Drill hole image rock mass structure identification method

A deep learning-based method with adaptive learning strategies enhances borehole image analysis by leveraging similar geological features, improving detection accuracy and robustness through phased training.

CN119091228AActive Publication Date: 2024-12-06INST OF ROCK & SOIL MECHANICS CHINESE ACAD OF SCI
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202411286021.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-13
Publication Date
2024-12-06
Estimated Expiration
2044-09-13

AI Technical Summary

Technical Problem

Current methods for interpreting borehole images rely heavily on manual recognition, leading to low efficiency, high workload, and poor accuracy, while traditional image detection methods and neural networks fail to leverage the similarity of geological conditions in the same area for improved detection.

Method used

A deep learning-based method using an adaptive learning strategy with a target detection network model, comprising a backbone, regional, and classification network, with adaptive layers to utilize similar features of rock structural surfaces, and a phased training approach to enhance detection accuracy and robustness.

Benefits of technology

The method significantly improves detection accuracy and robustness by leveraging similar geological features, stabilizing training, and avoiding overfitting, while maintaining high recognition performance across different tasks.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119091228A_ABST
    Figure CN119091228A_ABST
Patent Text Reader

Abstract

The invention provides a borehole image rock mass structure identification method, and relates to the technical field of geological exploration, and the method comprises the steps: constructing a target detection network model; collecting a drilling image data set, marking a target frame and a category of a rock mass structure, and making a sample set; initializing a target detection network model; freezing the adaptive network, and performing initial training on the target detection network model by using the sample set until the model converges to obtain an initially trained target detection network model; and obtaining a drilling image task set, carrying out task identification on the drilling image task set by using the initially trained target detection network model, unfreezing the adaptive network after a preset number of tasks is completed, and carrying out iterative training on the whole model by using the drilling image task set until the model converges, thereby obtaining a final target detection network model. According to the method, adaptive learning is introduced into a target detection task, so that efficient model training and strategy updating are realized while the detection precision and robustness are improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of geological exploration, and in particular to a method for identifying rock mass structure using a borehole image. Background Art

[0002] my country has a vast territory and its geological structure is very complex and diverse. With the continuous development of economic construction and the continuous advancement of engineering technology, detailed understanding of the geological structure of underground rock mass has become an irreplaceable task in various types of construction, geotechnical, geological, water conservancy and other projects, which is of great significance for the rational selection of engineering methods and the safety of engineering.

[0003] Borehole imaging technology is a technical method that uses optical principles and related equipment to obtain borehole wall images in the borehole. Using borehole imaging technology to obtain borehole images can accurately and quickly obtain rock structure information in the borehole. However, the current method of interpreting the structural surface of borehole images mainly relies on manual recognition, which has the following problems.

[0004] (1) At present, the method of analyzing borehole images and obtaining in-situ geological body structural surface information is generally that technicians subjectively identify the structural surface in the borehole image, manually read the structural surface feature point information, and use a computer program to fit it into a standard curve, and calibrate the geometric parameters of the structural surface through the parameters of the curve. This method has a large workload and low efficiency; at the same time, it relies too much on the experience of technicians and has poor accuracy and reliability.

[0005] (2) Traditional image detection methods such as filter detection, edge extraction, and template matching have poor application effects in rock structure detection of borehole images, low detection accuracy, and high false alarm rate. Due to the large differences between different borehole images, traditional image detection methods require manual definition of many hyperparameters. The hyperparameters vary greatly between boreholes, and the parameters need to be manually determined before detection. This method is difficult to improve detection efficiency and accuracy.

[0006] (3) In recent years, a series of structural surface detection methods based on neural networks have also emerged. Due to the similar geological conditions in the same area, the rock structure reflected in the borehole images obtained from engineering surveys also has similar characteristics in terms of lithology, failure mode, occurrence, etc. This feature helps to improve the accuracy and precision of rock structure identification. The current borehole image rock structure identification method does not take this feature into account. Summary of the invention

[0007] In view of this, the present invention proposes a method for identifying rock structure in borehole images, which performs target detection on structural surfaces on borehole images based on deep learning, and adopts an adaptive learning-based approach to fully utilize similar features of rock structural surfaces in the same area to improve target detection accuracy.

