TBM tunnel (hole) rock strength sensing method and system based on IPAN-efficient net network

Abstract

This application discloses a method and system for sensing rock strength in TBM tunnels (tunnels) based on the IPAN-EfficientNet network. The method includes: acquiring images of rock debris from the TBM tunnel (tunnel) and corresponding rebound strength data of the rock in the tunnel wall; establishing a rock debris image-rock strength level dataset; obtaining top-ranked indicators; constructing an IPAN-EfficientNet network; training the IPAN-EfficientNet network using the rock debris image-rock strength level dataset to obtain a rock strength level sensing model; and using the rock strength level sensing model to identify rock debris images, thereby achieving rock strength level sensing in the TBM tunnel (tunnel). The method described in this application achieves accurate sensing of rock strength at the TBM tunnel (tunnel) face, providing more scientific guidance for disaster early warning and tunneling support optimization during TBM tunnel (tunnel) construction, and ensuring TBM safety.

Description

Technical Field

[0001] This invention relates to the field of hard rock TBM tunnel (cavity) surrounding rock strength perception and computer vision technology, and in particular to a method and system for identifying the rock strength level of TBM tunnel (cavity) based on rock debris images. Background Technology

[0002] In the construction of deep and long tunnels, hard rock TBMs offer advantages over drill-and-blast methods in terms of safety, efficiency, and environmental friendliness. However, the tunneling performance of TBMs is extremely sensitive to geological changes, requiring timely optimization of tunneling and support parameters based on the encountered geological conditions to ensure safe and efficient TBM operation. TBM operators cannot directly observe the tunnel face, thus hindering accurate acquisition of geological information. Therefore, indirect methods are necessary to obtain this information. Rock debris is a direct byproduct of TBM rock breaking, and its morphology and particle size distribution contain rich information about the surrounding rock. In actual construction, operators typically rely on rock debris images to roughly assess the geological conditions of the tunnel face and make tunneling and support decisions, which is insufficient to guarantee safe and efficient TBM tunneling.

[0003] Using rock debris images to perceive the geology of tunnel faces is a hot research topic in the field. However, current research mainly focuses on identifying the grade of rough surrounding rock (with different combinations of rock strength and rock mass integrity within the same grade) or custom categories (such as the presence of large rock blocks), and rarely accurately obtains the two basic geological parameters of rock strength and rock mass integrity coefficient, thus failing to provide precise and scientific guidance for on-site construction.

[0004] Therefore, there is an urgent need for a method that can accurately sense the rock strength at the working face of a TBM tunnel. Summary of the Invention

[0005] To address the current limitations of rock strength assessment in TBM tunnel face recognition research, this application provides a method and system for sensing rock strength in TBM tunnels based on the IPAN-EfficientNet network.

[0006] Firstly, this application provides a method for sensing rock strength in TBM tunnels (cavities) based on the IPAN-EfficientNet network, including: Collect images of rock debris from the TBM tunnel (cavity) and data on the rebound strength of the rock in the tunnel wall; Rock strength is divided into multiple levels according to rebound strength value. Rock slag images are preprocessed to establish a rock slag image-rock strength level dataset. Based on the rock debris outline in the rock debris image, the correlation between rock debris morphological parameters and rock strength is analyzed, and the top-ranked indicators are obtained. Construct an IPAN-EfficientNet network by fusing the improved path aggregation network Improved PANet and EfficientNet; The IPAN-EfficientNet network was trained using a rock slag image-rock strength grade dataset to obtain a rock strength grade perception model. Rock strength level perception model is used to identify rock debris images in order to achieve rock strength level perception in TBM tunnels (caves).

[0007] In one possible implementation, the rock debris images and their corresponding back strength data are matched one-to-one with the recorded station number and time information.

[0008] In one possible implementation, the preprocessing includes size scaling, grayscale conversion, contrast enhancement, image denoising, and image enhancement.

[0009] In one possible implementation, the top-ranking metrics include: average radius ratio. 30th percentile roundness 60th percentile roundness Average roundness and maximum roundness .

