mass crack density;

[0030] The dangerous rock mass risk level assessment module is used to determine the dangerous rock mass risk level based on the dangerous rock mass crack density combined with the crack direction and intersection points.

[0031] In a third aspect, the present application further provides a computer device, wherein the computer device comprises a memory and a processor, wherein the memory stores a computer program, and the processor executes the steps of the methods described in the above embodiments.

[0032] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the method described in each of the above embodiments are implemented.

[0033] The above-mentioned dangerous rock mass risk assessment method based on crack density analysis first obtains the aerial image of the dangerous rock mass and performs preprocessing; then, the preprocessed aerial image of the dangerous rock mass is subjected to target detection using the YOLOv8 neural network to obtain a dangerous rock mass detection frame; then, based on the dangerous rock mass detection frame, the Fast-SCNN convolutional neural network is used to perform dangerous rock mass crack segmentation and density calculation to obtain the dangerous rock mass crack density; finally, the dangerous rock mass risk level is determined based on the dangerous rock mass crack density combined with the crack direction and intersection points. In other words, a dangerous rock mass risk assessment method based on deep learning is proposed, in which the high-resolution aerial image of the mountain is inferred by the YOLOv8 deep learning network to obtain the dangerous rock mass target frame, and within the target frame, the Fast-SCNN neural network model is used for semantic segmentation to extract the crack edge information and calculate the dangerous rock mass crack density. The dangerous rock mass crack density, i.e., the ratio of the effective area of ​​the crack in the dangerous rock mass target frame to the entire frame, is used as the main evaluation index, and the intersection points generated by different crack directions are combined as auxiliary judgment basis, which can efficiently evaluate the risk of dangerous rock mass, thereby judging the risk of rock mass disaster. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 This is an application environment diagram of a dangerous rock mass risk assessment method based on crack density analysis in one embodiment;

[0035] Figure 2 Schematic diagram of a flow chart of a dangerous rock mass risk assessment method based on crack density analysis in one embodiment;

[0036] Figure 3 A schematic diagram of a dangerous rock mass positioning process in one embodiment;

[0037] Figure 4 A schematic diagram of a dangerous rock mass fracture segmentation model training process in one embodiment;

[0038] Figure 5 A schematic diagram of calculating the density of cracks in a dangerous rock mass in one embodiment;

[0039] Figure 6 is a structural block diagram of a dangerous rock mass risk assessment device based on crack density analysis in one embodiment;

[0040] Figure 7 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0041] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0042] The dangerous rock mass risk assessment method based on crack density analysis provided in the embodiment of the present application can be applied to Figure 1 In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be integrated on the server 104, or it can be placed on the cloud or other network servers. Among them, the terminal 102 can be but is not limited to various personal computers, laptops, smart phones, tablets, Internet of Things devices and portable wearable devices. The Internet of Things devices can be smart speakers, smart TVs, smart air conditioners, smart car-mounted devices, etc. Portable wearable devices can be smart watches, smart bracelets, head-mounted devices, etc. The server 104 can be implemented with an independent server or a server cluster consisting of multiple servers.

[0043] In one embodiment, Figure 2 As shown in the figure, a risk assessment method for dangerous rock mass based on crack density analysis is provided. Figure 1 The server in the example is used to illustrate the following steps:

[0044] S201: Acquire an aerial image of a dangerous rock mass and perform preprocessing.

[0045] In the embodiment of the present application, first, an aerial image of a dangerous rock mass is obtained and preprocessed. Aerial images of dangerous rock mass refer to images of a mountain taken by a drone from different angles, ensuring that the neural network can learn the characteristics of the dangerous rock mass at different viewing angles, including horizontal rotation, vertical rotation, and changes in tilt angles. Preprocessing includes unifying the size, resolution, and information annotation, so that the aerial image of the dangerous rock mass meets the input conditions of the iterative learning network.

[0046] Specifically, in one embodiment of the present application, the preprocessing includes:

[0047] S301: Standardizing the aerial image of the dangerous rock mass.

[0048] S303: Classify and label the standardized dangerous rock mass aerial images.

