Method and system for detecting surrounding rock fragmentation degree based on machine vision

Abstract

The present disclosure relates to the technical field of detection of surrounding rock fragmentation, and proposes a detection method and system for surrounding rock fragmentation based on machine vision, including the following steps: acquiring tunnel surrounding rock images to be identified; preprocessing the acquired images; transmitting the preprocessed images in parallel to multiple trained VGGNet models, each VGGNet model outputs a fragmentation evaluation result, and multiple fragmentation evaluation results are obtained; the evaluation results are fused using a voting model fusion method to obtain a final evaluation result. Multiple recognition models are set for image recognition, and the final evaluation result is finally obtained by a voting method, which can avoid the problem of low recognition accuracy of a single model, and select a suitable anchor spraying support scheme according to the evaluation results, which can improve the safety and reliability of the support, save construction costs, and improve construction efficiency.

Description

Technical Field

[0001] The present disclosure relates to a technical field related to detection of surrounding rock fragmentation degree, and in particular, to a method and system for detecting surrounding rock fragmentation degree based on machine vision. Background Art

[0002] The statements in this section merely provide background information related to the present disclosure and do not necessarily constitute prior art.

[0003] With the continuous expansion of the scale of mines, tunnels and construction projects, tunnel construction has gradually shifted from a small burial depth and low ground stress environment to a large burial depth and high ground stress environment. The accurate assessment of the degree of surrounding rock fragmentation has become one of the key factors to ensure the safety, sustainability and economy of the project. Tunnel engineering plays a vital role in various fields such as urban transportation, water conservancy projects, underground facilities, etc., and the stability of the surrounding rock is crucial for the safe operation of the tunnel. Different types of surrounding rocks show different physical properties and degrees of fragmentation under different geographical and geological conditions, from hard rocks to soil and mixed sand, and the degree of fragmentation varies accordingly. This diversity means that different support schemes and construction strategies must be adopted in different situations. Therefore, in order to select appropriate support measures and ensure the long-term reliability of the tunnel, the degree of surrounding rock fragmentation must be accurately assessed. The accuracy of the degree of fragmentation assessment directly affects the selection of support schemes and the control of tunnel construction and maintenance costs.

[0004] The inventors found in their research that, at present, the degree of surrounding rock fragmentation usually relies on manual inspection or the use of traditional geological and geotechnical methods. These methods have some limitations, including the subjectivity and errors of manual evaluation, as well as the high cost of time and resources, and are not suitable for rapid decision-making and timely adjustments. In the prior art, some attempts have been made to use machine vision to evaluate the degree of surrounding rock fragmentation, but these methods have certain limitations. For example, some methods rely on manual image analysis, which may still lead to subjectivity and inconsistency. In addition, traditional machine vision methods may not be able to accurately capture complex surrounding rock structures and changes in the degree of fragmentation. Summary of the invention

[0005] In order to solve the above problems, the present invention proposes a method and system for detecting the degree of surrounding rock fragmentation based on machine vision. The method based on machine vision can efficiently and accurately evaluate the degree of surrounding rock fragmentation, thereby providing more reliable guidance and decision-making support for projects such as anchor spraying support, and formulating corresponding anchor spraying support schemes for different degrees of surrounding rock fragmentation, thereby achieving safe, economical and efficient tunnel support.

[0006] In order to achieve the above objectives, the present disclosure adopts the following technical solutions:

[0007] One or more embodiments provide a method for detecting the degree of surrounding rock fragmentation based on machine vision, comprising the following steps:

[0008] Acquire the tunnel surrounding rock image to be identified;

[0009] Preprocessing the acquired images;

[0010] The preprocessed images are transmitted in parallel to multiple trained VGGNet models, each VGGNet model outputs a fragmentation evaluation result, and multiple fragmentation evaluation results are obtained;

[0011] The voting model fusion method is used to fuse the evaluation results to obtain the final evaluation results.

[0012] One or more embodiments provide a system for detecting the degree of surrounding rock fragmentation based on machine vision, including:

[0013] Image acquisition module: configured to acquire an image of the tunnel surrounding rock to be identified;

[0014] Preprocessing module: configured to preprocess the acquired image;

[0015] Multi-model evaluation module: configured to transmit the preprocessed images to multiple trained VGGNet models in parallel, each VGGNet model outputs a fragmentation evaluation result, and obtains multiple fragmentation evaluation results;

[0016] Fusion module: configured to fuse the evaluation results using a voting model fusion method to obtain a final evaluation result.