[0010] In one possible implementation, the Improved PANet includes a feature pyramid network, a bottom-up path enhancement network, a compression and activation attention layer, and a classifier connected in sequence. The feature pyramid network comprises multiple input layers, a top-down upsampling network, and an intermediate fusion network. The multiple input layers, from bottom to top, include four feature maps F1, F2, F3, and F4, derived from EfficientNet. The top-down upsampling network includes four feature maps P4, P3, P2, and P1, and an upsampling layer positioned between the feature maps. This upsampling layer is composed of depthwise separable deconvolutional layers. The intermediate fusion network fuses the corresponding feature maps from the input layers and the top-down upsampling network by concatenation. Feature maps F1 and F4 are connected to P4 via jump chains, and feature map F1 is connected to F4 via a jump chain. The bottom-up path enhancement network includes four feature maps N1, N2, N3, and N4 in sequence; N1 is P1, N2 is obtained by downsampling N1 and concatenating it with P2, N3 is obtained by downsampling N2 and concatenating it with P3, and N4 is obtained by downsampling N3 and concatenating it with P4; N4 is short-circuited and downsampled with F1 and P1, and then aggregated through a depthwise separable deconvolution layer; The compressed and excited attention layer takes feature aggregation input as input and extracts attention between channels to obtain weighted fused multi-scale features; The classifier includes a global average pooling layer and a fully connected layer, used to output the classification results of rock strength levels.

[0011] Furthermore, the EfficientNet sequentially includes an input layer, a Gaussian noise layer, a stem layer, a first Fused MBConv module, a second Fused MBConv module, a first MBConv module, a second MBConv module, a third MBConv module, a fourth MBConv module, and a head layer; the head layer includes a convolutional layer, a BN layer, and a SiLU activation function. The second Fused MBConv module outputs F1', the first MBConv module outputs F2', the third MBConv module outputs F3', and the head layer outputs F4'; F1', F2', F3', and F4' are obtained by passing through convolutional layers to obtain F1, F2, F3, and F4, respectively.

[0012] Secondly, this application provides a TBM tunnel (cavity) rock strength sensing system based on the IPAN-EfficientNet network, including: The acquisition module is used to acquire images of rock debris from the TBM tunnel (cavity) and the corresponding rebound strength data of the rock in the tunnel wall; The dataset construction module is used to classify rock strength into multiple levels according to the rebound strength value, preprocess rock slag images, and establish a rock slag image-rock strength level dataset. The indicator filtering module is used to analyze the correlation between rock morphological parameters and rock strength based on the rock slag outline in the rock slag image, and to obtain the top-ranked indicators. The building block is used to construct the IPAN-EfficientNet network, which is formed by fusing the improved path aggregation network Improved PANet and EfficientNet. The training module is used to train the IPAN-EfficientNet network using a rock slag image-rock strength level dataset to obtain a rock strength level perception model. The output module is used to identify rock debris images using a rock strength level perception model in order to achieve rock strength level perception in TBM tunnels (caves).

[0013] Thirdly, this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the TBM tunnel (cavity) rock strength sensing method based on the IPAN-EfficientNet network as described in the first aspect.

[0014] Fourthly, a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the TBM tunnel (cavity) rock strength sensing method based on the IPAN-EfficientNet network as described in the first aspect.

[0015] Fifthly, a computer program product includes a computer program that, when executed by a processor, implements the TBM tunnel (cavity) rock strength sensing method based on the IPAN-EfficientNet network as described in the first aspect.

[0016] Compared with the prior art, the beneficial effects of this application are as follows: The method provided in this application, through the design of the IPAN-EfficientNet network, can accurately identify the morphological characteristics of rock debris of different sizes, and the perception model has excellent noise resistance. The accuracy of the provided perception model is superior to that of commonly used existing CNN models, including the EfficientNet model, and can provide detailed and scientific guidance for TBM construction.