[0049] In one embodiment of the present application, the PIL library provided in Python is used to process the size and resolution of the image, perform standardization, and apply labelImg to classify, confirm and label the standardized aerial images of the dangerous rock mass, and establish a training set and verification set for the deep learning network. The labeling of the dangerous rock mass is not limited to one scale, but also considers the labeling at different distances, so that the neural network can recognize the characteristics of the dangerous rock mass at different distances and sizes, and enhance its generalization ability. In the labeling process, data enhancement techniques such as random cropping, color transformation, noise addition, etc. are applied to further enrich the data diversity of the training set and verification set and improve the robustness of the model.

[0050] In one embodiment of the present application, the preprocessing further includes:

[0051] The standardized aerial image of the dangerous rock mass is affine transformed and then classified and labeled.

[0052] In one embodiment of the present application, in order to simulate the deformation in the natural environment, the image is subjected to an appropriate affine transformation before classification and labeling. Optionally, the affine transformation includes operations such as scaling, rotation, and translation. Taking scaling as an example, if the short side of the image is less than 640 pixels, the short side is multiplied by the corresponding ratio to be enlarged to 640 pixels, and the long side is multiplied by the same ratio; if the long side of the image is greater than 1280 pixels, the long side is multiplied by the corresponding ratio to be reduced to 1280 pixels, and the short side is multiplied by the same ratio. This ensures that the short side of the image is not shorter than 640 pixels, and the long side is not greater than 1280 pixels.

[0053] In this embodiment, by performing affine transformation on the standardized aerial image of the dangerous rock mass and then classifying and labeling it, the neural network can learn the characteristics of the deformed dangerous rock mass, thereby improving the recognition capability of the deformed target.

[0054] S203: Using the YOLOv8 neural network to perform target detection on the preprocessed dangerous rock mass aerial image to obtain a dangerous rock mass detection frame.

[0055] In an embodiment of the present application, the preprocessed aerial image of the dangerous rock mass is input into the YOLOv8 neural network for processing. YOLOv8 can be divided into four main parts: first, the backbone network (Backbone) extracts the feature map of the image through a set of basic convolutional layers, activation functions and pooling layers, and these feature maps are shared for subsequent feature pyramid networks (Feature Pyramid Network, FPN) and detection heads (Head); secondly, the feature pyramid network combines feature maps from different scales through multi-scale feature fusion to generate multi-scale feature maps to enhance the detection capability of targets of different sizes; then, the detection head predicts the position and size of the target through a predefined set of anchor points, predicts the category of the target, the coordinates of the bounding box and the confidence score for each anchor point, and adjusts the position and size of the anchor point through bounding box regression to obtain accurate detection results; finally, YOLOv8 uses a composite loss function to simultaneously optimize classification and positioning to obtain the final precise position of the detection box. The classification loss uses binary cross entropy or cross entropy loss, and the positioning loss uses GIoU or DIoU loss. Classification, confidence, and bounding box regression are comprehensively considered to ensure the balanced performance of the model on different tasks. The position coordinates (x, y, w, h) of the dangerous rock body are obtained through the above target detection, where (x, y) is the coordinate of the center point of the prediction box, w, h are the length and width of the prediction box respectively, and the content information of the prediction box (x) is obtained after the change. min ,y min , x max ,y max ), and use the detection box as the detection box for the next stage of semantic segmentation.

[0056] In one embodiment of the present application, the feature extraction and detection head part of the YOLOv8 neural network includes a rotation invariance module and a deformation perception module.

[0057] In one embodiment of the present application, because most mountain images are top-down views and targets often undergo rotational deformation, a rotational invariance module and a deformation perception module are introduced into the feature extraction and detection head of the YOLOv8 neural network to enhance the detection capability of rotated and deformed targets, thereby achieving higher detection accuracy and robustness in rock risk monitoring tasks.

[0058] The core of the rotation invariance module is to ensure that the neural network can still correctly identify the target when it is rotated in the image. This module consists of a direction-sensitive convolution layer and a rotation pooling layer. The rotation-invariant convolution kernel is used to capture the features of a specific direction in the dangerous rock image. At the same time, the rotation pooling layer is used to fuse the features at different rotation angles to ensure that the extracted features do not depend on the image rotation angle.

