Rock mass fracture identification method and device, electronic equipment and readable storage medium

Abstract

The invention provides a rock mass fracture recognition method and device, electronic equipment and a readable storage medium, and belongs to the field of mine rock mass intelligent recognition. The method comprises the following steps: inputting an obtained rock mass image to be detected into a fully trained rock mass fracture recognition network, carrying out feature coding on the rock mass image to be detected step by step to obtain a plurality of multi-level coding features, and carrying out double-branch attention enhancement on the multi-level coding features to obtain corresponding double-branch enhancement features, carrying out feature decoding on the multi-level coding features step by step and fusing corresponding-level double-branch enhancement features to obtain decoding output features; and predicting and outputting the decoded output features to obtain a rock mass fracture identification result. According to the method, through double-branch attention enhancement, the advantages of space channel attention and self-attention are combined, image context information can be globally integrated while the modeling ability and the spatial position sensing ability for different features are enhanced, and the accuracy of the rock mass fracture recognition method is effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent identification of mining rock masses, and in particular to a rock mass fracture identification method, device, electronic equipment and readable storage medium. Background Art

[0002] In mining engineering construction, cracks can cut the complete rock mass into pieces, forming potential slip surfaces, causing structural damage to the rock mass, while providing seepage channels for water and accelerating the weathering process, which has a serious impact on the stability of the rock mass. Traditional methods such as on-site investigation and manual measurement have many problems such as low efficiency, long time consumption, low accuracy and prominent safety hazards, which are difficult to meet the needs of the rapid development of mining and engineering construction.

[0003] Therefore, with the development of deep learning in the direction of image recognition, it can be applied to the task of rock crack identification. In rock crack identification, the existing technology usually takes into account that different crack patterns often vary greatly in shape and size, so the model is optimized to adapt to different types of crack patterns. However, in the task of rock crack identification, there are often connections between cracks in different areas. For example, if a certain area of ​​a rock mass is affected by the environment and cracks are generated, then other areas of the rock mass often also have cracks, and these cracks are affected by the weathering process, etc., and these cracks may also be connected by patterns. However, in the task of crack identification, the existing deep learning algorithms often only consider the characteristics of different cracks themselves, but do not consider the contextual connection between the crack characteristics in the global range, resulting in low accuracy in rock crack identification.

[0004] Therefore, the existing technology has a technical problem that in rock fracture identification, only the characteristics of different fractures themselves are considered, but the contextual connection of fracture characteristics in the global scope is not considered, which leads to low accuracy of rock fracture identification and needs to be improved. Summary of the invention

[0005] In view of this, it is necessary to provide a rock fracture identification method, device, electronic device and readable storage medium to solve the technical problem that in the prior art, only the characteristics of different fractures themselves are considered in rock fracture identification, while the contextual connection of fracture characteristics in the global scope is not considered, resulting in low accuracy of rock fracture identification.

[0006] In order to solve the above problems, on the one hand, the present invention provides a rock mass fracture identification method, comprising: The acquired rock mass image to be tested is input into a well-trained rock mass crack recognition network, and the rock mass image to be tested is feature encoded level by level to obtain a number of multi-level encoding features, and the multi-level encoding features are subjected to dual-branch attention enhancement to obtain corresponding dual-branch enhanced features, and the multi-level encoding features are feature decoded level by level and the dual-branch enhanced features of the corresponding levels are fused to obtain the decoded output features; The decoded output features are predicted and output to obtain the rock fracture identification results.

[0007] In a possible implementation, the rock mass image to be tested is subjected to step-by-step feature coding to obtain a number of multi-level coding features, including: Perform feature convolution, ReLU activation function and maximum pooling on the rock mass image to be tested step by step to obtain multi-level encoding features; Among them, the output features of each level of maximum pooling are used as the input features of the next level of feature convolution.

[0008] In a possible implementation, dual-branch attention enhancement is performed on the multi-level encoding features to obtain corresponding dual-branch enhanced features, including: Perform feature convolution and batch normalization on the multi-level encoding features to obtain the features to be enhanced; Perform spatial channel attention enhancement on the feature to be enhanced to obtain the first attention feature; Perform self-attention enhancement on the feature to be enhanced to obtain the second attention feature; Fusing the first attention feature and the second attention feature to obtain a fused attention feature; The fused attention features are convolved and batch normalized, and then concatenated with multi-level encoding features to obtain dual-branch enhanced features.