[0017] The method provided in this application has been tested and shown to be feasible and has great application potential in perceiving the surrounding rock strength of the working face of a TBM tunnel based on rock debris images. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0019] Figure 1 A flowchart of a TBM tunnel (cavity) rock strength sensing method based on IPAN-EfficientNet network provided for embodiments of this application; Figure 2 This is a schematic diagram of the IPAN-EfficientNet network structure according to an embodiment of this application; Figure 3 Rock slag images provided for embodiments of this application; Figure 4The confusion matrix output on the test set by the IPAN-EfficientNet network model provided in the embodiments of this application; Figure 5 This is a schematic diagram comparing the performance of the IPAN-EfficientNet network model of this application with other classic CNN models in a noisy environment. Figure 6 A structural diagram of a TBM tunnel (cavity) rock strength sensing system based on the IPAN-EfficientNet network provided in an embodiment of this application; Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0020] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0022] In the description of this application, it should also be noted that, unless otherwise expressly specified and limited, the terms "set up," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this technology based on the specific circumstances.

[0023] In the description of this application, spatial relation terms such as "below," "under," "below," "below," "above," "over," etc., are used herein to describe the relationship between one element or feature shown in the figures and other elements or features. It should be understood that, in addition to the orientation shown in the figures, spatial relation terms also include different orientations of the device in use and operation. For example, if the device in the figures is flipped, an element or feature described as "below" or "under" or "below" of other elements or features will be oriented "above" other elements or features. Therefore, the exemplary terms "below" and "under" can include both upper and lower orientations. Furthermore, the device may also include other orientations (e.g., rotated 90 degrees or other orientations), and the spatial descriptive terms used herein are interpreted accordingly.

[0024] In the description of this application, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use this application. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that this application can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid unnecessarily obscuring the description of this application. Therefore, this application is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0025] refer to Figure 1 A method for sensing rock strength in TBM tunnels (cavities) based on the IPAN-EfficientNet network includes the following steps: S101. Collect images of rock debris from the TBM tunnel (cavity) and data on the rebound strength of the rock in the tunnel wall.

[0026] In one possible implementation, in S101, the rock debris image and its corresponding back strength data are matched one-to-one with the recorded station number and time information.

[0027] In one possible implementation, in S101, rock slag images are acquired through a rock slag image acquisition system, which consists of an industrial camera, a data cable, and a data acquisition device (or computer). The industrial camera is located directly above the main unit belt or the rear supporting belt, and the data acquisition device controls the camera to acquire rock slag images. If the ambient light is insufficient, shadowless supplementary lighting should also be deployed near the camera.

[0028] In one possible implementation method, in S101, rock rebound strength data is collected through a rebound test. The cave wall rebound test is generally conducted manually using a rebound hammer on the cave wall (the impact rod of the rebound hammer should be perpendicular to the cave wall). A relatively intact cave wall should be selected for the test. The steps to obtain the rock rebound strength of the cave wall are as follows: (1) For a certain tunnel (cavity) section, if the tunnel wall rock mass is relatively intact, multiple measuring points are selected to carry out rebound tests. Before carrying out rebound tests, the test area of ​​the tunnel wall is ground flat by a grinding machine.

[0029] (2) The rebound test was carried out according to the method recommended by the International Society for Rock Mechanics. The test range for each measuring point was 20×20 cm. 2 The test area was divided into 20 test blocks, and each block underwent one rebound test. The distance between adjacent test points was ensured to be greater than the diameter of the rebound hammer's test rod during the rebound test. The formula for calculating the rebound strength at a given test point is as follows: (1) in, The i-th rebound strength value of a certain measuring point in descending order; This is a correction value for the rebound strength in the non-horizontal direction; This represents the average rebound strength at a certain measuring point.

[0030] (3) The final rebound strength value of a section is obtained by averaging the rebound strength of all measuring points on a certain section of the tunnel. The calculation formula is as follows; (2) in, Let j be the rebound strength value at the j-th measuring point; The number of measuring points on a certain cross section; This represents the rebound strength value of a certain cross-section.

[0031] S102. Based on the rebound strength value, the rock strength is divided into multiple levels. The rock slag image is preprocessed to establish a rock slag image-rock strength level dataset.

[0032] In one possible implementation, in S102, the rock strength is divided into three levels according to the rebound value based on the range of field-collected data, indicating high, medium, and low rock strength.

[0033] In one possible implementation, in S102, the preprocessing includes size scaling, grayscale conversion, contrast enhancement, image denoising, and image enhancement, wherein image enhancement includes flipping, rotating, translating, and scaling operations.