[0017] An electronic device includes a memory and a processor, and computer instructions stored in the memory and running on the processor. When the computer instructions are run by the processor, the steps in the above-mentioned method for detecting the degree of surrounding rock fragmentation based on machine vision are completed.

[0018] A computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the steps in the above-mentioned method for detecting the degree of surrounding rock fragmentation based on machine vision are completed.

[0019] Compared with the prior art, the present invention has the following beneficial effects:

[0020] In the present invention, multiple recognition models are set for image recognition, and the final evaluation result is finally obtained through a voting method, which can avoid the problem of low recognition accuracy of a single model. In addition, parallel recognition of images is realized by receiving recognition images in parallel by multiple models, which can achieve efficient and accurate evaluation results of the degree of surrounding rock fragmentation. According to the evaluation results, a suitable anchor spraying support scheme is selected, which can improve the safety and reliability of the support, save construction costs, and improve construction efficiency.

[0021] The advantages of the present disclosure and the advantages of additional aspects will be described in detail in the following specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The accompanying drawings constituting a part of the present disclosure are used to provide a further understanding of the present disclosure. The exemplary embodiments of the present disclosure and the description thereof are used to explain the present disclosure but do not constitute a limitation of the present disclosure.

[0023] Figure 1 is a flow chart of a method for detecting the degree of surrounding rock fragmentation according to Embodiment 1 of the present disclosure;

[0024] Figure 2 is a flow chart of surrounding rock crushing image feature learning and recognition evaluation in Example 1 of the present disclosure;

[0025] Figure 3 It is a schematic diagram of the surrounding rock crushing image training flow chart of the VGGNet network model of Example 1 of the present disclosure. DETAILED DESCRIPTION

[0026] The present disclosure is further described below in conjunction with the accompanying drawings and embodiments.

[0027] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present disclosure belongs.

[0028] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present disclosure. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof. It should be noted that, in the absence of conflict, the various embodiments in the present disclosure and the features in the embodiments can be combined with each other. The embodiments will be described in detail below in conjunction with the accompanying drawings.

[0029] Example 1

[0030] In the technical solutions disclosed in one or more embodiments, Figures 1 to 3 As shown, a method for detecting the degree of surrounding rock fragmentation based on machine vision includes the following steps:

[0031] Step 1: Obtain the tunnel surrounding rock image to be identified;

[0032] Step 2: preprocessing the acquired image;

[0033] Step 3: The preprocessed images are transmitted to multiple trained VGGNet models in parallel, and each VGGNet model outputs a fragmentation evaluation result, thereby obtaining multiple fragmentation evaluation results;

[0034] Step 4: Use the voting model fusion method to fuse the evaluation results to obtain the final evaluation results.

[0035] In this embodiment, multiple recognition models are set for image recognition, and the final evaluation result is finally obtained through the voting method, which can avoid the problem of low recognition accuracy of a single model. In addition, parallel recognition of images is realized by receiving recognition images in parallel by multiple models, which can achieve efficient and accurate evaluation results of the degree of surrounding rock fragmentation. According to the evaluation results, a suitable anchor spraying support scheme is selected, which can improve the safety and reliability of the support, save construction costs, and improve construction efficiency.

[0036] Furthermore, according to the evaluation results, a corresponding anchor spraying plan is formulated, including the arrangement density of anchor rods and the selection of spraying materials.

[0037] Specifically, the anchor spraying scheme includes the selection and setting of anchor rods, anchor spraying concrete, steel supports, and mesh spraying anchors; the support parameters are determined according to the degree of surrounding rock fragmentation, which may include the length, spacing and depth of the anchor rods, the thickness and strength of the anchor spraying concrete, the structural strength of the steel supports, and the density of the mesh anchor spraying.

[0038] Optionally, the degree of surrounding rock crushing may include five categories: complete, slightly crushed, moderately crushed, severely crushed and completely crushed.

[0039] Furthermore, it also includes visualizing the results of the fragmentation assessment, specifically:

[0040] According to the assessment results, the results are divided into complete, slightly broken, moderately broken, severely broken and completely broken. A scatter plot is used to show the assessment results of the degree of breakage of each image, with the horizontal axis representing the image index and the vertical axis representing the assessment results of the degree of breakage.