[0009] In a possible implementation, performing spatial channel attention enhancement on the feature to be enhanced to obtain a first attention feature includes: After performing channel dimension attention extraction on the features to be enhanced, merge the features to be enhanced to obtain preliminary enhanced features; After performing spatial dimension attention extraction on the preliminary enhanced features, the preliminary enhanced features are merged to obtain the first attention feature.

[0010] In a possible implementation, performing self-attention enhancement on the feature to be enhanced to obtain a second attention feature includes: The features to be enhanced are sequentially down-sampled, self-attention extracted and up-sampled to obtain the second attention features.

[0011] In a possible implementation, the multi-level coding features are decoded level by level and the corresponding level double-branch enhanced features are fused to obtain the decoded output features, including: The last level of multi-level coding features is upsampled level by level, and the corresponding level double-branch enhanced features and feature convolution are fused to obtain the decoded output features.

[0012] In a possible implementation, the trained complete rock mass fracture identification network is obtained by training an initial rock mass fracture identification network, and the training of the initial rock mass fracture identification network includes: The acquired rock fracture training data is input into the initial rock fracture identification network, and the parameters of the initial rock fracture identification network are optimized by gradient descent method based on the adaptive matrix estimation optimizer. The well-trained rock fracture identification network is obtained through the iterative optimization process.

[0013] On the other hand, the present invention also provides a rock mass fracture identification device, comprising: The feature extraction unit is used to input the acquired rock mass image to be tested into a well-trained rock mass fracture recognition network, perform feature encoding on the rock mass image to be tested level by level to obtain a number of multi-level encoding features, perform dual-branch attention enhancement on the multi-level encoding features to obtain corresponding dual-branch enhanced features, perform feature decoding on the multi-level encoding features level by level and fuse the dual-branch enhanced features of the corresponding levels to obtain decoded output features; The prediction output unit is used to predict and output the decoded output features to obtain rock fracture identification results.

[0014] On the other hand, the present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the above-mentioned rock fracture identification method is implemented.

[0015] On the other hand, the present invention further provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the above-mentioned rock fracture identification method is implemented.

[0016] The beneficial effects of the present invention are as follows: in the rock mass crack identification method provided by the present invention, the acquired rock mass image to be tested is first input into a well-trained rock mass crack identification network, the rock mass image to be tested is feature encoded step by step to obtain a number of multi-level coded features, the multi-level coded features are enhanced by dual-branch attention to obtain corresponding dual-branch enhanced features, the multi-level coded features are feature decoded step by step and the dual-branch enhanced features of the corresponding levels are fused to obtain decoded output features; then the decoded output features are predicted and output to obtain the rock mass crack identification results. The present invention combines the advantages of spatial channel attention and self-attention through dual-branch attention enhancement, and can globally integrate image context information while enhancing the modeling capabilities and spatial position perception capabilities of different features, thereby effectively improving the accuracy of the rock mass crack identification method. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0018] Figure 1 A schematic diagram of a flow chart of an embodiment of a rock mass fracture identification method provided by the present invention; Figure 2 A schematic diagram of a process of dual-branch attention enhancement according to an embodiment of the present invention; Figure 3 A schematic diagram of a process of enhancing spatial channel attention according to an embodiment of the present invention; Figure 4 A schematic structural diagram of an embodiment of a rock mass fracture identification device provided by the present invention; Figure 5 A schematic structural diagram of an embodiment of an electronic device provided by the present invention. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in 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 those skilled in the art without creative work are within the scope of protection of the present invention.

[0020] In the description of the embodiments of the present invention, unless otherwise specified, "multiple" means two or more than two. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" may mean: A exists alone, A and B exist at the same time, and B exists alone.

[0021] The descriptions of "first" and "second" in the embodiments of the present invention are only used for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the technical features defined as "first" and "second" may explicitly or implicitly include at least one of the features.

[0022] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present invention. The appearance of the phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0023] The present invention provides a rock mass fracture identification method, device, electronic device and readable storage medium, which are respectively described below.

[0024] Figure 1 A schematic diagram of a flow chart of an embodiment of a rock mass fracture identification method provided by the present invention is shown in FIG. Figure 1 As shown, the rock mass fracture identification method includes: S101, inputting the acquired rock mass image to be tested into a well-trained rock mass fracture recognition network, performing feature encoding on the rock mass image to be tested level by level to obtain a number of multi-level encoding features, performing dual-branch attention enhancement on the multi-level encoding features to obtain corresponding dual-branch enhanced features, performing feature decoding on the multi-level encoding features level by level and fusing the dual-branch enhanced features of the corresponding levels to obtain decoded output features; S102, predicting and outputting the decoded output features to obtain rock mass fracture identification results.