[0034] In one possible implementation, in S102, the rock slag image-rock strength grade dataset is divided into a training set, a validation set, and a test set; wherein, the training set is used to train the model, the validation set is used to adjust the model parameters, and the test set is used to verify the effectiveness of the model.

[0035] S103. Based on the rock debris outline in the rock debris image, analyze the correlation between rock debris morphological parameters and rock strength, and obtain the top-ranked indicators.

[0036] In one possible implementation, in S103, the morphological parameters of the rock debris include the ratio of the major axis to the minor axis radius and the roundness, the calculation of which is given by formulas (3) and (4); the relevant indicators include the 10th percentile radius ratio. 30th percentile radius ratio 60th percentile radius ratio Average radius ratio Maximum radius ratio 10th percentile roundness 30th percentile roundness 60th percentile roundness Average roundness and maximum roundness .

[0037] (3) (4) in: The radius of the major axis of the rock debris; The radius of the minor axis of the rock debris; The ratio of the major and minor axes radii; The perimeter of the rock debris outline; Area of ​​rock debris; This refers to the roundness of the rock debris.

[0038] It should be noted that if a single image of rock debris contains rock debris with different equivalent roundness or radius ratios, these rock debris pieces are sorted from smallest to largest according to their equivalent roundness or radius ratio. The roundness or radius ratios corresponding to the top 10% are the 10th quantile equivalent roundness or radius ratios. Rd 10 or Rr 10 Other parameters are similar.

[0039] average radius ratio This refers to the average radius ratio of rock cutting particles. Average roundness refers to the average roundness of rock cutting particles.

[0040] In one possible implementation, in S103, the correlation coefficients between each rock morphology index and rock rebound strength are determined using the Pearson linear correlation coefficient (PLCC) and the maximum information coefficient (MIC), and the five most correlated rock morphology indices are selected as follows: , , , and The index most relevant to rock strength includes the mean radius ratio. and roundness This indicates that rock strength is closely related to the morphology of rock fragments of all sizes. Therefore, the IPAN-EfficientNet network model constructed in subsequent steps should be able to accurately identify the morphological characteristics of rock fragments of different sizes, i.e., the structure and parameters of the modules should be adjusted accordingly.

[0041] S104. Construct the IPAN-EfficientNet network, which is formed by fusing the improved path aggregation network Improved PANet and EfficientNet.

[0042] In one possible implementation, see S104, see Figure 2 The improved PANet includes a feature pyramid network, a bottom-up path enhancement network, a compression and activation attention layer, and a classifier connected in sequence. The feature pyramid network comprises multiple input layers, a top-down upsampling network (P4→P1), and an intermediate fusion network. The multiple input layers, from bottom to top, include four feature maps F1, F2, F3, and F4, derived from EfficientNet. The top-down upsampling network includes four feature maps P4, P3, P2, and P1, and upsampling layers positioned between the feature maps. Each upsampling layer is composed of depthwise separable deconvolutional layers (which consist of depthwise separable convolutional layers and deconvolutional layers). The intermediate fusion network merges the corresponding feature maps from the input layers and the top-down upsampling network by concatenation. Feature maps F1 and F4 are connected to P4 via jump chains, and feature map F1 is connected to F4 via a jump chain. The bottom-up path enhancement network (N1→N4) includes four feature maps N1, N2, N3, and N4 in sequence; N1 is P1, N2 is obtained by downsampling N1 and concatenating it with P2, N3 is obtained by downsampling N2 and concatenating it with P3, and N4 is obtained by downsampling N3 and concatenating it with P4; N4 is short-circuited to F1 and P1, and features are aggregated through a depthwise separable convolutional layer (the depthwise separable deconvolutional layer is composed of depthwise separable convolutional layers and deconvolutional layers); The compressed and excited attention layer takes feature aggregation input as input and extracts attention between channels to obtain weighted fused multi-scale features; The classifier includes a global average pooling layer and a fully connected layer, used to output the classification results of rock strength levels.

[0043] It should be noted that the top-down and bottom-up descriptions are for consistency with the original PANet structure description. Specifically, Figure 2 In the middle, the left side is the bottom, and the right side is the top.