[0059] The core of the deformation perception module is to ensure that the neural network can still accurately identify the target when it is deformed (such as stretching or scaling) in the image. This module consists of an adaptive convolution layer and multi-scale feature fusion. Through the adaptive convolution kernel, the position of the convolution operation is dynamically adjusted to adapt to the deformation of the target. At the same time, through multi-scale feature fusion, the feature information at different scales is integrated to ensure that the extracted features can effectively cope with the deformation of the target.

[0060] In one embodiment of the present application, the method of using the YOLOv8 neural network for target detection includes:

[0061] A YOLOv8 neural network model is constructed and trained, and parameters of the YOLOv8 neural network model are adjusted in combination with a composite loss function, wherein the composite loss function includes a bounding box center coordinate error, a bounding box width and height error, a category probability error, and a confidence score error.

[0062] In one embodiment of the present application, a YOLOv8 neural network model is constructed, and the YOLOv8 neural network model is trained using a training set. Specifically, Figure 3 As shown in the figure, the feature map of the image is obtained through the feature extraction layer (including multiple convolution layers, activation functions and pooling layers) of the YOLOv8 neural network model. The convolution layer of YOLOv8 uses a 3x3 convolution kernel with a step size of 1 and a compensation of 1, which makes the length and width of the input and output unchanged. The pooling layer uses a 2x2 convolution kernel with a step size of 2, so that the length and width of the output become half of the input. After that, the obtained feature map is sent to the detection head of YOLOv8. The detection head contains three paths, which are used to predict categories, bounding boxes and confidences respectively. First, the feature map passes through a 3x3 convolution layer to generate a multi-scale feature map. Then, it passes through three convolution layers respectively to predict the category probability, bounding box regression value and confidence score of each anchor point. Finally, the results of the three paths are merged at the output layer. Through the non-maximum suppression (NMS) algorithm, the confidence and bounding box regression values ​​are integrated to filter out the area containing the detection target, eliminate the repeated and low-confidence detection results, and finally output the accurate target detection box and category label. The classification part uses the feature map containing the detected target that has been obtained, and calculates which category each area belongs to (such as dangerous rock mass, cracks, slopes, etc.) through the convolution layer and the softmax layer, and outputs their corresponding probability vectors; at the same time, the bounding box regression is used again to obtain the position offset of each detected target, which is used to regress a more accurate target detection frame. During the training process, the parameters of the YOLOv8 neural network model are adjusted in combination with the composite loss function, which includes the bounding box center coordinate error, bounding box width and height error, category probability error, and confidence score error. After the training is completed, the model is tested using the test set.

[0063] Specifically, in one embodiment of the present application, the formula of the composite loss function is:

[0064]

[0065] in, Indicates that the jth bounding box of the i-th grid contains the object, Indicates that the jth bounding box of the i-th grid does not contain the object, (x i ,y i ,w i ,h i ) represents the predicted bounding box parameters, represents the ground-truth bounding box parameters, C i represents the confidence of the prediction, Indicates the true confidence, p i (c) represents the predicted category probability, represents the true category probability, λ coord and λ noobj Represents the weight parameter of the loss function.

[0066] In one embodiment of the present application, the entire loss function includes four parts: (1) Calculating the error between the predicted bounding box center coordinates and the true value. This part is only valid when there is a target in the grid cell, and the bounding box with the highest IOU with the true value is selected. (2) Calculating the error between the square root of the predicted bounding box width and height and the square root of the true value to reduce the impact of small deviations of large bounding boxes. (3) Calculating the error between the predicted category probability and the true category probability. The classification error is calculated only when there is a target in the grid cell to avoid penalizing grid cells where there is no target. (4) Calculating the error between the confidence score of each bounding box prediction value and the true confidence score. The parameter λ is used to balance the different parts of the loss function to ensure the stability of the model, where the highest penalty is used for coordinate prediction and the lowest penalty is used for confidence prediction when the target is not detected. BCE (binary cross entropy) is used as the classification loss. Each category judges "whether it is this type" and outputs the confidence. The specific formula is as follows:

[0067]

[0068] Among them, L represents the average value of binary cross entropy loss of all samples, N is the total number of samples in the data set, and L i is the loss of the i-th data, y i is the true label of the i-th data, indicating the actual category of the sample, p i is the probability that the i-th data belongs to the positive class.