[0025] Compared with the prior art, the rock mass crack identification method provided by the present invention first inputs the acquired rock mass image to be tested into a well-trained rock mass crack identification network, performs feature encoding on the rock mass image to be tested step by step to obtain a number of multi-level encoded features, performs dual-branch attention enhancement on the multi-level encoded features to obtain corresponding dual-branch enhanced features, performs feature decoding on the multi-level encoded features step by step and fuses the dual-branch enhanced features of the corresponding levels to obtain decoded output features; then, the decoded output features are predicted and output to obtain the rock mass crack identification results. The present invention combines the advantages of spatial channel attention and self-attention through dual-branch attention enhancement, and can globally integrate image context information while enhancing the modeling capabilities and spatial position perception capabilities of different features, thereby effectively improving the accuracy of the rock mass crack identification method.

[0026] In some embodiments of the present invention, the rock mass image to be tested is subjected to step-by-step feature coding to obtain a plurality of multi-level coding features, including: Perform feature convolution, ReLU activation function and maximum pooling on the rock mass image to be tested step by step to obtain multi-level encoding features; Among them, the output features of each level of maximum pooling are used as the input features of the next level of feature convolution.

[0027] Specifically, the rock mass fracture recognition network is constructed using the U-Net network as a whole. In order to enable the model to extract richer information features, the embodiment uses the VGG16 network to replace the backbone network in the U-Net network. The constructed rock mass fracture recognition network includes a feature encoding module, a feature decoding module, and a prediction module, and introduces a dual-branch attention enhancement module to enhance the feature representation capability.

[0028] In the embodiment, the feature encoding module is a stack of convolutional layers and maximum pooling layers, and five multi-level encoding features are obtained by step-by-step feature extraction. The feature extraction of each layer consists of feature convolution, ReLU activation function and maximum pooling. The feature convolution of each level takes the rock mass image to be tested of the initial input network or the output feature after maximum pooling of the previous level as input. The sizes of the five multi-level encoding features are 512×512×64, 256×256×128, 128×128×256, 64×64×512 and 32×32×1024, respectively.

[0029] In some embodiments of the present invention, Figure 2 Schematic diagram of the process of dual-branch attention enhancement according to an embodiment of the present invention. Figure 2 The multi-level encoding features shown are subjected to dual-branch attention enhancement to obtain corresponding dual-branch enhanced features, including: S201, performing feature convolution and batch normalization on the multi-level coding features to obtain features to be enhanced; S202, performing spatial channel attention enhancement on the feature to be enhanced to obtain a first attention feature; S203, performing self-attention enhancement on the feature to be enhanced to obtain a second attention feature; S204, fusing the first attention feature and the second attention feature to obtain a fused attention feature; S205, perform feature convolution and batch normalization on the fused attention features, and then concatenate and merge the multi-level encoding features to obtain dual-branch enhanced features.

[0030] Specifically, in order to effectively enhance image features and capture image context information on a global scale, the embodiment performs feature enhancement on the obtained multi-level coding features through dual-branch attention enhancement. During the dual-branch attention enhancement process, the embodiment first performs feature convolution and batch normalization on the multi-level coding features, and then performs attention extraction on the spatial channel attention branch and the self-attention branch, respectively. The spatial channel attention branch can enhance the model's modeling ability and spatial position perception ability for different features, and the self-attention branch helps the model capture global context information and mine the correlation between crack features in different regions, so that the model can more accurately identify the boundaries and details of the cracks. Finally, the attention features obtained from the two branches are multiplied and fused, and then passed through the feature convolution layer and batch normalization again to obtain the dual-branch enhanced features.

[0031] In some embodiments of the present invention, Figure 3 Schematic diagram of the process of spatial channel attention enhancement according to an embodiment of the present invention. Figure 3 As shown, the spatial channel attention enhancement is performed on the feature to be enhanced to obtain the first attention feature, including: S301, performing channel dimension attention extraction on the features to be enhanced and merging the features to be enhanced to obtain preliminary enhanced features; S302, performing spatial dimension attention extraction on the preliminary enhanced features and merging the preliminary enhanced features to obtain a first attention feature.