[0044] Furthermore, in the feature pyramid network, the activation function of the depth-separable deconvolution layer of the top-down upsampling network is Swish.

[0045] Furthermore, in the bottom-up path enhancement network, the activation function of the separable deconvolution layer is Swish.

[0046] Furthermore, the activation function for all convolutional layers in the improved PANet is Swish.

[0047] Furthermore, the EfficientNet sequentially includes an input layer, a Gaussian noise layer, a stem layer, a first Fused MBConv module, a second Fused MBConv module, a first MBConv module, a second MBConv module, a third MBConv module, a fourth MBConv module, and a head layer; the head layer includes a convolutional layer, a BN layer, and a SiLU activation function. The second Fused MBConv module outputs F1', the first MBConv module outputs F2', the third MBConv module outputs F3', and the head layer outputs F4'; F1', F2', F3', and F4' are obtained by passing through convolutional layers to obtain F1, F2, F3, and F4, respectively.

[0048] Specifically, the EfficientNet is EfficientNetV2B3.

[0049] The inference process of the IPAN-EfficientNet network is described below.

[0050] See Figure 2The rock debris image is input into EfficientNet, processed by a Gaussian noise layer, and then sequentially passed through a stem layer, a first Fused MBConv module, a second Fused MBConv module, a first MBConv module, a second MBConv module, a third MBConv module, a fourth MBConv module, and a head layer, outputting relevant feature maps F1', F2', F3', and F4'. The Gaussian noise layer is used to enhance the model's noise resistance in environments with strong dust and water mist (noise) and improve the signal-to-noise ratio.

[0051] Feature maps F1, F2, F3, and F4 are input to Improved PANet. Here, F1 = Conv(F1'), F2 = Conv(F2'), F3 = Conv(F3'), and F4 is obtained by downsampling and concatenating Conv(F4') with F1. F4 adds local details by concatenating with F1. Conv(·) represents a 1×1 convolution.

[0052] P4 is obtained by concatenating Conv(F4) and F1 downsampled data. This operation adds local details by concatenating F1, which already possesses high global semantics. P3 is obtained by concatenating Conv(F3) and P4 upsampled data. P2 is obtained by concatenating Conv(F2) and P3 upsampled data. P1 is obtained by concatenating Conv(F1) and P2 upsampled data.

[0053] N1 is P1. N2 is obtained by downsampling N1 and concatenating it with Conv(P2). N3 is obtained by downsampling N2 and concatenating it with Conv(P3). N4 is obtained by downsampling N3 and concatenating it with Conv(P4). N4 is short-circuited with F1 and P1 (add) to ensure more global semantics and increase local details. Feature aggregation is obtained through a depthwise separable deconvolution layer. The depthwise separable deconvolution layer is used to extract more refined shape features.

[0054] The aggregated features are fed into a compression and activation attention layer (SE module) to extract inter-channel attention, resulting in weighted fused multi-scale features.

[0055] The weighted fusion of multi-scale features is passed through a classifier (Dropout layer + FC layer + softmax function) to output the classification result of rock strength level.

[0056] In the Improved PANet described in this application embodiment, the upsampling layers in the top-down upsampling network (P4→P1) all employ depthwise separable deconvolutional layers, with the Swish activation function used to extract finer shape features. The depthwise separable deconvolutional layer consists of depthwise separable convolutional layers and deconvolutional layers. The use of the Swish activation function instead of ReLU is to promote gradient (information) propagation, thereby improving model performance. In the Improved PANet described in this application embodiment, all convolutional layers use the Swish activation function instead of ReLU, again to promote gradient (information) propagation and thus improve model performance.

[0057] S105. The IPAN-EfficientNet network is trained using the rock slag image-rock strength level dataset to obtain a rock strength level perception model.

[0058] In one possible implementation, S105 includes: S105a. Freeze the first 30% of the loaded EfficientNet network layers (parameters cannot be trained) and compile the IPAN-EfficientNet network.

[0059] S105b. Divide the dataset into a training set, a validation set, and a test set. The training set is used for model training, the validation set is used for model parameter tuning, and the test set is used to verify the effectiveness of the model.