[0069] S205: Based on the dangerous rock mass detection frame, a Fast-SCNN convolutional neural network is used to perform dangerous rock mass crack segmentation and density calculation to obtain dangerous rock mass crack density.

[0070] In the embodiment of the present application, after obtaining the dangerous rock detection frame, the Fast-SCNN convolutional neural network is used to segment and calculate the density of the dangerous rock cracks. The Fast-SCNN convolutional neural network includes two main parts: (1) a feature extraction part, which extracts multi-scale features through convolution and pooling layers, including multiple scale levels including the initial image; (2) an upsampling and feature fusion part, which restores the spatial size of the feature map through the upsampling layer and fuses it with the corresponding scale features of the feature extraction part. The Fast-SCNN convolutional neural network can obtain a dangerous rock crack segmentation image, which provides the area size of the crack, and the dangerous rock crack density is the ratio of the crack area to the dangerous rock frame area.

[0071] Specifically, in one embodiment of the present application, the use of Fast-SCNN convolutional neural network to perform dangerous rock mass crack segmentation and density calculation includes:

[0072] S401: Select sample data based on the dangerous rock body detection frame, and perform convolution and down-sampling on the sample data.

[0073] S403: Use four-layer bilinear interpolation and deconvolution to obtain deep feature maps.

[0074] S405: Classification is performed based on the deep-level feature map to obtain a dangerous rock mass crack segmentation image.

[0075] S407: Calculating the crack density based on the dangerous rock mass crack segmentation image.

[0076] In one embodiment of the present application, first, based on the dangerous rock detection frame, images with rock targets that meet the characteristics of dangerous rock and diverse perspectives are selected as sample data, and labelme is applied to classify, confirm and annotate the selected sample images to establish a training set and verification set for the Fast-SCNN convolutional neural network. In specific applications, there are no special requirements for the size of the image samples in the data set, but in order to meet the input requirements of the Fast-SCNN convolutional neural network, the image samples of the dangerous rock positioning in the previous stage are uniformly processed into a uniform size of 512×512.

[0077] Fast-SCNN convolutional neural network includes multiple levels of convolution, batch normalization, activation function and maximum pooling operations. The entire network consists of a learning downsampling module, a global feature extraction module and a feature fusion module. Figure 4As shown in the figure, the sample data is convolved and downsampled. The learning downsampling module includes a standard convolution layer and two depthwise separable convolution layers, using a 3x3 convolution kernel with a step size of 2; the pooling operation also uses a 2x2 pooling kernel with a step size of 2. After that, the low-resolution feature map is gradually restored to its original size through a combination of quadratic linear interpolation, depthwise separable convolution (DWConv) and standard convolution (Conv). This process includes four bilinear interpolation and deconvolution operations to ensure high-resolution restoration of the feature map and retention of detail information for further calculation. The global feature extraction module captures multi-scale global features through multiple depthwise separable convolution layers and inverted residual bottleneck blocks, and uses a pyramid pooling module to enhance feature extraction. The feature fusion module restores the low-resolution feature map to high resolution through the upsampling layer and fuses it with the feature map of the downsampled part to effectively extract and fuse the detail features and semantic features of the dangerous rock mass, and generate an accurate feature map of the dangerous rock mass crack.