[0032] Specifically, in spatial channel attention enhancement, for the input tensor, the embodiment will infer the attention map along the two independent dimensions of channel and space in sequence, and multiply and merge the inferred attention map each time with the input feature map. In this way, the embodiment learns the weights between channels through channel attention to enhance the model's modeling ability for different types of features, and learns the weights of different spatial positions through spatial attention to enhance the model's spatial perception of features, so that the model can effectively enhance various types of crack features with huge differences in shape and size, thereby improving the accuracy of crack recognition.

[0033] In some embodiments of the present invention, performing self-attention enhancement on the feature to be enhanced to obtain a second attention feature includes: The features to be enhanced are sequentially down-sampled, self-attention extracted and up-sampled to obtain the second attention features.

[0034] Specifically, in self-attention enhancement, for the input tensor, the embodiment first downsamples it, then uses the self-attention mechanism to perform self-attention extraction in a global scope, and finally performs an upsampling operation to make the obtained features match the output size obtained by the spatial channel attention branch, so as to facilitate subsequent feature fusion operations. In the process, the embodiment will judge the number of input channels to determine the number of downsampling and upsampling to improve the robustness of the algorithm model.

[0035] In some embodiments of the present invention, multi-level coding features are decoded level by level and the corresponding level double-branch enhanced features are fused to obtain decoded output features, including: The last level of multi-level coding features is upsampled level by level, and the corresponding level double-branch enhanced features and feature convolution are fused to obtain the decoded output features.

[0036] Specifically, in the decoding stage of the rock fracture recognition network, the embodiment first upsamples the multi-level coding features obtained at the last level, and then merges them with the double-branch enhanced features after double-branch attention enhancement at the corresponding level of the encoder, and then performs feature convolution decoding and inputs them into the decoding layer of the next level. Finally, the embodiment transmits the decoded output features obtained from the last level decoding layer to the prediction module to perform binary classification of background and fracture for each feature point to obtain the rock fracture recognition result.

[0037] In some embodiments of the present invention, the trained complete rock mass fracture identification network is obtained by training an initial rock mass fracture identification network, and the training of the initial rock mass fracture identification network includes: The acquired rock fracture training data is input into the initial rock fracture identification network, and the parameters of the initial rock fracture identification network are optimized by gradient descent method based on the adaptive matrix estimation optimizer. The well-trained rock fracture identification network is obtained through the iterative optimization process.

[0038] Specifically, in the model training, in the embodiment, 80% of the acquired rock fracture training data is used as a training set, 10% as a test set, and 10% as a validation set. The size of the input image in the training data is 512×512. The Adam (Adaptive Moment Estimation) optimizer is used to train the initial rock fracture identification network through the gradient descent algorithm to obtain a fully trained rock fracture identification network.

[0039] In order to verify the effectiveness of the present invention, the embodiment substitutes the test set into the rock crack recognition network model of the present invention and compares it with the traditional FCN (Fully Convolutional Network), PSANet (Point-wise Spatial Attention Network), PSPNet (Pyramid Scene Parsing Network), CCNet (Criss-Cross Attention Network), DeeplabV3 network, DeeplabV3+ network and UperNet (Unified Perceptual Parsing Network), and uses the accuracy (Accuracy), intersection over union (IoU), mean intersection over union (MIoU), mean pixel accuracy and other indicators to evaluate the model. Among them, the specific formula of accuracy is:

[0040] The specific formula of intersection and union ratio is:

[0041] The specific formula of average intersection-union ratio is:

[0042] The specific formula for average pixel accuracy is:

[0043] in, Represents the number of actual positive examples and predicted positive examples; Indicates the number of actual negative examples that are predicted to be negative examples; Indicates the number of negative examples that are actually predicted to be positive examples; Indicates the number of positive examples that are actually predicted to be negative examples. is the total number of categories.

[0044] Finally, after calculation, the rock fracture identification network model of the present invention has an mPA of 89.94%, an accuracy of 99.45%, an IoU of 67.66, 99.44, and an MIoU of 83.55%. However, traditional models such as the CCNet model have an mPA of 87.29%, an accuracy of 97.96%, an IoU of 53.2, 99.07, and an MIoU of 76.14%. The UperNet model has an mPA of 87.21%, an accuracy of 98.63%, an IoU of 64.73, 99.40, and an MIoU of 82.06%. From the results, it can be seen that the basic model has been significantly improved and has a higher segmentation accuracy.