[0008] The technical solution of the present invention is achieved in this way:

[0009] The present invention provides a method for identifying rock mass structure using a borehole image, comprising the following steps:

[0010] S1 builds the target detection network model, including the backbone network, regional network, adaptive network and classification network;

[0011] S2 collects the borehole image dataset, annotates the target boxes and categories of the rock mass structure, and produces a sample set;

[0012] S3 initializes the target detection network model;

[0013] S4 freezes the adaptive network, uses the sample set to perform initial training on the target detection network model, and iteratively updates the backbone network, regional network, and classification network according to the loss function until the model converges to obtain the initially trained target detection network model;

[0014] S5 obtains the drilling image task set, and uses the initially trained target detection network model to perform recognition tasks on the drilling image task set. After completing the preset number of tasks, the adaptive network is unfrozen, and the entire model is iteratively trained using the drilling image task set until the model converges to obtain the final target detection network model.

[0015] On the basis of the above technical solution, preferably, the drilling image task set and the sample set are image data of the same batch in the same area.

[0016] Based on the above technical solution, preferably, the backbone network adopts the ResNet-50 network and removes the last fully connected layer and softmax layer, the regional network includes a convolutional layer, an anchor box generator, a bounding box regression layer, a target scoring layer and a ROI pooling layer, the adaptive network includes a hidden layer A and a hidden layer B, and the classification network includes a fully connected layer, a classification layer and a bounding box regression layer.

[0017] Based on the above technical solution, preferably, in the adaptive network, the weight matrix and bias vector of the hidden layer A and the hidden layer B are expressed as follows:

[0018]

[0019]

[0020] Where W A is the weight matrix of hidden layer A, b A is the bias vector of hidden layer A, W B is the weight matrix of hidden layer B, b B is the bias vector of the hidden layer B, R represents the real number field, dinput Represents the dimension of the input feature vector, d A represents the number of neurons in the hidden layer A, d B Represents the number of neurons in the hidden layer B.

[0021] On the basis of the above technical solution, preferably, when initializing the target detection network model, the weight matrix W of the hidden layer A is A and the weight matrix W of the hidden layer B B All are initialized to all-1 matrices.

[0022] On the basis of the above technical solution, preferably, in each training of the target detection network model, forward propagation is used to calculate the output of each layer to obtain the prediction result, and the gradient of the loss function with respect to the network parameters is calculated on the basis of the forward propagation, and back propagation is performed to update the weight matrix and bias vector of each layer, and the model is judged to have converged by continuous iteration of forward propagation and back propagation until the value of the loss function reaches the minimum.

[0023] Based on the above technical solution, preferably, during forward propagation, the outputs of hidden layer A and hidden layer B are:

[0024] a=h A (W A ·x)

[0025] b=h B (W B a)

[0026] h A (z)=max(0,z)

[0027] h B (z) = sigmoid(z)

[0028] In the formula, a represents the output of hidden layer A, b represents the output of hidden layer B, and W A is the weight matrix of the hidden layer A, W B is the weight matrix of the hidden layer B, h A represents the activation function of the hidden layer A, h B represents the activation function of the hidden layer B, x represents the input feature vector, and z represents the input value of the activation function.

[0029] Based on the above technical solution, preferably, in each training, the gradient calculation method of the loss function for the hidden layer A and the hidden layer B is:

[0030]

[0031] In the formula, Represents the weight matrix W of the loss function L for the hidden layer AA The partial derivative of Represents the weight matrix W of the loss function L on the hidden layer B B The partial derivative of Represents the bias vector b of the loss function L on the hidden layer A A The partial derivative of Represents the bias vector b of the loss function L on the hidden layer B B The partial derivative of represents the partial derivative of the loss function L with respect to the output a of the hidden layer A, Represents the weight matrix W of the output a of hidden layer A to hidden layer A A The partial derivative of represents the partial derivative of the loss function L with respect to the output b of the hidden layer B, Represents the weight matrix W of the output b of hidden layer B to hidden layer B B The partial derivative of Represents the bias vector b of the output a of hidden layer A to hidden layer A A The partial derivative of Represents the bias vector b of the output b of hidden layer B to hidden layer B B The partial derivative of , where and By back-propagation, and is the derivative of the activation function, and its value depends on the input of hidden layer A and hidden layer B.

[0032] Based on the above technical solution, preferably, the update formulas of hidden layer A and hidden layer B are:

[0033]

[0034] In the formula, α represents the learning rate.