[0078] At the end of the model, the maximum numerical description of the pixel position in the two images is used as the classification of the pixel, thus generating a segmented image and achieving end-to-end. The method for calculating this probability is: first, a pixel-wise softmax classifier is used, that is, the output corresponding to each pixel is softmaxed separately. The pixel-wise softmax classifier function is:

[0079]

[0080] where a k (x) represents the activation value of the pixel at position x in the kth layer of the feature map, that is, the score of each pixel point (x) corresponding to the feature channel (k); K is the total number of pixel categories; p k (x) is the classification result of pixel x for class k, that is, the probability; the weighted Loss algorithm is selected to monitor the loss during the training model, and the weighted Loss is:

[0081]

[0082] Where l:Ω→1,…,K is the true label of each pixel, ω(x) is the weight, from which the loss rate of each iterative learning can be obtained to monitor the learning process of the model.

[0083] The crack area ratio is a key indicator when assessing the risk level of dangerous rock mass. The larger the crack area ratio, the more serious the degree of fracture of the rock mass, the worse the stability of the rock mass, and the higher the risk of disaster. The dangerous rock mass area obtained by target detection is semantically segmented, the number of crack pixels is counted, and the cracks are further screened to remove invalid cracks with small areas and little impact on structural stability. The total area and proportion of effective cracks in the detection frame are calculated. If the effective crack area ratio is high, it means that the degree of fracture of the rock mass is relatively serious, and the integrity and stability of the structure have been significantly damaged.

[0084] S207: Determine the risk level of the dangerous rock mass based on the density of the cracks in the dangerous rock mass combined with the crack directions and intersections.

[0085] In the embodiments of the present application, crack density is a key indicator when evaluating the risk level of dangerous rock mass. The larger the crack area ratio, the more serious the degree of rupture of the rock mass, the worse the stability, and the increased risk of disaster. By semantically segmenting the dangerous rock mass area, counting the number of crack pixels, and removing invalid cracks that are too small and have little impact on structural stability, the total area of ​​effective cracks and their proportion are finally calculated. At the same time, the crack direction and intersection are used as auxiliary judgment factors. When the crack directions are inconsistent and there are intersections, it usually indicates that the weak points of the rock mass have increased, which further increases the instability. The risk level of the dangerous rock mass is comprehensively evaluated by combining the crack area ratio and the distribution of crack directions and intersections.

[0086] The risk level is divided according to the proportion of crack area to the detection frame area: low risk (crack ratio ≤ 5%): it indicates that the overall rock structure is relatively intact, cracks have little destructive effect on it, and the risk of collapse is low; medium risk (5% < crack ratio ≤ 30%): it indicates that cracks have expanded significantly, the rock structure has been weakened, there is a potential risk of collapse, and regular monitoring is required; high risk (30% < crack ratio ≤ 50%): the rock structure has been severely damaged, the risk of collapse has increased significantly, and preventive measures need to be taken; extremely high risk (crack ratio > 50%): the rock structure is severely damaged, the integrity is almost lost, and it may collapse at any time, and emergency protective measures need to be taken immediately.

[0087] like Figure 5As shown in the figure, the specific process of calculating the density of cracks in dangerous rock mass is as follows. First, the area where the cracks are located is identified by the target detection algorithm, and the detection frame is marked in the upper right corner of the image. The area of ​​the detection frame is 11618 pixels. Then, the crack area is accurately extracted by the semantic segmentation algorithm, and the actual area of ​​the crack is calculated to be 1579 pixels in the lower left corner image. Finally, in the lower right corner image, it is calculated that the proportion of the crack area to the total area of ​​the detection frame is 13.5%. This ratio reflects the density of the cracks in the detection frame. At the same time, the direction of the cracks is extracted, and the number of intersections between the cracks is obtained. It can be found that there are intersections between the cracks, which verifies that the crack density is an effective crack density, and the dangerous rock is judged to be of medium risk.

