Detection method and device of tunnel support steel grating, electronic equipment and storage medium

Abstract

The invention provides a tunnel support steel grating detection method and device, electronic equipment and a storage medium, and the method comprises the steps: inputting a tunnel construction picture into a feature detection module of a steel grating detection model, and determining a steel grating cross feature in the tunnel construction picture as a region of interest; inputting the region of interest and preset reinforcing steel bar priori knowledge into a region processing module, and performing expansion splicing on the region of interest to obtain a steel grating detection region; inputting the steel grating detection area into an edge detection module to obtain edge information of each reinforcing steel bar of the tunnel support steel grating; inputting the edge information of each reinforcing steel bar into a distance detection module, and calculating to obtain the distance between the reinforcing steel bars; and based on the distance between the steel bars and a preset distance threshold value, whether the tunnel supporting steel grating is correctly installed or not is determined. Thus, through the steel grating detection model, more accurate steel grating edge information can be obtained to detect the distance between the steel bars to judge whether the steel grating is correctly installed, and the accuracy of tunnel support steel grating detection is improved.

Description

Technical Field

[0001] The present application relates to the technical field of tunnel construction, and particularly to a detection method, device, electronic device and storage medium for tunnel support steel grids. Background Art

[0002] In order to achieve the coordinated and integrated development of rail transit and the city, the shallow tunneling method of constructing a tunnel by digging a hole underground without excavating the ground surface has become the preferred construction method for carrying out rail transit engineering construction in areas with dense surface buildings and heavy traffic. Among them, the steel grid, as an important strengthening support means in tunnel excavation, is used to pre-bear the surrounding rock pressure and restrain deformation before the shotcrete reaches the design strength, realize the support of the poor section of the chamber, and ensure the stability of the chamber structure and construction safety. However, in actual construction operations, there may be a situation where the depth of a single excavation is too deep and the installation and support of the steel grid are not carried out in a timely manner, which is extremely likely to cause a tunnel collapse accident. Therefore, it is necessary to detect whether the tunnel support steel grid is correctly installed to ensure the safety of construction.

[0003] The existing installation detection of tunnel support steel grids uses the Canny operator to perform edge detection on the construction site pictures, and judges whether the steel bars forming the steel grid are detected by extracting the edge parts of the steel bars. If no steel bars are detected, an alarm will be issued. However, the Canny operator mainly detects the edge information of objects. Due to the overly messy construction site, it is easy to capture other invalid information, resulting in the inability to determine whether the steel grid is correctly installed, and the accuracy of tunnel support steel grid detection is relatively low. Summary of the Invention

[0004] In view of this, the embodiments of the present application at least provide a detection method, device, electronic device and storage medium for tunnel support steel grids. Through the steel grid detection model, more accurate steel grid edge information can be obtained to detect the distance between steel bars to judge whether the steel grid is correctly installed, improving the accuracy of tunnel support steel grid detection.

[0005] The present application mainly includes the following aspects:

[0006] In a first aspect, the embodiments of the present application provide a detection method for a tunnel support steel grid, the method including:

[0007] Input a tunnel construction picture into the feature detection module of the steel grid detection model, and determine the steel grid cross feature in the tunnel construction picture as the region of interest of the tunnel construction picture; the steel grid detection model further includes a region processing module, an edge detection module and a distance detection module;

[0008] Input the region of interest and the preset prior knowledge of steel bars into the region processing module, expand and splice the region of interest to obtain the steel grid detection region of the tunnel construction picture;

[0009] Input the steel grid detection region into the edge detection module to obtain the edge information of each steel bar of the tunnel support steel grid in the tunnel construction picture;

[0010] Input the edge information of each steel bar into the distance detection module to calculate the distance between each steel bar;

[0011] Based on the distance between each steel bar and the preset distance threshold, determine whether the tunnel support steel grid is correctly installed.

[0012] In a second aspect, an embodiment of the present application further provides a detection device for a tunnel support steel grid, and the detection device for the tunnel support steel grid includes:

[0013] A feature extraction module, configured to input a tunnel construction picture into the feature detection module of the steel grid detection model, and determine the steel grid cross feature in the tunnel construction picture as the region of interest of the tunnel construction picture; the steel grid detection model further includes a region processing module, an edge detection module, and a distance detection module;

[0014] An expansion processing module, configured to input the region of interest and the preset prior knowledge of steel bars into the region processing module, expand and splice the region of interest to obtain the steel grid detection region of the tunnel construction picture;

[0015] A steel bar extraction module, configured to input the steel grid detection region into the edge detection module to obtain the edge information of each steel bar of the tunnel support steel grid in the tunnel construction picture;

[0016] A distance calculation module, configured to input the edge information of each steel bar into the distance detection module to calculate the distance between each steel bar;

[0017] A distance judgment module, configured to determine whether the tunnel support steel grid is correctly installed based on the distance between each steel bar and the preset distance threshold.

[0018] In a third aspect, an embodiment of the present application further provides an electronic device, including: a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device runs, the processor communicates with the memory through the bus. When the machine-readable instructions are run by the processor, the steps of the above-mentioned detection method for the tunnel support steel grid are executed.

[0019] Fourthly, an embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is run by a processor, it executes the steps of the detection method of the tunnel support steel grid as described above.

[0020] For the detection method, device, electronic device and storage medium of the tunnel support steel grid provided by the embodiment of the present application, a tunnel construction picture is input into the feature detection module of the steel grid detection model to determine the cross feature of the steel grid in the tunnel construction picture as the region of interest of the tunnel construction picture. The steel grid detection model further includes a region processing module, an edge detection module and a distance detection module. The region of interest and the preset prior knowledge of steel bars are input into the region processing module to expand and splice the region of interest to obtain the steel grid detection region of the tunnel construction picture. The steel grid detection region is input into the edge detection module to obtain the edge information of each steel bar of the tunnel support steel grid in the tunnel construction picture. The edge information of each steel bar is input into the distance detection module to calculate the distance between each steel bar. Based on the distance between each steel bar and the preset distance threshold, it is determined whether the tunnel support steel grid is correctly installed. In this way, through the steel grid detection model, more accurate edge information of the steel grid can be obtained to detect the distance between steel bars to judge whether the steel grid is correctly installed, improving the accuracy of the detection of the tunnel support steel grid.