[0035] Based on the above technical solution, preferably, the calculation formula of the loss function is:

[0036]

[0037] Where L represents the loss function, N is the batch size of the data, C is the number of categories, and y i,j is the true result. When the i-th data belongs to the j-th category, y i,j =1, otherwise, y i,j =0, is the prediction result, that is, the probability that the model predicts that the i-th data belongs to the j-th category.

[0038] The system of the present invention has the following beneficial effects compared with the prior art:

[0039] By introducing adaptive learning into the target detection task, efficient model training and updating strategies are implemented while improving detection accuracy and robustness.

[0040] By adding two fully connected hidden layers and designing an adaptive learning strategy for them, the similar features of rock mass structural surfaces in the same area are fully utilized to significantly improve the accuracy and robustness of target detection;

[0041] Adaptive learning strategies can adjust model parameters according to new data, so that the model can maintain high recognition performance in different batches of tasks and has strong adaptability.

[0042] Freezing the initial stage of the hidden layer can stabilize the training of other layers, which not only speeds up the convergence of the model but also avoids the overfitting problem in the absence of sufficient data. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0044] Figure 1 is a flow chart of the method of the present invention;

[0045] Figure 2 This is a model architecture diagram of the present invention;

[0046] Figure 3 This is the model training graph of the present invention. DETAILED DESCRIPTION

[0047] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0048] See also Figure 1 The method for identifying rock mass structure in borehole images provided by the present invention comprises the following steps:

[0049] S1 builds the target detection network model, including the backbone network, regional network, adaptive network and classification network;

[0050] S2 collects the borehole image dataset, annotates the target boxes and categories of the rock mass structure, and produces a sample set;

[0051] S3 initializes the target detection network model;

[0052] S4 freezes the adaptive network, uses the sample set to perform initial training on the target detection network model, and iteratively updates the backbone network, regional network, and classification network according to the loss function until the model converges to obtain the initially trained target detection network model;

[0053] S5 obtains the drilling image task set, and uses the initially trained target detection network model to perform recognition tasks on the drilling image task set. After completing the preset number of tasks, the adaptive network is unfrozen, and the entire model is iteratively trained using the drilling image task set until the model converges to obtain the final target detection network model.

[0054] like Figure 2 As shown, the technical implementation process of the present invention is as follows: a target detection network model is designed, including a backbone network, a regional network, an adaptive network and a classification network. The backbone network adopts a ResNet-50 network and removes the last fully connected layer and softmax layer. The regional network includes a convolutional layer, an anchor box generator, a bounding box regression layer, a target scoring layer and a ROI pooling layer. The adaptive network includes a hidden layer A and a hidden layer B. The classification network includes a fully connected layer, a classification layer and a bounding box regression layer. First, a picture of any size is input into the backbone feature extraction network to obtain a feature map. The original picture is searched to obtain a regional proposal box, and each corresponding feature box is found in the feature map. Each feature box is pooled to a uniform size in the ROI pooling layer. After that, the data is processed sequentially by two fully connected hidden layers A and B. This is an additional hidden layer added in the proposed scheme, which is used to use the similarity of the target features of the structural surface in the same area to carry out network adaptive learning. Finally, the data is subjected to softmax classification and bbox regression respectively to obtain the target classification and target box.

[0055] Specifically, Figure 3 As shown, in one embodiment of the present invention, the training strategy of the model is as follows:

[0056] Create a dataset of rock mass structural surfaces in borehole images. Manually mark rock mass structural surfaces on a certain number of borehole images, which are the recognition targets. Conduct training work on this dataset.

[0057] First, the hidden layers A and B are frozen to keep the weights of the hidden layers A and B unchanged at the initial values. The batch gradient descent method is used to update the weights of other layers of the network, and the network is trained until the model converges.

[0058] The model is used to perform borehole image recognition tasks of the same batch in the same area. After a certain number of tasks are executed, the hidden layers A and B are unfrozen, and the weights of the hidden layers A and B are updated using the task data of the batch. In this way, the adaptive learning of the model is achieved.

[0059] Since the geological conditions in the same area are similar, the rock mass structure reflected in the borehole images obtained by engineering survey also has similar characteristics in terms of lithology, failure mode, occurrence, etc. Therefore, the borehole image task set and sample set collected by the present invention are image data of the same batch in the same area.

[0060] Let's take a specific implementation example to illustrate:

[0061] 1. Collect borehole image datasets, annotate the target boxes and categories of rock structures, and create sample sets.