[0088] In the above-mentioned dangerous rock mass risk assessment method based on crack density analysis, first, the aerial image of the dangerous rock mass is obtained and preprocessed; then, the preprocessed aerial image of the dangerous rock mass is subjected to target detection using the YOLOv8 neural network to obtain a dangerous rock mass detection frame; then, based on the dangerous rock mass detection frame, the Fast-SCNN convolutional neural network is used to segment and calculate the crack density of the dangerous rock mass to obtain the crack density of the dangerous rock mass; finally, the risk level of the dangerous rock mass is determined based on the crack density of the dangerous rock mass combined with the crack direction and intersection points. In other words, a dangerous rock mass risk assessment method based on deep learning is proposed, in which the high-resolution aerial image of the mountain is inferred by the YOLOv8 deep learning network to obtain a dangerous rock mass target frame, and within the target frame, the Fast-SCNN neural network model is used for semantic segmentation, the crack edge information is extracted, and the crack density of the dangerous rock mass is calculated. The crack density of the dangerous rock mass, that is, the ratio of the effective area of ​​the crack in the dangerous rock mass target frame to the entire frame, is used as the main evaluation index, and the intersection points generated by different crack directions are combined as auxiliary judgment basis, which can efficiently evaluate the risk size of the dangerous rock mass, thereby judging the risk of rock mass disaster.

[0089] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0090] Based on the same inventive concept, the embodiment of the present application also provides a dangerous rock mass risk assessment device based on crack density analysis for implementing the dangerous rock mass risk assessment method based on crack density analysis mentioned above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in the embodiments of one or more dangerous rock mass risk assessment devices based on crack density analysis provided below can refer to the limitations of the dangerous rock mass risk assessment method based on crack density analysis above, and will not be repeated here.

[0091] In one embodiment, Figure 6 As shown, a dangerous rock mass risk assessment device 600 based on crack density analysis is provided, comprising: a data acquisition module 601, a dangerous rock mass positioning module 603, a dangerous rock mass crack density calculation module 605 and a dangerous rock mass risk level assessment module 607, wherein:

[0092] The data acquisition module 601 is used to acquire the aerial image of the dangerous rock mass and perform preprocessing.

[0093] The dangerous rock body positioning module 603 is used to perform target detection on the pre-processed dangerous rock body aerial image using the YOLOv8 neural network to obtain a dangerous rock body detection frame.

[0094] The dangerous rock mass crack density calculation module 605 is used to perform dangerous rock mass crack segmentation and density calculation based on the dangerous rock mass detection frame using a Fast-SCNN convolutional neural network to obtain the dangerous rock mass crack density.

[0095] The dangerous rock mass risk level assessment module 607 is used to determine the dangerous rock mass risk level based on the dangerous rock mass crack density combined with the crack direction and intersection points.

[0096] In one embodiment of the present application, the data acquisition module is further used to:

[0097] Standardizing the aerial image of the dangerous rock mass;

[0098] The standardized dangerous rock mass aerial images are classified and labeled.

[0099] In one embodiment of the present application, the data acquisition module is further used to:

[0100] The standardized aerial image of the dangerous rock mass is affine transformed and then classified and labeled.

[0101] In one embodiment of the present application, the feature extraction and detection head part of the YOLOv8 neural network includes a rotation invariance module and a deformation perception module.

[0102] In one embodiment of the present application, the dangerous rock mass positioning module is also used for:

[0103] A YOLOv8 neural network model is constructed and trained, and parameters of the YOLOv8 neural network model are adjusted in combination with a composite loss function, wherein the composite loss function includes a bounding box center coordinate error, a bounding box width and height error, a category probability error, and a confidence score error.

[0104] In one embodiment of the present application, the formula of the composite loss function is:

[0105]

[0106] in, Indicates that the jth bounding box of the i-th grid contains the object, Indicates that the jth bounding box of the i-th grid does not contain the object, (x i ,y i ,w i ,h i ) represents the predicted bounding box parameters, represents the ground-truth bounding box parameters, C i represents the confidence of the prediction, Indicates the true confidence, p i (c) represents the predicted category probability, represents the true category probability, λ coord and λ noobj Represents the weight parameter of the loss function.

[0107] In one embodiment of the present application, the dangerous rock mass crack density calculation module is also used for:

[0108] Selecting sample data based on the dangerous rock body detection frame, and convolving and downsampling the sample data;

[0109] Four-layer bilinear interpolation and deconvolution are used to obtain deep feature maps;

[0110] Classification is performed based on the deep-level feature map to obtain a dangerous rock mass crack segmentation image;

[0111] The crack density is calculated based on the dangerous rock mass crack segmentation image.