[0021] To make the above objects, features and advantages of the present application more obvious and understandable, the following specifically gives preferred embodiments and cooperates with the attached drawings for detailed description as follows. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] To more clearly illustrate the technical solutions of the embodiments of the present application, the following briefly introduces the drawings required in the embodiments. It should be understood that the following drawings only show some embodiments of the present application and should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts.

[0023] Figure 1 Shows a flowchart of a detection method of a tunnel support steel grid provided by an embodiment of the present application;

[0024] Figure 2 Shows a schematic structural diagram of a steel grid detection model in an embodiment of the present application;

[0025] Figure 3 Shows a schematic structural diagram of a steel grid detection model in another embodiment of the present application;

[0026] Figure 4 Shows a functional module diagram of a detection device of a tunnel support steel grid provided by an embodiment of the present application;

[0027] Figure 5 The schematic structural diagram of an electronic device provided by an embodiment of the present application is shown. Specific implementation manners

[0028] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Apparently, the described embodiments are only some of the embodiments of the present application, rather than all of the embodiments. Usually, the components of the embodiments of the present application described and illustrated herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the present application claimed, but merely represents selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative efforts fall within the protection scope of the present application.

[0029] For the convenience of understanding the present application, the technical solutions provided by the present application will be described in detail below with reference to specific embodiments.

[0030] The embodiments of the present application provide a detection method, device, electronic device, and storage medium for a tunnel support steel grid. Through the steel grid detection model, more accurate steel grid edge information can be obtained to detect the distance between steel bars to determine whether the steel grid is installed correctly, improving the accuracy of tunnel support steel grid detection.

[0031] Please refer to Figure 1 , Figure 1 which is a flowchart of a detection method for a tunnel support steel grid provided by an embodiment of the present application. As Figure 1 shown, the detection method for a tunnel support steel grid provided by an embodiment of the present application includes the following steps:

[0032] S101, input a tunnel construction picture into the feature detection module of the steel grid detection model, and determine the steel grid cross feature in the tunnel construction picture as the region of interest of the tunnel construction picture; the steel grid detection model further includes a region processing module, an edge detection module, and a distance detection module.

[0033] Here, first, a camera arranged at the tunnel excavation site is used to capture on-site construction photos, and the real steel bar layout conditions of each construction scene under different conditions are collected as tunnel construction pictures, which are uploaded to the server terminal for detection. Then please refer to Figure 2 , Figure 2 which is a schematic structural diagram of the steel grid detection model in an embodiment of the present application. As Figure 2As shown in the figure, the steel grid detection model 200 includes a feature detection module 210, a region processing module 220, an edge detection module 230, and a distance detection module 240. The collected tunnel construction pictures are input into the feature detection module 210 to determine the steel grid cross features in the tunnel construction pictures, which are used as the regions of interest (ROIs) of the tunnel construction pictures. Among them, the steel grid cross features refer to the cross features formed between the steel bars that make up the steel grid. The region of interest (ROI) refers to the part of the image that contains the target object or key information. In this solution, it refers to the part of the image that contains the steel grid cross features.

[0034] S102: Input the region of interest and the preset prior knowledge of steel bars into the region processing module, expand and splice the region of interest to obtain the steel grid detection region of the tunnel construction picture.

[0035] Here, after obtaining the region of interest, the region of interest and the preset prior knowledge of steel bars are input into the region processing module 220 to obtain the steel grid detection region of the tunnel construction picture. Specifically, using the preset prior knowledge of steel bar thickness and bending degree, etc., expand the ROI regions at the heads and tails of the parallel steel bars on the plane, that is, expand and connect the corresponding ROI regions at the head and tail regions of each steel bar to obtain the steel grid detection region of the tunnel construction picture.

[0036] S103: Input the steel grid detection region into the edge detection module to obtain the edge information of each steel bar of the tunnel support steel grid in the tunnel construction picture.

[0037] Here, the steel grid detection region is input into the edge detection module 230 to obtain the edge information of each steel bar of the tunnel support steel grid in the tunnel construction picture. Specifically, in the embodiments of the present application, the edge detection module 230 adopts the EDTER (Efficient Dual-Transformer Edge Refinement) algorithm, which can utilize the complete image context information and detailed local features of the steel grid detection region to extract clear target boundaries and meaningful edges of each steel bar of the tunnel support steel grid in the tunnel construction picture.

[0038] S104: Input the edge information of each steel bar into the distance detection module, and calculate the distance between each steel bar.

[0039] Here, the edge information of each steel bar is input into the distance detection module 240 to calculate the actual distance between each steel bar. Specifically, in the embodiments of the present application, using the principle of similar triangles, combined with the pixel width, actual width of the steel bar, and the camera focal length, calculate the distance from each steel bar to the camera, and then obtain the spacing between each steel bar.

[0040] S105. Determine whether the tunnel support steel grid is correctly installed based on the distance between each steel bar and a preset distance threshold.

[0041] Here, based on the calculated distance between each steel bar and the preset distance threshold, it is determined whether the tunnel support steel grid is correctly installed. If the distance between the steel bars is within the preset safe range, it is considered that the steel grid is correctly installed; otherwise, an alarm is issued to prompt the construction personnel to make adjustments to ensure construction safety.

[0042] Further, the feature detection module includes a feature extraction module, a feature enhancement module, a detection frame positioning module, and a detection frame optimization module; inputting the tunnel construction picture into the feature detection module of the steel grid detection model to determine the steel grid cross feature in the tunnel construction picture as the region of interest of the tunnel construction picture includes:

[0043] Step a1. Input the tunnel construction picture into the feature extraction module to obtain a multi-scale feature map of the tunnel construction picture.