[0062] 2. Initialize the target detection network model and set the weight matrix W of the hidden layer A A and the weight matrix W of the hidden layer B B All are initialized to full 1 matrices. To simplify the model and avoid the same features learned across layers, the bias vectors of hidden layer A and hidden layer B can be initialized to full 0 vectors to help the network update parameters more smoothly at the beginning of training.

[0063] 3. Use the sample set to perform initial training on the target detection network model. At this time, freeze the hidden layers A and B (i.e., do not update their weights and biases), and train the remaining layers normally. Freezing the hidden layers A and B can prevent the model from overfitting the data in the initial training phase.

[0064] In each training, forward propagation is used to calculate the output of each layer to obtain the prediction result. Based on the forward propagation, the gradient of the loss function with respect to the network parameters is calculated, and back propagation is performed to update the weight matrix and bias vector of each layer. Through continuous iteration of forward propagation and back propagation, the model is judged to have converged until the value of the loss function reaches the minimum.

[0065] During forward propagation, the input image is passed through the feature extraction network to obtain a feature map. The search generates a region proposal (RoI) and finds the corresponding feature box on the feature map. ROI pooling is performed to pool the feature boxes to a uniform size. The feature data passes through two fully connected hidden layers A and B. At this time, the weights and biases are not updated. This is to ensure that the model can first learn some basic features without being affected by hidden layers A and B.

[0066] The weight matrix and bias vector of hidden layer A and hidden layer B are expressed as follows:

[0067]

[0068]

[0069] Where W A is the weight matrix of hidden layer A, b A is the bias vector of hidden layer A, W B is the weight matrix of hidden layer B, b B is the bias vector of the hidden layer B, R represents the real number field, d input Represents the dimension of the input feature vector, d A represents the number of neurons in the hidden layer A, d B Represents the number of neurons in the hidden layer B.

[0070] The outputs of hidden layer A and hidden layer B are:

[0071] a=h A (W A ·x)

[0072] b=h B (W B a)

[0073] h A (z)=max(0,z)

[0074] h B (z) = sigmoid(z)

[0075] In the formula, a represents the output of hidden layer A, b represents the output of hidden layer B, and W A is the weight matrix of the hidden layer A, W B is the weight matrix of the hidden layer B, h A represents the activation function of the hidden layer A, h B represents the activation function of the hidden layer B, x represents the input feature vector, and z represents the input value of the activation function.

[0076] Then classification and bounding box (bbox) regression are performed.

[0077] The loss is calculated using the following loss function:

[0078]

[0079] Where L represents the loss function, N is the batch size of the data, C is the number of categories, and y i,j is the true result. When the i-th data belongs to the j-th category, y i,j =1, otherwise, y i,j =0, is the prediction result, that is, the probability that the model predicts that the i-th data belongs to the j-th category.

[0080] Calculate the partial derivatives of the loss function L for all non-frozen layer parameters (backbone network, regional network, and classification network) and perform back propagation. Use the gradient descent method to update the weights and biases of all non-frozen layers until the model converges on the sample set.

[0081] 4. Obtain a drilling image task set. In one embodiment, the drilling image task set is labeled, and supervised learning is used to further train the model. Labeled data can directly tell the model what the correct output is, thereby more accurately updating parameters. In another embodiment, the drilling image task set is unlabeled. In this case, an unsupervised learning or semi-supervised learning strategy is adopted. Unsupervised learning: Use the initially trained model to predict the unlabeled data and generate pseudo labels. These pseudo labels are then used as temporary labels to continue supervised training. Semi-supervised learning: Similar to the pseudo-label method, the model makes inferences on unlabeled data, and then uses high-confidence prediction results as new training data. This method gradually improves model performance in iterations.

[0082] 5. Set the task quantity interval, such as checking whether adaptive learning and retraining are needed after executing a batch or a certain number of recognition tasks.

[0083] By regularly judging the task stage, the model can be updated adaptively. After completing a certain number of tasks, the new data batch will expose the model to more new feature information, unfreeze and retrain the hidden layers A and B, so that their weights and biases can be actively adjusted to better fit the new task requirements. By switching the training strategy in stages, the model can be continuously optimized and gradually improve its performance and accuracy on specific tasks. After each stage, the hidden layer is retrained so that it can absorb and adapt to the features contained in the new data, avoiding the model from being too dependent on the initial weight settings.