[0041] Optionally, the preprocessing method in step 2 includes:

[0042] Step 21, using mean filtering to perform noise reduction on the image;

[0043] Step 22: Using histogram equalization to redistribute the grayscale distribution of pixels, enhance the contrast of the image, and enhance the image;

[0044] Step 23: Use feature point matching to detect and match feature points in the image to achieve image alignment; rotate and reverse the image to increase data diversity and improve the generalization ability of the model.

[0045] Furthermore, a normalization processing method is also included, which uses a maximum and minimum value normalization method to normalize the image, find the minimum pixel value min and the maximum pixel value max of the image, and perform normalization calculation on each pixel value x in the image:

[0046]

[0047] Among them, min(x) and max(x) represent the minimum and maximum values ​​of image pixels respectively.

[0048] Among them, the voting model fusion method is as follows: for each sample, the number of categories appearing in the prediction results of the three models is counted, and the category with the largest number of occurrences is selected as the final prediction result.

[0049] Furthermore, the VGGNet model training process is as follows: Figure 3 As shown, the following steps are included:

[0050] S1. Obtain tunnel surrounding rock crushing image data to build a data set and annotate the image;

[0051] S2, preprocess the images in the dataset, including denoising, image enhancement, image alignment, rotation and flipping;

[0052] S3, normalizing the preprocessed image;

[0053] S4. Use the pre-trained VGGNet model as the initial parameters to perform transfer learning on the image to obtain a fine-tuned model. The fine-tuned model extracts the features required for surrounding rock fragmentation assessment;

[0054] S5. Construct multiple VGGNet models, extract features of different scales for evaluating the degree of surrounding rock fragmentation, and use different scale features to correspond to different VGGNet models. Use the features extracted by the fine-tuned model and the labeled image data to train multiple VGGNet models, establish a correlation model between the degree of surrounding rock fragmentation and image features, and obtain the trained VGGNet model.

[0055] In this embodiment, the pre-trained VGGNet model is first used for transfer learning to obtain a fine-tuned model. Then, the fine-tuned model is used to extract the features required for surrounding rock fragmentation assessment. These features, together with the labeled surrounding rock fragmentation data, are used to train the newly constructed VGGNet model. The accuracy of the trained model in recognition can be improved.

[0056] S6, evaluate the performance of each trained VGGNet model, select the candidate model as the trained VGGNet model, and use the voting model fusion method to evaluate the degree of surrounding rock fragmentation;

[0057] Step S1 specifically collects a large amount of historical image data about surrounding rocks, including surrounding rock images with different degrees of fragmentation and integrity, and divides them into a training set, a test set, and a validation set.

[0058] Specifically, the data are annotated and each image is assigned a label to indicate its degree of fragmentation, using a value between [0-1], where 0 represents intact surrounding rock and 1 represents completely fragmented surrounding rock, and the intermediate values, i.e., decimals between 0 and 1, represent different degrees of fragmentation.

[0059] The preprocessing method of step S2 is the same as that of step 2, and step S3 is the same as the normalization method described above, which will not be repeated here.

[0060] Step S4, using the pre-trained VGGNet model as the initial parameters, performs transfer learning on the image. Specifically, the pre-trained VGGNet model is used as a feature extractor, and then fine-tuned on the surrounding rock fragmentation degree recognition task. By freezing the weights of the pre-trained model, only the classifier part is trained, and transfer learning training is performed. After that, the model is applied to new image tasks.

[0061] In this embodiment, high-level features learned by a pre-trained model on large-scale data are used as initial parameters to enable the model to converge and learn features of new tasks more quickly, thereby reducing training time and sample requirements.

[0062] In this embodiment, step S5 is specifically as follows:

[0063] S51. Establish multiple VGGNet network models for identifying the degree of surrounding rock fragmentation, and different scale features correspond to different VGGNet models;

[0064] Among them, Figure 2As shown in the figure, the VGGNet network model includes sequentially connected convolutional layers, pooling layers, fully connected layers and output layers. A small-sized convolution kernel (3x3) is used in the convolutional layer, and the ReLU activation function is used for nonlinear transformation. Batch normalization is added between convolutional layers to accelerate the training process and improve the robustness of the model; the pooling layer is used to reduce the dimension of the features; the fully connected layer is used for classification; and the output layer is used to output the evaluation results.