[0044] Here, please refer to Figure 3 , Figure 3 which is the second structural schematic diagram of the steel grid detection model in the embodiment of the present application. As Figure 3 shown, the feature detection module 210 includes a feature extraction module 211, a feature enhancement module 212, a detection frame positioning module 213, and a detection frame optimization module 214. Input the tunnel construction picture into the feature extraction module 211 to obtain a multi-scale feature map of the tunnel construction picture.

[0045] Specifically, the feature extraction module 211 of the embodiment of the present application adopts the backbone network of YOLOv7, mainly using the ELAN (Efficient Layer Aggregation Network) structure and the MP (Multi-Path) structure. Among them, the ELAN structure is an efficient neural network architecture mainly used for feature extraction and processing, which can improve the performance and efficiency of the model by optimizing the hierarchical structure and feature fusion method of the network. The MP structure is a network module for downsampling, which can improve the efficiency and effect of feature extraction by combining multiple downsampling methods.

[0046] In the embodiment of the present application, after adjusting the tunnel construction picture into a three-channel image of 640×640, it is sent to the feature extraction module 211. First, after extracting the shallow features of the steel bars through 4 CBS (Convolution-BatchNorm-SiLU) modules, it is sent to the ELAN structure. Among them, the CBS module consists of three main parts: a convolutional layer (Convolution), a batch normalization layer (Batch Normalization), and a SiLU activation function (Sigmoid Linear Unit), which are used to efficiently extract image features.

[0047] The ELAN structure can effectively learn and converge deeper networks by controlling the shortest and longest gradient paths, realizing the ability to continuously enhance the network learning ability without destroying the original gradient path. Specifically, in the embodiment of the present application, the picture C after extracting features from the tunnel construction picture is copied to obtain c. After c undergoes two 3*3 convolutions, c1 is generated. Then, after c1 undergoes two 3*3 convolutions, c2 is generated. Finally, c, c1, and c2 are concatenated and the deep features are extracted through a 1*1 convolutional kernel and sent to the MP structure.

[0048] The function of the MP structure is downsampling. The deep feature map D generated by ELAN is copied as d. Max pooling (max pooling) and convolution (conv) with a stride of 2 (stride = 2) are used simultaneously in two steps. Finally, the results of the two methods are concatenated, and better results can be obtained compared to using them separately.

[0049] By repeatedly calling ELAN and MP, on-site steel bar feature maps with different depths can be obtained. In the embodiment of the present application, a total of 3 depths are used. Upsampling the deeper features and concatenating them with the shallower features can obtain an on-site map that better shows the position of the steel bars. At the same time, in this upsampling process, the same method as above is used to fuse features of different depths multiple times, that is, concatenating the deepest layer feature map with the upsampled map, and then concatenating the deepest layer feature map with the twice upsampled map and sending it to the next stage again. This is beneficial for the manifestation of deep features and finally obtains a multi-scale feature map.

[0050] Step a2, input the multi-scale feature map into the feature enhancement module, and enhance the key information of the steel grid cross feature in the multi-scale feature map based on the channel attention mechanism and the spatial attention mechanism to obtain an enhanced multi-scale feature map.

[0051] Here, the feature enhancement module 212 processes the multi-scale feature maps through a channel attention mechanism and a spatial attention mechanism. In the embodiment of the present application, the feature enhancement module 212 is a CBAM (Convolutional Block Attention Module) module. Among them, the CBAM module is an attention mechanism module used to enhance the key information in the feature map. By combining channel attention and spatial attention, it improves the model's perception ability of important features, thereby enhancing the performance of the model. Corresponding to the embodiment of the present application, based on the channel attention mechanism and the spatial attention mechanism, the key information of the steel grid cross feature in the multi-scale feature map is enhanced to obtain the enhanced multi-scale feature map.

[0052] Step a3: Input the enhanced multi-scale feature map into the detection box localization module to predict the position and confidence of the steel grid cross feature, and generate multiple candidate detection boxes.

[0053] Here, the detection box localization module 213, based on the enhanced multi-scale feature map, adopts an anchor-based structure and predicts the position and confidence of the steel grid cross feature through the detection head (Head) of YOLOv7. This module outputs multiple candidate detection boxes, and each detection box contains position information (such as the center point coordinates, width, and height) and confidence, which are used for subsequent detection box optimization processing.

[0054] Step a4: Input the multiple candidate detection boxes into the detection box optimization module, and based on a preset confidence threshold and the non-maximum suppression algorithm, select at least one target detection box from the multiple candidate detection boxes as the region of interest of the tunnel construction picture.

[0055] Here, the multiple candidate detection boxes are input into the detection box optimization module 214. The detection box optimization module 214 first filters out the candidate detection boxes with higher confidence according to the preset confidence threshold and removes the misdetected candidate detection boxes with low confidence. Subsequently, through the non-maximum suppression (NMS) algorithm, the redundant detection boxes with high overlap are removed. Finally, at least one target detection box is retained as the region of interest of the tunnel construction picture. Among them, the NMS algorithm is used to remove redundant detection boxes and retain the optimal detection result to ensure that each target is detected only once and avoid multiple detection boxes pointing to the same target.

[0056] Furthermore, if the detection box localization module 213 does not detect any candidate detection boxes, that is, it does not detect the steel grid cross feature of the steel bars, it indicates that there is no steel grid in the tunnel construction picture, and an alarm will be issued to prompt the construction personnel to make adjustments to ensure construction safety.

[0057] Furthermore, the feature enhancement module includes a channel attention module, a spatial attention module, and a feature fusion module; inputting the multi-scale feature map into the feature enhancement module, and enhancing the key information of the steel grid cross feature in the multi-scale feature map based on the channel attention mechanism and the spatial attention mechanism to obtain the enhanced multi-scale feature map, including:

[0058] Step b1: Input the multi-scale feature map into the channel attention module to enhance the channel information of the steel grid cross feature in the multi-scale feature map, and obtain the multi-scale feature map after channel attention processing.