[0084] 6. Adaptive learning phase. Unfreeze the hidden layers A and B so that their weights and biases can be updated. Train the entire model using the drilling image task set.

[0085] Specifically, during training, the forward propagation process is as follows: the input image generates a feature map through the feature extraction network, passes through the ROI pooling layer, passes through two fully connected hidden layers A and B, and then to the classification layer and regression layer to calculate the loss. The weight matrix and bias vector of hidden layer A and hidden layer B are expressed as follows:

[0086]

[0087] Where W A is the weight matrix of hidden layer A, b A is the bias vector of hidden layer A, W B is the weight matrix of hidden layer B, b Bis the bias vector of the hidden layer B, R represents the real number field, d input Represents the dimension of the input feature vector, d A represents the number of neurons in the hidden layer A, d B Represents the number of neurons in the hidden layer B.

[0088] The outputs of hidden layer A and hidden layer B are:

[0089] a=h A (W A x)

[0090] b=h B (W B a)

[0091] h A (z)=max(0,z)

[0092] h B (z) = sigmoid(z)

[0093] In the formula, a represents the output of hidden layer A, b represents the output of hidden layer B, and W A is the weight matrix of the hidden layer A, W B is the weight matrix of the hidden layer B, h A represents the activation function of the hidden layer A, h B represents the activation function of the hidden layer B, x represents the input feature vector, and z represents the input value of the activation function.

[0094] Then classification and bounding box (bbox) regression are performed.

[0095] The loss is calculated using the following loss function:

[0096]

[0097] Where L represents the loss function, N is the batch size of the data, C is the number of categories, and y i,j is the true result. When the i-th data belongs to the j-th category, y i,j =1, otherwise, y i,j =0, is the prediction result, that is, the probability that the model predicts that the i-th data belongs to the j-th category.

[0098] Calculate the partial derivatives of the loss function with respect to the weights and biases of all layers (including hidden layers A and B), and update the parameters by backpropagation.

[0099] The gradient calculation method of the loss function for hidden layer A and hidden layer B is:

[0100]

[0101]

[0102] In the formula, Represents the weight matrix W of the loss function L for the hidden layer A A The partial derivative of Represents the weight matrix W of the loss function L on the hidden layer B B The partial derivative of Represents the bias vector b of the loss function L on the hidden layer A A The partial derivative of Represents the bias vector b of the loss function L on the hidden layer B B The partial derivative of represents the partial derivative of the loss function L with respect to the output a of the hidden layer A, Represents the weight matrix W of the output a of hidden layer A to hidden layer A A The partial derivative of represents the partial derivative of the loss function L with respect to the output b of the hidden layer B, Represents the weight matrix W of the output b of hidden layer B to hidden layer B B The partial derivative of Represents the bias vector b of the output a of hidden layer A to hidden layer A A The partial derivative of Represents the bias vector b of the output b of hidden layer B to hidden layer B B The partial derivative of , where and By back-propagation, and is the derivative of the activation function, and its value depends on the input of hidden layer A and hidden layer B.

[0103] The update formulas for hidden layer A and hidden layer B are:

[0104]

[0105] In the formula, α represents the learning rate.

[0106] Repeat the above steps for iterative training until the model converges.

[0107] The present invention performs training in a phased approach. During the initial training, the weights of hidden layers A and B are frozen. The goal is to fully train other layers until the model converges, improving the stability and accuracy of the initial training. When a certain number of tasks are performed, when the model is used in the same area, the weights of hidden layers A and B are unfrozen, and the batch of data is used to update hidden layers A and B. This operation allows the model to be fine-tuned for new data, further improving detection accuracy through adaptive learning.

[0108] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for identifying rock mass structure from borehole images, characterized in that: The steps include: S1 builds the target detection network model, including the backbone network, regional network, adaptive network and classification network; S2 collects the borehole image dataset, annotates the target boxes and categories of the rock mass structure, and produces a sample set; S3 initializes the target detection network model; S4 freezes the adaptive network, uses the sample set to perform initial training on the target detection network model, and iteratively updates the backbone network, regional network, and classification network according to the loss function until the model converges to obtain the initially trained target detection network model; S5 obtains the drilling image task set, and uses the initially trained target detection network model to perform recognition tasks on the drilling image task set. After completing the preset number of tasks, the adaptive network is unfrozen, and the entire model is iteratively trained using the drilling image task set until the model converges to obtain the final target detection network model.