[0060] S106. Use the rock strength level perception model to identify rock debris images in order to realize the perception of rock strength level in TBM tunnels (caves).

[0061] The following description, in conjunction with more detailed embodiments, provides further details.

[0062] A method for sensing rock strength in TBM tunnels (cavities) based on the IPAN-EfficientNet network includes the following steps: Step 1: Using a rock debris image acquisition system installed in a TBM tunnel in Southwest China, a rebound test was conducted on the tunnel wall to obtain rock debris images and rock strength data.

[0063] in: The rock slag image acquisition system consists of an industrial camera installed above the conveyor belt near the main control room, a high-speed data cable, and a computer deployed in the main control room. The rock slag image acquisition software controls the camera to acquire rock slag images at certain time intervals. In order to ensure that the captured rock slag images have sufficient brightness, four strip lights are also installed near the camera. Rebound tests were conducted on relatively intact rock sections within the tunnel. Following the operating procedures, a rebound hammer (produced by PROCEQ, Switzerland, with a range of 20-150 MPa and a rebound energy of 2.207 J) was used to obtain the rock rebound strength. The steps are as follows: (1) For a certain tunnel (cavity) section, if the tunnel wall rock mass is relatively intact, multiple measuring points are selected to carry out rebound tests. Before carrying out rebound tests, the test area of ​​the tunnel wall is ground flat by a grinding machine.

[0064] (2) The rebound test was carried out according to the method recommended by the International Society for Rock Mechanics. The test range for each measuring point was 20×20 cm. 2 The test area is divided into 20 test blocks. Each block is subjected to a rebound test. The distance between adjacent test points is greater than the diameter of the test rod of the rebound hammer during the rebound test. The rebound strength of a certain test point is calculated according to formula (1).

[0065] (3) The average value of the rebound strength of all measuring points on a certain section of the tunnel (hole) is used to obtain the final rebound strength value of the section, which is calculated according to formula (2).

[0066] Figure 3 An image of rock debris is shown.

[0067] Step 2: Divide the rock strength into multiple levels according to the rock rebound strength value and establish a rock slag image-rock strength level dataset; the rock slag images in the rock slag image-rock strength level dataset need to be preprocessed, and the dataset is divided into training set, validation set and test set. The training set is used to train the model, the validation set is used to adjust the model parameters, and the test set is used to verify the effectiveness of the model; the rock slag image preprocessing steps include size scaling, grayscale conversion, contrast enhancement, image noise reduction and image enhancement, where image enhancement includes flipping, rotation, translation and scaling operations.

[0068] Step 3: Based on the rock slag outline in the rock slag image, obtain relevant indices of rock slag morphology parameters, analyze the correlation between rock slag indices and rock strength, and provide guidance for the subsequent design of rock slag image recognition models.

[0069] The outline of the rock debris in the image was delineated using Image-Pro Plus 6.0 software. The software can automatically obtain the morphological parameters of the rock debris, including the ratio of the major axis to the minor axis radius and the roundness. The calculation of the radius ratio and roundness is shown in formulas (3) and (4). The relevant indicators include the 10th percentile radius ratio. 30th percentile radius ratio 60th percentile radius ratio Average radius ratio Maximum radius ratio 10th percentile roundness 30th percentile roundness 60th percentile roundness Average roundness and maximum roundness .

[0070] The correlation coefficients between various rock morphology indices and rock rebound strength were determined using the Pearson linear correlation coefficient (PLCC) and maximum information coefficient (MIC). The five most correlated rock morphology indices were selected as follows: , , , and The index most relevant to rock strength includes the mean radius ratio. and roundness This indicates that rock strength is closely related to the morphology of rock debris of all sizes, meaning that the rock debris image recognition network model designed subsequently should be able to accurately identify the morphological characteristics of rock debris of different sizes.

[0071] Step 4: Construct the IPAN-EfficientNet network, which is formed by fusing the improved path aggregation network Improved PANet and EfficientNet.

[0072] Step 5: Train the IPAN-EfficientNet network using the rock slag image-rock strength grade dataset to obtain a rock strength grade perception model.