[0059] Here, as Figure 3 shown, the feature enhancement module 212 includes a channel attention module 2121, a spatial attention module 2122, and a feature fusion module 2123. The channel attention module 2121 performs global average pooling and global max pooling on the input multi-scale feature map to extract the global information of each channel. The two pooling results are processed by a shared multi-layer perceptron (MLP), and finally, the attention weights of each channel are generated through the Sigmoid activation function. These weights are multiplied with the input multi-scale feature map channel by channel to enhance the channel information of the steel grid cross feature in the multi-scale feature map, and obtain the multi-scale feature map after channel attention processing.

[0060] Step b2: Input the feature map after channel attention processing into the spatial attention module to enhance the spatial position information of the steel grid cross feature in the multi-scale feature map, and obtain the multi-scale feature map after spatial attention processing.

[0061] Here, the spatial attention module 2122 first performs average pooling and max pooling on the multi-scale feature map after channel attention processing in the channel dimension to extract the important features in the spatial dimension. Then, the two pooling results are concatenated and processed by a convolutional layer, and finally, the attention weights of each spatial position are generated through the Sigmoid activation function. These weights are multiplied with the multi-scale feature map after channel attention processing pixel by pixel to enhance the spatial position information of the steel grid cross feature in the multi-scale feature map, and obtain the multi-scale feature map after spatial attention processing.

[0062] Step b3: Input the multi-scale feature map after channel attention processing and the multi-scale feature map after spatial attention processing into the feature fusion module to obtain the enhanced multi-scale feature map.

[0063] Here, the feature fusion module 2123 fuses the multi-scale feature maps processed by channel attention and the multi-scale feature maps processed by spatial attention. By means of weighted summation or concatenation, etc., the attention information in the channel and spatial dimensions is combined to generate enhanced multi-scale feature maps. This process retains the feature information of key channels and spatial positions, providing a richer feature representation for subsequent detection tasks.

[0064] Further, inputting the multiple candidate detection boxes into the detection box optimization module, and screening at least one target detection box from the multiple candidate detection boxes based on a preset confidence threshold and a non-maximum suppression algorithm as the region of interest of the tunnel construction picture includes:

[0065] Step c1, sorting the multiple candidate detection boxes in descending order of confidence, and selecting the candidate detection box with the highest confidence as the reference detection box.

[0066] Here, first, all candidate detection boxes are sorted according to confidence to ensure the processing order is from high confidence to low confidence, so as to preferentially retain the detection box with the highest confidence as the reference detection box.

[0067] Step c2, calculating the intersection over union (IoU) of the confidence of the reference detection box and other candidate detection boxes.

[0068] Here, by calculating the intersection over union (IoU) of the confidence of the reference detection box and the remaining candidate detection boxes, the overlap degree between them is evaluated, providing a basis for subsequent threshold adjustment. Among them, the IoU is used to measure the overlap degree between two bounding boxes.

[0069] Step c3, for any candidate detection box other than the reference detection box, dynamically adjusting the IoU threshold according to the confidence difference between the reference detection box and the candidate detection box.

[0070] Here, the IoU threshold is dynamically adjusted according to the confidence difference to adapt to detection boxes with different confidences. For example, if the confidence of the reference detection box is 0.9 and the confidence of another candidate detection box is 0.8, the IoU threshold may be relaxed from 0.5 to 0.6 to allow the retention of adjacent targets. Here, the purpose of dynamically adjusting the IoU threshold is to use a looser threshold for high-confidence detection boxes (small confidence difference) to allow the retention of adjacent targets; for low-confidence detection boxes (large confidence difference), use a stricter threshold to reduce false detections.

[0071] Step c4, if the intersection over union of the confidence of the reference detection box and the candidate detection box is greater than the dynamically adjusted IoU threshold, then remove the candidate detection box.

[0072] Here, for each candidate detection box, if its IoU with the reference detection box exceeds the dynamically adjusted IoU threshold, it is considered that these two detection boxes point to the same target. Therefore, the candidate detection box is removed to reduce false detections.

[0073] Step c5: Repeat until all candidate detection boxes except the reference detection box are processed, and at least one target detection box is selected as the region of interest of the tunnel construction picture.

[0074] Here, steps c3 and c4 are repeated to process the remaining candidate detection boxes in turn until all candidate detection boxes are processed, and finally at least one target detection box is selected as the region of interest of the tunnel construction picture.

[0075] Furthermore, the edge detection module includes a global feature extraction module, a local feature refinement module, a gating fusion module, and an edge generation module; inputting the steel grid detection area into the edge detection module to obtain the edge information of each steel bar of the tunnel support steel grid in the tunnel construction picture, including:

[0076] Step d1: Input the steel grid detection area into the global feature extraction module to extract the global features of the steel grid detection area.

[0077] Here, as Figure 3 shown, the edge detection module 230 includes a global feature extraction module 231, a local feature refinement module 232, a gating fusion module 233, and an edge generation module 234. First, the global feature extraction module 231 divides the steel grid detection area into multiple coarse-grained regions (patches) of 16×16 pixels and converts these regions into embedding vectors as the input of the encoder. In the embodiment of the present application, the global feature extraction module 231 uses a global Transformer encoder module to act on these embeddings, calculates global attention, and extracts the part that is really a steel bar in the ROI area. Then, the edge information of the global features is extracted through a BiMLA (Bilinear Multi-Level Aggregation) decoder module. Among them, the Transformer encoder is used to process the input sequence through the self-attention mechanism and the feed-forward neural network (FFNN), and can effectively capture the long-range dependencies in the sequence. The BiMLA decoder can effectively extract edge information and generate high-quality feature maps through the fusion of multi-scale features.

[0078] In the embodiment of the present application, the BiMLA module increases the channels of each input patch (assuming there are 4, namely a, b, c, and d) through 1×1 convolution.

[0079] For each patch, 3×3 convolution is performed to extract shallow features, and after deconvolution, 3×3 convolution is performed again to extract deep features, and finally deconvolution is performed.