2. The method for identifying rock mass structure using borehole images according to claim 1, characterized in that: The borehole image task set and sample set are image data of the same batch in the same area.

3. The method for identifying rock mass structure using borehole images according to claim 1, characterized in that: The backbone network uses the ResNet-50 network and removes the last fully connected layer and softmax layer. The regional network includes a convolutional layer, an anchor box generator, a bounding box regression layer, a target scoring layer, and a ROI pooling layer. The adaptive network includes hidden layer A and hidden layer B. The classification network includes a fully connected layer, a classification layer, and a bounding box regression layer.

4. The method for identifying rock mass structure using borehole images according to claim 1, characterized in that: In the adaptive network, the weight matrix and bias vector of hidden layer A and hidden layer B are expressed as follows: Where W A is the weight matrix of hidden layer A, b A is the bias vector of hidden layer A, W B is the weight matrix of hidden layer B, b B is the bias vector of the hidden layer B, R represents the real number field, d input Represents the dimension of the input feature vector, d A represents the number of neurons in the hidden layer A, d B Represents the number of neurons in the hidden layer B.

5. The method for identifying rock mass structure using borehole images according to claim 4, characterized in that: When initializing the target detection network model, the weight matrix W of the hidden layer A is A and the weight matrix W of the hidden layer B B All are initialized to all-1 matrices.

6. The method for identifying rock mass structure from borehole images according to claim 4, characterized in that: In each training of the target detection network model, forward propagation is used to calculate the output of each layer to obtain the prediction result. Based on the forward propagation, the gradient of the loss function with respect to the network parameters is calculated, and back propagation is performed to update the weight matrix and bias vector of each layer. Through continuous iteration of forward propagation and back propagation, the model is judged to have converged until the value of the loss function reaches the minimum.

7. The method for identifying rock mass structure from borehole images according to claim 6, characterized in that: During forward propagation, the outputs of hidden layer A and hidden layer B are: a=h A (W A ·x) b=h B (W B ·a) h A (z)=max(0,z) h B (z)=sigmoid(z) In the formula, a represents the output of hidden layer A, b represents the output of hidden layer B, and W A is the weight matrix of the hidden layer A, W B is the weight matrix of the hidden layer B, h A represents the activation function of the hidden layer A, h B represents the activation function of the hidden layer B, x represents the input feature vector, and z represents the input value of the activation function.

8. The method for identifying rock mass structure using borehole images according to claim 7, characterized in that: In each training, the gradient of the loss function for hidden layer A and hidden layer B is calculated as: In the formula, Represents the weight matrix W of the loss function L for the hidden layer A A The partial derivative of Represents the weight matrix W of the loss function L on the hidden layer B B The partial derivative of Represents the bias vector b of the loss function L on the hidden layer A A The partial derivative of Represents the bias vector b of the loss function L on the hidden layer B B The partial derivative of represents the partial derivative of the loss function L with respect to the output a of the hidden layer A, Represents the weight matrix W of the output a of hidden layer A to hidden layer A A The partial derivative of represents the partial derivative of the loss function L with respect to the output b of the hidden layer B, Represents the weight matrix W of the output b of hidden layer B to hidden layer B B The partial derivative of Represents the bias vector b of the output a of hidden layer A to hidden layer A A The partial derivative of Represents the bias vector b of the output b of hidden layer B to hidden layer B B The partial derivative of , where and By back-propagation, and is the derivative of the activation function, and its value depends on the input of hidden layer A and hidden layer B.

9. The method for identifying rock mass structure from borehole images according to claim 8, characterized in that: The update formulas for hidden layer A and hidden layer B are: In the formula, α represents the learning rate.

10. The method for identifying rock mass structure from borehole images according to claim 8, characterized in that: The loss function is calculated as: Where L represents the loss function, N is the batch size of the data, C is the number of categories, and y i,j is the true result. When the i-th data belongs to the j-th category, y i,j =1, otherwise, y i,j =0, is the prediction result, that is, the probability that the model predicts that the i-th data belongs to the j-th category.

Citation Information

Patent Citations

  • Drilling image structural plane automatic identification and parameter extraction method

    CN104915640A

  • Deep learning and transfer learning-based rock stratum structure intelligent detection and classification method

    CN112836075A

  • Target detection method for remote sensing image noise labeling based on adaptive weight

    CN116343050A

  • Intelligent grading method for tunnel surrounding rock based on image information

    CN117853323A