[0065] S52, adjusting the size and resolution of the image, inputting it into the corresponding VGGNet model, extracting the size and structural features of the surrounding rock crushing area, and inputting the features extracted by the fine-tuned model and the labeled surrounding rock crushing image training set into each VGGNet model for training;

[0066] S53, by means of forward propagation and back propagation algorithms and optimizers, the weights and biases of the VGGNet model are continuously adjusted so that the VGGNet model can learn the characteristics and patterns of the degree of surrounding rock fragmentation, thereby obtaining a trained VGGNet model;

[0067] Furthermore, the gradient of each parameter is calculated from the output layer to the input layer through the back-propagation algorithm, and the stochastic gradient descent optimization algorithm is used to update the parameters of the model according to the direction and magnitude of the gradient to minimize the loss function.

[0068] In this embodiment, by continuously performing forward propagation and back propagation and updating parameters using the stochastic gradient descent algorithm, the VGGNet model will be gradually optimized to improve the accuracy of identifying the degree of surrounding rock fragmentation.

[0069] Optionally, a method for updating model parameters using a stochastic gradient descent optimization algorithm is as follows:

[0070] S5.1. Initialize the parameters of the VGGNet model, including the weights and biases of the convolutional and fully connected layers.

[0071] S5.2, randomly select training samples, input them into the VGGNet model, perform forward propagation, and calculate the loss function;

[0072] Furthermore, the method of randomly selecting training samples may be: using K-fold cross validation to divide the training set into multiple subsets, and selecting data in the subsets as training samples.

[0073] Specifically, K-fold cross validation is used to divide the training set into 10 subsets, 9 of which are used as training data each time, and the remaining 1 subset is used as validation data. The training data is used to train the VGGNet model, and the validation data is used to evaluate the performance of the model.

[0074] S5.3. Calculate the gradient of model parameters according to the loss function;

[0075] S5.4. Update the parameters of the VGGNet model according to the direction and size of the learning rate and gradient;

[0076] Specifically, the update rule of the parameters of the VGGNet model is: new parameters = old parameters - learning rate × gradient;

[0077] S5.5. After traversing all training samples, one training iteration is completed. Repeat multiple iterations until the predetermined number of training rounds is reached or the convergence condition is met, and the optimal parameters of the VGGNet model are obtained.

[0078] In step S6, the performance of each trained VGGNet model is evaluated, and a candidate model with better performance is selected as the trained VGGNet model;

[0079] Among them, the performance evaluation indicators can be indicators such as the accuracy, precision and recall rate of the model.

[0080] Specifically, use the test set and validation set to evaluate the performance of each trained VGGNet model, calculate the model's accuracy, precision, recall and other indicators to evaluate the performance and effect of the model, select candidate models with better performance, and record the performance indicators and evaluation results of each model on the validation set.

[0081] The voting model fusion method is used to fuse the evaluation results to obtain the final evaluation results.

[0082] The above process is described below with a specific example.

[0083] Data preparation:

[0084] Input data: training set is X_train, Y_train, validation set is X_val, Y_val), test set (X_test)

[0085] Training set size: N_train, validation set size: N_val, test set size: N_test

[0086] Model training:

[0087] Model 1 training: Use the training set (X_train, Y_train) for training to obtain the model parameter W1;

[0088] Model 2 training: Use the training set (X_train, Y_train) for training to obtain model parameters W2;

[0089] Model 3 training: Use the training set (X_train, Y_train) for training to obtain model parameters W3;

[0090] Model fusion preparation:

[0091] Make predictions on the validation set:

[0092] Model 1 prediction: Y_val_pred1 = VGGNet1.predict(X_val);

[0093] Model 2 prediction: Y_val_pred2 = VGGNet2.predict(X_val);

[0094] Model 3 prediction: Y_val_pred3 = VGGNet3.predict(X_val);

[0095] Model Fusion:

[0096] Voting prediction: Y_val_pred_vote=vote(Y_val_pred1,Y_val_pred2,Y_val_pred3);

[0097] The calculation process of voting prediction is as follows:

[0098] For each sample, the number of categories appearing in the prediction results of the three models is counted, and the category with the most occurrences is selected as the final prediction result.

[0099] The voting prediction is expressed as:

[0100] Y_val_pred_vote=vote(Y_val_pred1,Y_val_pred2,Y_val_pred3);

[0101] The specific voting model fusion formula is as follows:

[0102] vote(y1,y2,y3)=argmax(count(y1),count(y2),count(y3));

[0103] Among them, y1, y2, y3 represent the prediction results of model 1, model 2 and model 3 respectively, and count(y) represents the number of times the y category appears in the prediction results.