[0080] Between different patches, feature fusion is required. Therefore, before the first convolution, the four patches are concatenated and processed to generate a, a + b, a + b + c, a + b + c + d, which are denoted as a1, b1, c1, d1; before the second convolution, a1 + b1 + c1 + d1, b1 + c1 + d1, c1 + d1, d1 are concatenated. Finally, all the results are concatenated and passed through three 3×3 convolutions and one 1×1 convolution to change the number of channels to obtain the final result. In this way, the features of each block of the steel bar image and a relatively rough overall edge are obtained respectively, as the global features of the steel grating detection area.

[0081] Step d2: Input the global feature into the local feature refinement module to extract the local feature of the global feature.

[0082] Here, the local feature refinement module 232 uses a sliding window of H / 2×W / 2 to divide the global feature into multiple fine-grained patches of 4×4 pixels. Each window calculates local attention through a local Transformer encoder to generate local high-resolution features. These local features are fused with the global feature to further refine the edge information and obtain the local feature of the global feature.

[0083] Specifically, a sliding window of H / 2×W / 2 is used to divide the input image into a sequence of {X1, X2, X3, X4}. For each window, it is divided into 4×4 fine-grained patches, and then local attention is calculated through a local Transformer encoder. Then, the attention features of all windows are concatenated. Next, similar to the global BiMLA module, it is input into the local BiMLA module to generate local high-resolution features. This can enable the model to detect a more refined steel bar edge and add edge details on the basis of the previous global one.

[0084] Step d3: Input the global feature and the local feature into the gated fusion module, and fuse the global feature and the local feature based on the gated mechanism to obtain the comprehensive feature of the steel grating detection area.

[0085] Here, the gated fusion module 233 fuses the global feature F and the local feature f. Specifically, feature concatenation is performed based on the global feature F and the local feature f, and a gated residual path is added beside the concatenation path. The fusion ratio of the global feature F and the local feature f is dynamically adjusted through a gating mechanism, enhancing the expression ability of key features while retaining the basic information of the original global feature to ensure the adaptability of the model to different regional features. Finally, the generated comprehensive feature can more accurately represent the edge information of the steel grating detection area, providing a richer feature representation for subsequent edge detection.

[0086] Step d4, input the comprehensive feature into the edge generation module to generate the edge information of each steel bar of the tunnel support steel grating in the tunnel construction picture.

[0087] Here, the edge generation module 234 uses the comprehensive feature to generate the final edge detection result. Specifically, the comprehensive feature generates an edge map through a convolution operation, and then the edge map is converted into a binary edge map through a thresholding operation, so as to obtain clear steel bar edge information, which serves as the edge information of each steel bar of the tunnel support steel grating in the tunnel construction picture.

[0088] Further, the inputting the global feature and the local feature into the gated fusion module, fusing the global feature and the local feature based on a gating mechanism, and obtaining the comprehensive feature of the steel grating detection area includes:

[0089] Step e1, concatenate the global feature and the local feature along the channel dimension to generate a mixed feature tensor of the global feature and the local feature.

[0090] Here, first, two convolution operations are performed on the global feature F to obtain a deep global feature F1, and then the deep global feature F1 and the local feature f are concatenated along the channel dimension in the concatenation path. Specifically, after taking the dot product of the deep global feature F1 and the local feature f, preliminary steel bar edge information is obtained, and then the preliminary steel bar edge information and the deep global feature F1 are concatenated to obtain a mixed feature tensor containing global semantics and local details. This step retains the basic combination of the global feature and the local feature, providing a rich feature representation for subsequent processing.

[0091] Step e2, perform a channel compression convolution operation on the mixed feature tensor to generate a gating weight map.

[0092] Here, a gated residual path is added beside the concatenated path, and a 1×1 convolution operation is performed on the concatenated hybrid feature tensor to compress the number of channels to be the same as that of the deep global feature F1. Then, through the Sigmoid activation function, a gated weight map G with a value range between [0,1] is generated. The larger the spatial position value of this weight map, the higher the contribution degree of the local detail features at the corresponding position to the final edge prediction.

[0093] Step e3, multiply the gated weight map by the hybrid feature tensor to obtain a weighted feature tensor.

[0094] Here, multiply the gated weight map G by the hybrid feature tensor to obtain a weighted feature tensor. At this time, G acts as a feature selector: in the edge region (such as the steel bar boundary), G is close to 1, and the detailed information of the local feature f is fully retained; in the non-edge region (such as the background), G is close to 0, and the irrelevant details are filtered.

[0095] Step e4, perform a residual connection between the weighted feature tensor and the global feature to obtain the comprehensive feature of the steel grating detection area.

[0096] Here, add the gated-fused feature to the original global feature F1 for residual connection to form a residual structure, and obtain the comprehensive feature of the steel grating detection area through two convolution operations. Among them, the introduction of the residual connection ensures two key features: smooth gradient propagation: avoiding the problem of gradient disappearance during the training of deep networks; protection of original information: even if the gated weight judgment is incorrect, the basic edge structure can still be retained through F1.

[0097] Further, the step of inputting the edge information of each steel bar into the distance detection module and calculating the distance between each steel bar includes:

[0098] Step f1, based on the edge information of each steel bar, determine the pixel width of each steel bar in the tunnel construction picture.

[0099] Here, the distance detection module 240 calculates the pixel width of each steel bar in the image through the edge information generated by the edge detection module 230. Specifically, the pixel width of each steel bar can be obtained by calculating the pixel distance between the steel bar boundaries in the edge information.

[0100] Step f2, based on the pixel width, the width of the steel bar, and the focal length of the camera that captured the tunnel construction picture, determine the distance between each steel bar; the focal length of the camera that captured the tunnel construction picture is determined according to a reference object with a preset known width, the preset distance from the reference object to the camera, and the pixel width of the reference object in the tunnel construction picture.

[0101] Here, the principle of similar triangles is used to calculate the actual distance between steel bars. Suppose there is a reference object with a width of W. Then place this reference object at a position D away from the camera that takes pictures of the tunnel construction. Take a picture of the reference object with the camera and measure the pixel width P of the reference object. The formula for the focal length of the camera is obtained: F = (P x D) / W. When the camera continues to move closer to or farther away from the object or target, the distance between the object and the camera can be calculated using similar triangles: D' = (W x F) / P. Furthermore, based on the pixel width, the width of the steel bars, and the focal length of the camera that takes pictures of the tunnel construction, the distance between each steel bar can be determined.