[0045] In summary, the rock mass fracture identification method provided by the present invention first inputs the acquired rock mass image to be tested into a well-trained rock mass fracture identification network, performs feature encoding on the rock mass image to be tested step by step to obtain a number of multi-level coded features, performs dual-branch attention enhancement on the multi-level coded features to obtain corresponding dual-branch enhanced features, performs feature decoding on the multi-level coded features step by step and fuses the dual-branch enhanced features of the corresponding levels to obtain decoded output features; then the decoded output features are predicted and output to obtain the rock mass fracture identification results. The present invention combines the advantages of spatial channel attention and self-attention through dual-branch attention enhancement, and can globally integrate image context information while enhancing the modeling capabilities and spatial position perception capabilities of different features, thereby effectively improving the accuracy of the rock mass fracture identification method.

[0046] In order to better implement the rock mass fracture identification method in the embodiment of the present invention, based on the rock mass fracture identification method, correspondingly, Figure 4 As shown, the present invention also provides a rock mass crack identification device, and the rock mass crack identification device 400 includes: The feature extraction unit 401 is used to input the acquired rock mass image to be tested into a well-trained rock mass fracture recognition network, perform feature encoding on the rock mass image to be tested level by level to obtain a plurality of multi-level encoded features, perform dual-branch attention enhancement on the multi-level encoded features to obtain corresponding dual-branch enhanced features, perform feature decoding on the multi-level encoded features level by level and fuse the dual-branch enhanced features of the corresponding levels to obtain decoded output features; The prediction output unit 402 is used to predict and output the decoded output features to obtain rock fracture identification results.

[0047] The rock fracture identification device 400 provided in the above embodiment can implement the technical solution described in the above rock fracture identification method embodiment. The specific implementation principles of the above modules or units can refer to the corresponding contents in the above rock fracture identification method embodiment, which will not be repeated here.

[0048] like Figure 5 As shown, the present invention also provides an electronic device 500. The electronic device 500 includes a processor 501, a memory 502 and a display 503. Figure 5 Only some components of the electronic device 500 are shown, but it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0049] In some embodiments, the processor 501 may be a central processing unit (CPU), a microprocessor or other data processing chip, used to run program codes or process data stored in the memory 502, such as the rock fracture identification method of the present invention.

[0050] In some embodiments, the processor 501 may be a single server or a server group. The server group may be centralized or distributed. In some embodiments, the processor 501 may be local or remote. In some embodiments, the processor 501 may be implemented in a cloud platform. In one embodiment, the cloud platform may include a private cloud, a public cloud, a hybrid cloud, a community cloud, a distributed cloud, an internal cloud, a multi-cloud, etc., or any combination thereof.

[0051] In some embodiments, the memory 502 may be an internal storage unit of the electronic device 500, such as a hard disk or memory of the electronic device 500. In other embodiments, the memory 502 may also be an external storage device of the electronic device 500, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the electronic device 500.

[0052] Furthermore, the memory 502 may include both an internal storage unit of the electronic device 500 and an external storage device. The memory 502 is used to store application software installed in the electronic device 500 and various data.

[0053] In some embodiments, the display 503 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode) touch device, etc. The display 503 is used to display information of the electronic device 500 and to display a visual user interface. The components 501-503 of the electronic device 500 communicate with each other through a system bus.

[0054] In one embodiment, when the processor 501 executes the rock fracture identification program in the memory 502, the following steps may be implemented: The acquired rock mass image to be tested is input into a well-trained rock mass crack recognition network, and the rock mass image to be tested is feature encoded level by level to obtain a number of multi-level encoding features, and the multi-level encoding features are subjected to dual-branch attention enhancement to obtain corresponding dual-branch enhanced features, and the multi-level encoding features are feature decoded level by level and the dual-branch enhanced features of the corresponding levels are fused to obtain the decoded output features; The decoded output features are predicted and output to obtain the rock fracture identification results.

[0055] It should be understood that: when the processor 501 executes the rock fracture identification program in the memory 502, in addition to the above functions, other functions can also be implemented. For details, please refer to the description of the corresponding method embodiment above.

[0056] Accordingly, an embodiment of the present application also provides a computer-readable storage medium, which is used to store computer-readable programs or instructions. When the program or instructions are executed by a processor, the steps or functions in the rock fracture identification method provided in the above-mentioned method embodiments can be implemented.