[0112] Each module in the dangerous rock mass risk assessment device based on crack density analysis can be implemented in whole or in part by software, hardware and a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to each module.

[0113] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 7 As shown. The computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a dangerous rock mass risk assessment method based on crack density analysis is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a key, trackball or touchpad set on the computer device shell, or an external keyboard, touchpad or mouse.

[0114] Those skilled in the art will understand that Figure 7 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0115] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0116] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0117] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0118] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0119] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0120] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0121] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A dangerous rock mass risk assessment method based on crack density analysis, characterized in that: The method comprises: Obtain aerial images of dangerous rock masses and perform preprocessing; The YOLOv8 neural network is used to detect the preprocessed dangerous rock mass aerial images and obtain the dangerous rock mass detection frame; Based on the dangerous rock mass detection frame, a Fast-SCNN convolutional neural network is used to segment and calculate the density of dangerous rock mass cracks to obtain the density of dangerous rock mass cracks; The risk level of the dangerous rock mass is determined based on the crack density of the dangerous rock mass combined with the crack direction and intersection points.

2. The dangerous rock mass risk assessment method based on crack density analysis according to claim 1 is characterized in that: The pre-processing comprises: Standardizing the aerial image of the dangerous rock mass; The standardized dangerous rock mass aerial images are classified and labeled.

3. The dangerous rock mass risk assessment method based on crack density analysis according to claim 2 is characterized in that: The pre-processing further comprises: The standardized aerial image of the dangerous rock mass is affine transformed and then classified and labeled.

4. The dangerous rock mass risk assessment method based on crack density analysis according to claim 1 is characterized in that: The feature extraction and detection head part of the YOLOv8 neural network includes a rotation invariance module and a deformation perception module.

5. The dangerous rock mass risk assessment method based on crack density analysis according to claim 1 is characterized in that: The target detection using the YOLOv8 neural network includes: A YOLOv8 neural network model is constructed and trained, and parameters of the YOLOv8 neural network model are adjusted in combination with a composite loss function, wherein the composite loss function includes a bounding box center coordinate error, a bounding box width and height error, a category probability error, and a confidence score error.

6. The dangerous rock mass risk assessment method based on crack density analysis according to claim 5 is characterized in that: The formula of the composite loss function is: in, Indicates that the jth bounding box of the i-th grid contains the object, Indicates that the jth bounding box of the i-th grid does not contain the object, (x i ,y i ,w i ,h i ) represents the predicted bounding box parameters, represents the ground-truth bounding box parameters, C i represents the confidence of the prediction, Indicates the true confidence, p i (c) represents the predicted category probability, represents the true category probability, λ coord and λ noobj Represents the weight parameter of the loss function.

7. The dangerous rock mass risk assessment method based on crack density analysis according to claim 1 is characterized in that: The use of Fast-SCNN convolutional neural network to perform dangerous rock mass crack segmentation and density calculation includes: Selecting sample data based on the dangerous rock body detection frame, and convolving and downsampling the sample data; Four-layer bilinear interpolation and deconvolution are used to obtain deep feature maps; Classification is performed based on the deep-level feature map to obtain a dangerous rock mass crack segmentation image; The crack density is calculated based on the dangerous rock mass crack segmentation image.

8. A dangerous rock mass risk assessment device based on crack density analysis, characterized in that: The device comprises: A data acquisition module is used to acquire aerial images of dangerous rock masses and perform preprocessing; The dangerous rock mass positioning module is used to detect the pre-processed dangerous rock mass aerial images using the YOLOv8 neural network to obtain the dangerous rock mass detection frame; A dangerous rock mass crack density calculation module is used to perform dangerous rock mass crack segmentation and density calculation based on the dangerous rock mass detection frame using a Fast-SCNN convolutional neural network to obtain the dangerous rock mass crack density; The dangerous rock mass risk level assessment module is used to determine the dangerous rock mass risk level based on the dangerous rock mass crack density combined with the crack direction and intersection points.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images