[0102] A detection method for a tunnel support steel grid provided by an embodiment of the present application includes: inputting a tunnel construction picture into a feature detection module of a steel grid detection model to determine the steel grid cross feature in the tunnel construction picture as the region of interest of the tunnel construction picture; the steel grid detection model further includes a region processing module, an edge detection module, and a distance detection module; inputting the region of interest and preset prior knowledge of steel bars into the region processing module to expand and splice the region of interest to obtain the steel grid detection region of the tunnel construction picture; inputting the steel grid detection region into the edge detection module to obtain the edge information of each steel bar of the tunnel support steel grid in the tunnel construction picture; inputting the edge information of each steel bar into the distance detection module to calculate the distance between each steel bar; and determining whether the tunnel support steel grid is correctly installed based on the distance between each steel bar and a preset distance threshold. In this way, through the steel grid detection model, more accurate steel grid edge information can be obtained to detect the distance between steel bars to determine whether the steel grid is correctly installed, improving the accuracy of tunnel support steel grid detection.

[0103] Based on the same inventive concept, an embodiment of the present application also provides a detection device for a tunnel support steel grid corresponding to the detection method for a tunnel support steel grid provided in the above embodiment. Since the principle of solving problems by the device in the embodiment of the present application is similar to the detection method for a tunnel support steel grid in the above embodiment of the present application, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be described again.

[0104] Please refer to Figure 4 , which is a functional module diagram of a detection device for a tunnel support steel grid provided by an embodiment of the present application. As Figure 4 shown, the detection device 400 for a tunnel support steel grid includes:

[0105] A feature extraction module 410, configured to input a tunnel construction picture into a feature detection module of a steel grid detection model to determine the steel grid cross feature in the tunnel construction picture as the region of interest of the tunnel construction picture; the steel grid detection model further includes a region processing module, an edge detection module, and a distance detection module.

[0106] An expansion processing module 420 is configured to input the region of interest and preset prior knowledge of steel bars into the region processing module, expand and splice the region of interest, and obtain the steel grid detection region of the tunnel construction picture.

[0107] A steel bar extraction module 430 is configured to input the steel grid detection region into the edge detection module to obtain the edge information of each steel bar of the tunnel support steel grid in the tunnel construction picture.

[0108] A distance calculation module 440 is configured to input the edge information of each steel bar into the distance detection module to calculate the distance between each steel bar.

[0109] A distance judgment module 450 is configured to determine whether the tunnel support steel grid is correctly installed based on the distance between each steel bar and a preset distance threshold.

[0110] Further, the feature detection module includes a feature extraction module, a feature enhancement module, a detection frame positioning module, and a detection frame optimization module; when the feature extraction module 410 is configured to input the tunnel construction picture into the feature detection module of the steel grid detection model to determine the steel grid cross feature in the tunnel construction picture as the region of interest of the tunnel construction picture, the feature extraction module 410 is specifically configured to:

[0111] Input the tunnel construction picture into the feature extraction module to obtain the multi-scale feature map of the tunnel construction picture;

[0112] Input the multi-scale feature map into the feature enhancement module, enhance the key information of the steel grid cross feature in the multi-scale feature map based on the channel attention mechanism and the spatial attention mechanism, and obtain the enhanced multi-scale feature map;

[0113] Input the enhanced multi-scale feature map into the detection frame positioning module, predict the position and confidence of the steel grid cross feature, and generate a plurality of candidate detection frames;

[0114] Input the plurality of candidate detection frames into the detection frame optimization module, and screen out at least one target detection frame from the plurality of candidate detection frames as the region of interest of the tunnel construction picture based on a preset confidence threshold and the non-maximum suppression algorithm.

[0115] Further, the feature enhancement module includes a channel attention module, a spatial attention module, and a feature fusion module; when the feature extraction module 410 is used to input the multi-scale feature map into the feature enhancement module and enhance the key information of the steel grid cross feature in the multi-scale feature map based on the channel attention mechanism and the spatial attention mechanism to obtain the enhanced multi-scale feature map, the feature extraction module 410 is specifically used for:

[0116] Input the multi-scale feature map into the channel attention module to enhance the channel information of the steel grid cross feature in the multi-scale feature map, and obtain the multi-scale feature map after channel attention processing;

[0117] Input the feature map after channel attention processing into the spatial attention module to enhance the spatial position information of the steel grid cross feature in the multi-scale feature map, and obtain the multi-scale feature map after spatial attention processing;

[0118] Input the multi-scale feature map after channel attention processing and the multi-scale feature map after spatial attention processing into the feature fusion module to obtain the enhanced multi-scale feature map.

[0119] Further, when the feature extraction module 410 is used to input the multiple candidate detection frames into the detection frame optimization module and screen out at least one target detection frame from the multiple candidate detection frames based on the preset confidence threshold and the non-maximum suppression algorithm as the region of interest of the tunnel construction picture, the feature extraction module 410 is specifically used for:

[0120] Sort the multiple candidate detection frames in descending order of confidence, and select the candidate detection frame with the highest confidence as the reference detection frame;

[0121] Calculate the intersection over union (IoU) of the confidence of the reference detection frame and other candidate detection frames;

[0122] For any candidate detection frame other than the reference detection frame, dynamically adjust the IoU threshold according to the confidence difference between the reference detection frame and the candidate detection frame;

[0123] If the IoU of the confidence of the reference detection frame and the candidate detection frame is greater than the dynamically adjusted IoU threshold, remove the candidate detection frame;

[0124] Repeat until all candidate detection frames other than the reference detection frame are processed, and screen out at least one target detection frame as the region of interest of the tunnel construction picture.