[0104] Model Evaluation:

[0105] Calculate accuracy: accuracy = (1 / N_val) * sum(1{Y_val_pred_vote == Y_val});

[0106] Model tuning:

[0107] Tune the model based on the evaluation results, adjust model parameters, add regularization, increase the amount of data, etc.

[0108] Model Application:

[0109] Make predictions on the test set:

[0110] Model 1 prediction: Y_test_pred1 = VGGNet1.predict(X_test);

[0111] Model 2 prediction: Y_test_pred2 = VGGNet2.predict(X_test);

[0112] Model 3 prediction: Y_test_pred3 = VGGNet3.predict(X_test);

[0113] Voting prediction: Y_test_pred_vote=vote(Y_test_pred1,Y_test_pred2,Y_test_pred3).

[0114] Example 2

[0115] Based on Example 1, this embodiment provides a system for detecting the degree of surrounding rock fragmentation based on machine vision, including:

[0116] Image acquisition module: configured to acquire an image of the tunnel surrounding rock to be identified;

[0117] Preprocessing module: configured to preprocess the acquired image;

[0118] Multi-model evaluation module: configured to transmit the preprocessed images to multiple trained VGGNet models in parallel, each VGGNet model outputs a fragmentation evaluation result, and obtains multiple fragmentation evaluation results;

[0119] Fusion module: configured to fuse the evaluation results using a voting model fusion method to obtain a final evaluation result.

[0120] It should be noted here that the various modules in this embodiment correspond one-to-one to the various steps in Example 1, and the specific implementation process is the same, which will not be repeated here.

[0121] Example 3

[0122] Based on Example 1, this embodiment provides an electronic device, including a memory and a processor, and computer instructions stored in the memory and running on the processor. When the computer instructions are executed by the processor, the steps in the method for detecting the degree of surrounding rock fragmentation based on machine vision described in Example 1 are completed.

[0123] Example 4

[0124] Based on Example 1, this embodiment provides a computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, the steps in the method for detecting the degree of surrounding rock fragmentation based on machine vision described in Example 1 are completed.

[0125] The above description is only a preferred embodiment of the present disclosure and is not intended to limit the present disclosure. For those skilled in the art, the present disclosure may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

[0126] Although the above describes the specific implementation methods of the present disclosure in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present disclosure. Technical personnel in the relevant field should understand that on the basis of the technical solution of the present disclosure, various modifications or variations that can be made by those skilled in the art without creative work are still within the scope of protection of the present disclosure.

Claims

1. A method for detecting the degree of surrounding rock fragmentation based on machine vision, characterized in that: The steps include: Acquire the tunnel surrounding rock image to be identified; Preprocessing the acquired images; The preprocessed images are transmitted in parallel to multiple trained VGGNet models, each VGGNet model outputs a fragmentation evaluation result, and multiple fragmentation evaluation results are obtained; The evaluation results are integrated by using the voting model fusion method to obtain the final evaluation result; The voting model fusion method is as follows: for each sample, the number of fragmentation degree categories appearing in the prediction results of multiple models is counted, and the category with the most occurrences is selected as the final prediction result; The fragmentation assessment results are visualized and a scatter plot is used to display the fragmentation assessment results of each image. The horizontal axis represents the image index and the vertical axis represents the fragmentation assessment results. The VGGNet model training process includes the following steps: S1. Obtain tunnel surrounding rock crushing image data to build a data set and annotate the image; S2, preprocess the images in the dataset, including denoising, image enhancement, image alignment, rotation and flipping; S3, normalizing the preprocessed image; S4. Use the pre-trained VGGNet model as the initial parameters to perform transfer learning on the image to obtain a fine-tuned model, which extracts the features required for surrounding rock fragmentation assessment; S5. Construct multiple VGGNet models, extract features of different scales for evaluating the degree of surrounding rock fragmentation, and use different scale features to correspond to different VGGNet models. Use the features extracted by the fine-tuned model and the labeled image data to train multiple VGGNet models, establish a correlation model between the degree of surrounding rock fragmentation and image features, and obtain the trained VGGNet model. S6, evaluate the performance of each trained VGGNet model, select the candidate model as the trained VGGNet model, and use the voting model fusion method to evaluate the degree of surrounding rock fragmentation; Using the pre-trained VGGNet model as the initial parameters, transfer learning is performed on the image as follows: the pre-trained VGGNet model is used as a feature extractor, and then fine-tuned on the surrounding rock fragmentation degree recognition task. Transfer learning training is performed by freezing the weights of the pre-trained model and training the weights of the classifier part.