[0073] Step 6: Use the rock strength level perception model to identify rock debris images in order to realize the rock strength level perception of TBM tunnels (caves).

[0074] The performance test results of the aforementioned method are presented below.

[0075] Figure 4 This is the confusion matrix output by the IPAN-EfficientNet network in this embodiment of the application on the test set. Labels 0, 1, and 2 represent three non-overlapping rock rebound intensity ranges, respectively. The numerical values ​​in the cells represent the sample size, and the percentages represent the proportions.

[0076] Figure 5This diagram illustrates the performance comparison between the IPAN-EfficientNet network model of this application and other classic CNN models in a noisy environment. It can be observed that when the noise level in the test set is Gaussian noise with a standard deviation of 0.04, the F-measure of the proposed IPAN-EfficientNet model still reaches 0.878, which is 7.6%, 66.0%, 66.9%, 69.1%, 69.1%, 46.2%, 41.1%, 58.9%, and 67.1% higher than the AlexNet, Xception, VGG16, ResNet101, MobileNetV2, DenseNet121, NASNetMobile, InceptionV3, and EfficientNetV2B3 models, respectively. Therefore, the TBM tunnel (cavity) rock strength level perception method and system based on rock debris images and the IPAN-EfficientNet network provided in this application is effective and superior. Corresponding to the aforementioned embodiment of a TBM tunnel (cavity) rock strength sensing method based on rock debris images and IPAN-EfficientNet network, this application also provides a TBM tunnel (cavity) rock strength sensing system based on rock debris images and IPAN-EfficientNet network.

[0077] Figure 6 This is a schematic diagram of the structure of a TBM tunnel (cavity) rock strength sensing system based on rock debris images and the IPAN-EfficientNet network, as provided in an embodiment of this application. Figure 6 As shown, it includes: a data acquisition module 61, a dataset construction module 62, an indicator selection module 63, a construction module 64, a training module 65, and an output module 66, wherein: The acquisition module 61 is used to acquire images of rock debris from the TBM tunnel (cavity) and data on the rebound strength of the rock in the tunnel wall. The dataset construction module 62 is used to classify rock strength into multiple levels according to the rebound strength value, preprocess rock slag images, and establish a rock slag image-rock strength level dataset. The index screening module 63 is used to analyze the correlation between rock morphological parameters and rock strength based on the rock slag outline of the rock slag image and to obtain the top-ranked indicators. Module 64 is used to build the IPAN-EfficientNet network, which is formed by fusing the improved path aggregation network Improved PANet and EfficientNet. Training module 65 is used to train the IPAN-EfficientNet network using the rock slag image-rock strength level dataset to obtain a rock strength level perception model. Output module 66 is used to identify rock debris images using a rock strength level perception model in order to realize the rock strength level perception of TBM tunnels (caves).

[0078] Figure 7 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 7 As shown, the electronic device may include a processor 710, a communications interface 720, a memory 730, and a communications bus 740. The processor 710, communications interface 720, and memory 730 communicate with each other via the communications bus 740. The processor 710 can call logical instructions from the memory 730 to execute a TBM tunnel (cavity) rock strength sensing method based on rock debris images and the IPAN-EfficientNet network.

[0079] Furthermore, the logical instructions in the aforementioned memory 730 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0080] On the other hand, this application also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute the TBM tunnel (cavity) rock strength sensing method based on rock debris images and IPAN-EfficientNet network provided by the above methods.

[0081] In another aspect, this application also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, is implemented to perform the TBM tunnel (cavity) rock strength sensing method based on rock debris images and IPAN-EfficientNet network provided by the above methods.

[0082] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0083] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0084] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for sensing rock strength in TBM tunnels (cavities) based on IPAN-EfficientNet network, characterized in that, include: Collect images of rock debris from the TBM tunnel (cavity) and data on the rebound strength of the rock in the tunnel wall; Rock strength is divided into multiple levels according to rebound strength value. Rock slag images are preprocessed to establish a rock slag image-rock strength level dataset. Based on the rock debris outline in the rock debris image, the correlation between rock debris morphological parameters and rock strength is analyzed, and the top-ranked indicators are obtained. Construct an IPAN-EfficientNet network by fusing the improved path aggregation network Improved PANet and EfficientNet; The IPAN-EfficientNet network was trained using a rock slag image-rock strength grade dataset to obtain a rock strength grade perception model. Rock strength level perception model is used to identify rock debris images in order to achieve rock strength level perception in TBM tunnels (caves).