[0125] Further, the edge detection module includes a global feature extraction module, a local feature refinement module, a gated fusion module, and an edge generation module; when the steel bar extraction module 430 is used to input the steel grid detection area into the edge detection module to obtain the edge information of each steel bar of the tunnel support steel grid in the tunnel construction picture, the steel bar extraction module 430 is specifically used for:

[0126] Input the steel grid detection area into the global feature extraction module to extract the global features of the steel grid detection area;

[0127] Input the global features into the local feature refinement module to extract the local features of the global features;

[0128] Input the global features and the local features into the gated fusion module, and fuse the global features and the local features based on the gated mechanism to obtain the comprehensive features of the steel grid detection area;

[0129] Input the comprehensive features into the edge generation module to generate the edge information of each steel bar of the tunnel support steel grid in the tunnel construction picture.

[0130] Further, when the steel bar extraction module 430 is used to input the global features and the local features into the gated fusion module, and fuse the global features and the local features based on the gated mechanism to obtain the comprehensive features of the steel grid detection area, the steel bar extraction module 430 is specifically used for:

[0131] Concatenate the global features and the local features along the channel dimension to generate a mixed feature tensor of the global features and the local features;

[0132] Perform a channel compression convolution operation on the mixed feature tensor to generate a gated weight map;

[0133] Multiply the gated weight map by the mixed feature tensor to obtain a weighted feature tensor;

[0134] Perform a residual connection on the weighted feature tensor and the global features to obtain the comprehensive features of the steel grid detection area.

[0135] Further, when the distance calculation module 440 is used to input the edge information of each steel bar into the distance detection module to calculate the distance between each steel bar, the distance calculation module 440 is specifically used for:

[0136] Based on the edge information of each steel bar, determine the pixel width of each steel bar in the tunnel construction picture;

[0137] Determine the distance between each steel bar based on the pixel width, the width of the steel bar, and the focal length of the camera that captures the tunnel construction picture; the focal length of the camera that captures the tunnel construction picture is determined based on a reference object with a preset known width, the preset distance from the reference object to the camera, and the pixel width of the reference object in the tunnel construction picture.

[0138] A detection device for a tunnel support steel grid provided by an embodiment of the present application includes: a feature extraction module, configured to input a tunnel construction picture into the feature detection module of the steel grid detection model to determine the steel grid cross feature in the tunnel construction picture as the region of interest of the tunnel construction picture; the steel grid detection model further includes a region processing module, an edge detection module, and a distance detection module; an expansion processing module, configured to input the region of interest and preset prior knowledge of the steel bar into the region processing module to expand and splice the region of interest to obtain the steel grid detection region of the tunnel construction picture; a steel bar extraction module, configured to input the steel grid detection region into the edge detection module to obtain the edge information of each steel bar of the tunnel support steel grid in the tunnel construction picture; a distance calculation module, configured to input the edge information of each steel bar into the distance detection module to calculate the distance between each steel bar; a distance judgment module, configured to determine whether the tunnel support steel grid is correctly installed based on the distance between each steel bar and a preset distance threshold. In this way, through the steel grid detection model, more accurate edge information of the steel grid can be obtained to detect the distance between the steel bars to determine whether the steel grid is correctly installed, improving the accuracy of the tunnel support steel grid detection.

[0139] Based on the same inventive concept, please refer to Figure 5 , Figure 5 which is a schematic structural diagram of an electronic device provided by an embodiment of the present application. As Figure 5 shown, the electronic device 500 includes: a processor 510, a memory 520, and a bus 530.

[0140] The memory 520 stores machine-readable instructions executable by the processor 510. When the electronic device 500 runs, the processor 510 communicates with the memory 520 through the bus 530. When the machine-readable instructions are run by the processor 510, the steps of the detection method for the tunnel support steel grid provided in the above embodiment are executed. The specific implementation manner can be seen in the method embodiment and will not be elaborated here.

[0141] Based on the same inventive concept, an embodiment of the present application further provides a computer-readable storage medium. A computer program is stored on the computer-readable storage medium. When the computer program is run by a processor, the steps of the detection method for the tunnel support steel grid provided in the above embodiment are executed. The specific implementation manner can be seen in the method embodiment and will not be elaborated here.

[0142] Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working processes of the above-described devices and units can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0143] In the embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division, and there can be other division methods in actual implementation. For another example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection between each other can be through some communication interfaces. The indirect coupling or communication connection of the devices or units can be in electrical, mechanical or other forms.

[0144] The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0145] In addition, each functional unit in the embodiments provided in the present application can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit.

[0146] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art or part of this technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0147] It should be noted that like reference numerals and letters denote like items in the following figures. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. In addition, the terms "first", "second", "third", etc. are only used for descriptive distinction and should not be construed as indicating or implying relative importance.

[0148] Finally, it should be noted that the above-described embodiments are only specific embodiments of the present application, which are used to illustrate the technical solutions of the present application, rather than to limit it. The protection scope of the present application is not limited thereto. Although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that any person skilled in the art within the technical scope disclosed by the present application can still modify the technical solutions described in the foregoing embodiments or can easily think of changes, or make equivalent replacements for some of the technical features; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application. All should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A detection method for a tunnel support steel grid, characterized in that, The method includes: Inputting the tunnel construction picture into the feature detection module of the steel grid detection model to determine the steel grid cross feature in the tunnel construction picture as the region of interest of the tunnel construction picture; the steel grid detection model further includes a region processing module, an edge detection module, and a distance detection module; Inputting the region of interest and the preset prior knowledge of steel bars into the region processing module to expand and splice the region of interest to obtain the steel grid detection region of the tunnel construction picture; Inputting the steel grid detection region into the edge detection module to obtain the edge information of each steel bar of the tunnel support steel grid in the tunnel construction picture; Inputting the edge information of each steel bar into the distance detection module to calculate the distance between each steel bar; Based on the distance between each steel bar and the preset distance threshold, determining whether the tunnel support steel grid is correctly installed.