[0057] Those skilled in the art will appreciate that all or part of the processes of the above-mentioned embodiments can be implemented by instructing related hardware (such as a processor, a controller, etc.) through a computer program, and the computer program can be stored in a computer-readable storage medium, wherein the computer-readable storage medium is a disk, an optical disk, a read-only storage memory, or a random access memory, etc.

[0058] The rock fracture identification method, device, electronic device and storage medium provided by the present invention are introduced in detail above. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for technical personnel in this field, according to the idea of ​​the present invention, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A rock mass fracture identification method, characterized in that: include: Input the acquired rock mass image to be tested into a well-trained rock mass fracture recognition network, perform feature encoding on the rock mass image to be tested level by level to obtain a plurality of multi-level encoding features, perform dual-branch attention enhancement on the multi-level encoding features to obtain corresponding dual-branch enhanced features, perform feature decoding on the multi-level encoding features level by level and fuse the dual-branch enhanced features of the corresponding levels to obtain decoded output features; The decoded output features are predicted and output to obtain rock mass fracture identification results.

2. The rock mass fracture identification method according to claim 1, characterized in that: The step-by-step feature coding of the rock mass image to be tested to obtain a plurality of multi-level coding features includes: Performing feature convolution, ReLU activation function and maximum pooling on the rock mass image to be tested step by step to obtain multi-level coding features; Among them, the output features of each level of maximum pooling are used as the input features of the next level of feature convolution.

3. The rock mass fracture identification method according to claim 1, characterized in that: The performing dual-branch attention enhancement on the multi-level coding features to obtain corresponding dual-branch enhanced features includes: Performing feature convolution and batch normalization on the multi-level encoding features to obtain features to be enhanced; Performing spatial channel attention enhancement on the feature to be enhanced to obtain a first attention feature; Performing self-attention enhancement on the feature to be enhanced to obtain a second attention feature; Fusing the first attention feature and the second attention feature to obtain a fused attention feature; The fused attention features are subjected to feature convolution and batch normalization, and then spliced ​​and combined with the multi-level encoding features to obtain dual-branch enhanced features.

4. The rock mass fracture identification method according to claim 3, characterized in that: The step of performing spatial channel attention enhancement on the feature to be enhanced to obtain a first attention feature includes: After performing channel dimension attention extraction on the features to be enhanced, the features to be enhanced are merged to obtain preliminary enhanced features; After performing spatial dimension attention extraction on the preliminary enhanced features, the preliminary enhanced features are merged to obtain a first attention feature.

5. The rock mass fracture identification method according to claim 3, characterized in that: The step of performing self-attention enhancement on the feature to be enhanced to obtain a second attention feature includes: The features to be enhanced are sequentially down-sampled, self-attention extracted and up-sampled to obtain a second attention feature.

6. The rock mass fracture identification method according to claim 1, characterized in that: The step of decoding the multi-level coding features level by level and fusing the dual-branch enhanced features of the corresponding levels to obtain decoded output features includes: The multi-level coding features of the last level are up-sampled level by level, and the double-branch enhanced features and feature convolution of the corresponding level are fused to obtain the decoded output features.

7. The rock mass fracture identification method according to claim 1, characterized in that: The fully trained rock mass fracture identification network is obtained by training an initial rock mass fracture identification network, wherein the trained initial rock mass fracture identification network comprises: The acquired rock mass fracture training data is input into the initial rock mass fracture identification network, and the parameters of the initial rock mass fracture identification network are optimized by gradient descent method based on the adaptive matrix estimation optimizer, and a fully trained rock mass fracture identification network is obtained through an iterative optimization process.

8. A rock mass fracture identification device, characterized in that: include: A feature extraction unit is used to input the acquired rock mass image to be tested into a well-trained rock mass fracture recognition network, perform feature encoding on the rock mass image to be tested level by level to obtain a plurality of multi-level encoding features, perform dual-branch attention enhancement on the multi-level encoding features to obtain corresponding dual-branch enhanced features, perform feature decoding on the multi-level encoding features level by level and fuse the dual-branch enhanced features of the corresponding level to obtain decoded output features; The prediction output unit is used to predict and output the decoded output features to obtain rock fracture identification results.

9. An electronic device comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the rock fracture identification method according to any one of claims 1 to 7 is 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 rock fracture identification method according to any one of claims 1 to 7 is implemented.

Images