2. The method according to claim 1, characterized in that, The rock debris images and their corresponding recovery strength data are matched one-to-one with the recorded station number and time information.

3. The method according to claim 1, characterized in that, The preprocessing includes size scaling, grayscale conversion, contrast enhancement, image noise reduction, and image enhancement.

4. The method according to claim 1, characterized in that, The top-ranking indicators include: average radius ratio 30th percentile roundness 60th percentile roundness Average roundness and maximum roundness .

5. The method according to claim 1, characterized in that, The improved PANet comprises a feature pyramid network, a bottom-up path enhancement network, a compression and activation attention layer, and a classifier connected in sequence. The feature pyramid network comprises multiple input layers, a top-down upsampling network, and an intermediate fusion network. The multiple input layers, from bottom to top, include four feature maps F1, F2, F3, and F4, derived from EfficientNet. The top-down upsampling network includes four feature maps P4, P3, P2, and P1, and an upsampling layer positioned between the feature maps. This upsampling layer is composed of depthwise separable deconvolutional layers. The intermediate fusion network fuses the corresponding feature maps from the input layers and the top-down upsampling network by concatenation. Feature maps F1 and F4 are connected to P4 via jump chains, and feature map F1 is connected to F4 via a jump chain. The bottom-up path enhancement network includes four feature maps N1, N2, N3, and N4 in sequence; N1 is P1, N2 is obtained by downsampling N1 and concatenating it with P2, N3 is obtained by downsampling N2 and concatenating it with P3, and N4 is obtained by downsampling N3 and concatenating it with P4; N4 is short-circuited to F1 and P1, and features are aggregated through a depthwise separable deconvolution layer; The compressed and excited attention layer takes feature aggregation input as input and extracts attention between channels to obtain weighted fused multi-scale features; The classifier includes a global average pooling layer and a fully connected layer, used to output the classification results of rock strength levels.

6. The method according to claim 5, characterized in that, The EfficientNet comprises, in sequence, an input layer, a Gaussian noise layer, a stem layer, a first Fused MBConv module, a second Fused MBConv module, a first MBConv module, a second MBConv module, a third MBConv module, a fourth MBConv module, and a head layer; the head layer comprises a convolutional layer, a BN layer, and a SiLU activation function. The second Fused MBConv module outputs F1', the first MBConv module outputs F2', the third MBConv module outputs F3', and the head layer outputs F4'; F1', F2', F3', and F4' are obtained by passing through convolutional layers to obtain F1, F2, F3, and F4, respectively.

7. A TBM tunnel (cavity) rock strength sensing system based on IPAN-EfficientNet network, characterized in that, include: The acquisition module is used to acquire images of rock debris from the TBM tunnel (cavity) and the corresponding rebound strength data of the rock in the tunnel wall; The dataset construction module is used to classify rock strength into multiple levels according to the rebound strength value, preprocess rock slag images, and establish a rock slag image-rock strength level dataset. The indicator filtering module is used to analyze the correlation between rock morphological parameters and rock strength based on the rock slag outline in the rock slag image, and to obtain the top-ranked indicators. The building block is used to construct the IPAN-EfficientNet network, which is formed by fusing the improved path aggregation network Improved PANet and EfficientNet. The training module is used to train the IPAN-EfficientNet network using a rock slag image-rock strength level dataset to obtain a rock strength level perception model. The output module is used to identify rock debris images using a rock strength level perception model in order to achieve rock strength level perception in TBM tunnels (caves).

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the TBM tunnel (cavity) rock strength sensing method based on the IPAN-EfficientNet network as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the TBM tunnel (cavity) rock strength sensing method based on the IPAN-EfficientNet network as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the TBM tunnel (cavity) rock strength sensing method based on the IPAN-EfficientNet network as described in any one of claims 1 to 6.

Images