2. The detection method of the tunnel support steel grid according to claim 1, characterized in that The feature detection module includes a feature extraction module, a feature enhancement module, a detection box positioning module, and a detection box optimization module; the step of inputting the tunnel construction picture into the feature detection module of the steel grid detection model to determine the steel grid cross feature in the tunnel construction picture as the region of interest of the tunnel construction picture includes: Inputting the tunnel construction picture into the feature extraction module to obtain the multi-scale feature map of the tunnel construction picture; Inputting the multi-scale feature map into the feature enhancement module to enhance the key information of the steel grid cross feature in the multi-scale feature map based on the channel attention mechanism and the spatial attention mechanism to obtain the enhanced multi-scale feature map; Inputting the enhanced multi-scale feature map into the detection box positioning module to predict the position and confidence of the steel grid cross feature and generate multiple candidate detection boxes; Inputting the multiple candidate detection boxes into the detection box optimization module to screen out at least one target detection box from the multiple candidate detection boxes as the region of interest of the tunnel construction picture based on the preset confidence threshold and the non-maximum suppression algorithm.

3. The inspection method of the tunnel support steel grid according to claim 2, characterized in that, The feature enhancement module includes a channel attention module, a spatial attention module, and a feature fusion module; the step of inputting the multi-scale feature map into the feature enhancement module to enhance the key information of the steel grid cross feature in the multi-scale feature map based on the channel attention mechanism and the spatial attention mechanism to obtain the enhanced multi-scale feature map includes: Inputting the multi-scale feature map into the channel attention module to enhance the channel information of the steel grid cross feature in the multi-scale feature map to obtain the multi-scale feature map processed by the channel attention; Inputting the feature map processed by the channel attention into the spatial attention module to enhance the spatial position information of the steel grid cross feature in the multi-scale feature map to obtain the multi-scale feature map processed by the spatial attention; Inputting the multi-scale feature map processed by the channel attention and the multi-scale feature map processed by the spatial attention into the feature fusion module to obtain the enhanced multi-scale feature map.

4. The inspection method of the tunnel support steel grid according to claim 2, characterized in that Inputting the multiple candidate detection frames into the detection frame optimization module, and screening at least one target detection frame from the multiple candidate detection frames based on a preset confidence threshold and a non-maximum suppression algorithm as the region of interest of the tunnel construction picture, includes: Sorting the multiple candidate detection frames in descending order of confidence, and selecting the candidate detection frame with the highest confidence as the reference detection frame; Calculating the intersection over union (IoU) of the confidence of the reference detection frame and other candidate detection frames; For any candidate detection frame other than the reference detection frame, dynamically adjusting the IoU threshold according to the confidence difference between the reference detection frame and the candidate detection frame; If the IoU of the confidence of the reference detection frame and the candidate detection frame is greater than the dynamically adjusted IoU threshold, removing the candidate detection frame; Repeating until all candidate detection frames other than the reference detection frame are processed, and screening at least one target detection frame as the region of interest of the tunnel construction picture.

5. The detection method of the tunnel support steel grid according to claim 1, characterized in that The edge detection module includes a global feature extraction module, a local feature refinement module, a gated fusion module, and an edge generation module; inputting the steel grid detection region into the edge detection module to obtain the edge information of each steel bar of the tunnel support steel grid in the tunnel construction picture, includes: Inputting the steel grid detection region into the global feature extraction module to extract the global features of the steel grid detection region; Inputting the global features into the local feature refinement module to extract the local features of the global features; Inputting the global features and the local features into the gated fusion module, and fusing the global features and the local features based on a gated mechanism to obtain the comprehensive features of the steel grid detection region; Inputting the comprehensive features into the edge generation module to generate the edge information of each steel bar of the tunnel support steel grid in the tunnel construction picture.

6. The detection method of the tunnel support steel grid according to claim 5, characterized in that Inputting the global features and the local features into the gated fusion module, and fusing the global features and the local features based on a gated mechanism to obtain the comprehensive features of the steel grid detection region, includes: Concatenating the global features and the local features along the channel dimension to generate a mixed feature tensor of the global features and the local features; Performing a channel compression convolution operation on the mixed feature tensor to generate a gated weight map; Multiplying the gated weight map by the mixed feature tensor to obtain a weighted feature tensor; Performing a residual connection on the weighted feature tensor and the global features to obtain the comprehensive features of the steel grid detection region.

7. The inspection method of the tunnel support steel grid according to claim 1, characterized in that Inputting the edge information of each steel bar into the distance detection module, and calculating the distance between each steel bar, includes: Based on the edge information of each steel bar, determining the pixel width of each steel bar in the tunnel construction picture; Determine the distance between each steel bar based on the pixel width, the width of the steel bar, and the focal length of the camera that captures the tunnel construction picture; the focal length of the camera that captures the tunnel construction picture is determined according to a reference object with a preset known width, the preset distance from the reference object to the camera, and the pixel width of the reference object in the tunnel construction picture.

8. A detection device for a tunnel support steel grid, characterized in that, The detection device for the tunnel support steel grid includes: A feature extraction module, configured to input the tunnel construction picture into the feature detection module of the steel grid detection model, and determine the steel grid cross feature in the tunnel construction picture as the region of interest of the tunnel construction picture; the steel grid detection model further includes a region processing module, an edge detection module, and a distance detection module; An expansion processing module, configured to input the region of interest and the preset prior knowledge of the steel bar into the region processing module, expand and splice the region of interest to obtain the steel grid detection region of the tunnel construction picture; A steel bar extraction module, configured to input the steel grid detection region into the edge detection module to obtain the edge information of each steel bar of the tunnel support steel grid in the tunnel construction picture; A distance calculation module, configured to input the edge information of each steel bar into the distance detection module to calculate the distance between each steel bar; A distance judgment module, configured to determine whether the tunnel support steel grid is correctly installed based on the distance between each steel bar and a preset distance threshold.

9. An electronic device, characterized in that, including: A processor, a memory, and a bus, where the memory stores machine-readable instructions executable by the processor. When the electronic device runs, the processor communicates with the memory through the bus. When the machine-readable instructions are run by the processor, the steps of the detection method for the tunnel support steel grid according to any one of claims 1 to 7 are executed.

10. A computer-readable storage medium, characterized in that, A computer program is stored on the computer-readable storage medium. When the computer program is run by the processor, the steps of the detection method for the tunnel support steel grid according to any one of claims 1 to 7 are executed.

Images