2. The method for detecting the degree of surrounding rock fragmentation based on machine vision according to claim 1, characterized in that: Pre-processing methods include: The image is denoised using mean filtering; Histogram equalization is used to redistribute the grayscale distribution of pixels, enhance the contrast of the image, and enhance the image; Feature point matching is used to detect and match feature points in the image to achieve image alignment; rotating and reversing the image increases data diversity and improves the generalization ability of the model.

3. The method for detecting the degree of surrounding rock fragmentation based on machine vision according to claim 1, characterized in that: Step S5 is specifically as follows: Establish multiple VGGNet network models for identifying the degree of surrounding rock fragmentation, with different scale features corresponding to different VGGNet models; Adjust the size and resolution of the image and input it into the corresponding VGGNet model to extract the size and structural features of the surrounding rock fragmentation area. Input the features extracted by the fine-tuned model and the labeled surrounding rock fragmentation image training set into each VGGNet model for training; Through the forward propagation and back propagation algorithms and optimizers, the weights and biases of the VGGNet model are continuously adjusted so that the VGGNet model can learn the characteristics and patterns of the degree of surrounding rock fragmentation and obtain the trained VGGNet model.

4. The method for detecting the degree of surrounding rock fragmentation based on machine vision according to claim 3, characterized in that: The back-propagation algorithm is used to calculate the gradient of each parameter from the output layer to the input layer, and the stochastic gradient descent optimization algorithm is used to update the model parameters according to the direction and magnitude of the gradient to minimize the loss function.

5. The detection system of surrounding rock fragmentation degree based on machine vision is characterized by: include: Image acquisition module: configured to acquire an image of the tunnel surrounding rock to be identified; Preprocessing module: configured to preprocess the acquired image; Multi-model evaluation module: configured to transmit the preprocessed images to multiple trained VGGNet models in parallel, each VGGNet model outputs a fragmentation evaluation result, and obtains multiple fragmentation evaluation results; Fusion module: configured to fuse the evaluation results using a voting model fusion method to obtain a final evaluation result; The voting model fusion method is as follows: for each sample, the number of fragmentation degree categories appearing in the prediction results of multiple models is counted, and the category with the most occurrences is selected as the final prediction result; The fragmentation assessment results are visualized and a scatter plot is used to display the fragmentation assessment results of each image. The horizontal axis represents the image index and the vertical axis represents the fragmentation assessment results. The VGGNet model training process includes the following steps: S1. Obtain tunnel surrounding rock crushing image data to build a data set and annotate the image; S2, preprocess the images in the dataset, including denoising, image enhancement, image alignment, rotation and flipping; S3, normalizing the preprocessed image; S4. Use the pre-trained VGGNet model as the initial parameters to perform transfer learning on the image to obtain a fine-tuned model, which extracts the features required for surrounding rock fragmentation assessment; S5. Construct multiple VGGNet models, extract features of different scales for evaluating the degree of surrounding rock fragmentation, and use different scale features to correspond to different VGGNet models. Use the features extracted by the fine-tuned model and the labeled image data to train multiple VGGNet models, establish a correlation model between the degree of surrounding rock fragmentation and image features, and obtain the trained VGGNet model. S6, evaluate the performance of each trained VGGNet model, select the candidate model as the trained VGGNet model, and use the voting model fusion method to evaluate the degree of surrounding rock fragmentation; Using the pre-trained VGGNet model as the initial parameters, transfer learning is performed on the image as follows: the pre-trained VGGNet model is used as a feature extractor, and then fine-tuned on the surrounding rock fragmentation degree recognition task. Transfer learning training is performed by freezing the weights of the pre-trained model and training the weights of the classifier part.

6. An electronic device, characterized in that: The invention comprises a memory and a processor and computer instructions stored in the memory and executed on the processor. When the computer instructions are executed by the processor, the steps in the method for detecting the degree of surrounding rock fragmentation based on machine vision as described in any one of claims 1 to 4 are completed.

7. A computer-readable storage medium, characterized in that: Used to store computer instructions, which, when executed by a processor, complete the steps of the method for detecting the degree of surrounding rock fragmentation based on machine vision as described in any one of claims 1 to 4.

Images