Tunnel water inrush automatic identification method and system based on water inrush target detection model

Abstract

The invention provides a tunnel water inrush automatic identification method and system based on a water inrush target detection model, and belongs to the technical field of tunnel water inrush disaster monitoring, prevention and control. The method comprises the steps of extracting video image features in a tunnel; mixing the space attention weight and the channel attention weight of the features, and updating the features; fusing the updated features; and decoding the fused features to obtain a target frame and a category result. According to the method, context features are coupled through a content awareness attention mechanism, the generalization of a monitoring model is improved, and accurate recognition of the tunnel gushing water under various tunnel conditions is guaranteed; omnibearing high-precision detection and statistical analysis of the water inrush of the tunnel are further realized, and whether early warning prompt in the tunnel is carried out is judged, so that the evolution condition of the water inrush is predicted, and a guarantee is provided for safe construction of the tunnel.

Description

Technical Field

[0001] The present invention relates to the technical field of monitoring and preventing and controlling water inrush disasters in tunnels, and in particular to a method and system for automatically identifying water inrush in tunnels based on a water inrush target detection model. Background Art

[0002] During the construction of deep-buried drill-and-blast tunnels, disasters such as sudden water inrush in tunnels are often caused by crossing high mountain canyon areas, strong water-rich areas, fault fracture zones, karst caves and other environments. Due to the complexity of the geological conditions in the tunnel construction area and the uncertainty in the construction process, more accurate monitoring of sudden water inrush disasters during tunnel construction is particularly important to ensure safe tunnel construction and avoid casualties. Monitoring is an indispensable step in dealing with sudden water inrush disasters in tunnels.

[0003] The patent "Unmanned Monitoring and Early Warning Method, Device and System for Tunnel Water Burst by Machine Vision", patent number "CN202310786056.3" uses the method of calculating optical flow vectors to calculate tunnel water bursts, which is easily affected by light and dust. When the dust in the tunnel is large, the accuracy of this method is greatly reduced, and the staff cannot accurately find the water burst point from the video.

[0004] Therefore, a method for accurately identifying tunnel water inrush is needed. Summary of the invention

[0005] In view of this, the present invention provides a method and system for automatically identifying sudden water surges in tunnels based on a sudden water surge target detection model. By constructing a sudden water surge target detection model for automatic identification of sudden water surges, accurate identification of sudden water surges in tunnels under various tunnel conditions is ensured, thereby achieving all-round high-precision detection and statistical analysis of sudden water surges in tunnels.

[0006] To this end, the present invention provides the following technical solutions:

[0007] A method for automatically identifying sudden water inrush in a tunnel based on a sudden water inrush target detection model, comprising:

[0008] Construct a water inrush target detection model;

[0009] The sudden water target detection model includes:

[0010] Extract video image features in tunnels;

[0011] Mix the spatial attention weights and channel attention weights of the features and update the features;

[0012] Fusion of updated features;

[0013] The fused features are decoded to obtain the target frame and category results.

[0014] Furthermore, the extracting of video image features in the tunnel includes:

[0015] Feature extraction is performed through the ResNet50 network to obtain 4 layers of features;

[0016] The encoder obtains new features of the fourth layer of features among the four layers of features.

[0017] Furthermore, the spatial attention weight and channel attention weight update features of the hybrid feature include:

[0018] Obtaining new features of the fourth layer of features and channel attention weights of the first layer of features, the second layer of features, and the third layer of features in the four layers of features through convolution, global average pooling, and channel shuffling;

[0019] Obtain new features of the fourth layer of features and spatial attention weights of the first layer features, the second layer features, and the third layer features in the four layers of features through global average pooling, global maximum pooling, and channel shuffling;

[0020] Mixing the channel attention weight and the spatial attention weight;

[0021] Update features via content-aware attention mechanism.

[0022] Furthermore, the updated features are fused through a multi-layer feature fusion module:

[0023]

[0024] f l =f l-1 +f l+1 ,l=7

[0025] f k =f k-4 +DS(f k-1 ),k=8,9

[0026] F = PConv(f 10 )+PConv(f9)+PConv(DS(f8))+PConv(DS(f7)

[0027] Among them, PConv represents the partial convolution operation, f i represents features, DS represents downsampling operation, and BL represents bilinear interpolation upsampling operation.

[0028] Furthermore, decoding the fused features to obtain a target frame and a category result includes:

[0029] The fused feature F is passed through the transformer decoder module and the feedforward neural network to obtain the target box and category results.

[0030] Furthermore, the category results include:

[0031] Water surge points and water surges.

[0032] Furthermore, before constructing the water inrush target detection model, the method further includes: constructing a water inrush target detection model training data set.

[0033] Furthermore, the construction of a water inrush target detection model training data set includes:

[0034] Collect water inrush video frames and perform preprocessing;

[0035] Annotate and divide the preprocessed video frames as training data sets for water inrush target detection models;

[0036] The pre-processing comprises:

[0037] Capture the video frame of sudden water inrush in the tunnel surveillance video;

[0038] The water inrush video frames are decomposed into sub-bands of different frequencies by wavelet denoising;

[0039] Threshold processing is performed on sub-bands of different frequencies to remove noise from the water inrush video frames.

[0040] A tunnel water inrush automatic identification system based on a water inrush target detection model, comprising:

[0041] The backbone module is used to extract the features of the video images in the tunnel;

[0042] Content-aware module, which mixes the spatial attention weights and channel attention weights of features and updates features;

[0043] Multi-layer feature fusion module, which fuses the updated features;

[0044] The recognition module decodes the fused features to obtain the target box and category results.

[0045] Advantages and positive effects of the present invention:

[0046] The present invention monitors the video in the tunnel in real time through an industrial camera; and uses a water gushing detection target detection model to perform target detection on the detected video image, and automatically identifies water gushing points and water gushing. In the present invention, the locations where water gushing may occur, such as the tunnel face and the arch waist, are detected; and a fill light is provided to improve the brightness of the video. For the water gushing target detection model in the present invention, a real water gushing target detection model training set is constructed by obtaining frames of the actual water gushing video; the content-aware attention mechanism is used to obtain specific important spatial information in each channel, and to ensure the full mixing of spatial and channel attention features, thereby ensuring information interaction; and the problem that the tunnel water gushing points are small targets in the image, occupy a small proportion of the total pixels, and the foreground and background colors of the image are almost the same, which is not conducive to capturing the target, is solved. The generalization ability of the monitoring model is improved, and the accurate identification of tunnel water gushing under various tunnel conditions is guaranteed; further, all-round high-precision detection and statistical analysis of tunnel water gushing are realized, and it is determined whether to carry out early warning prompts in the tunnel, thereby predicting the evolution of water gushing, and providing guarantee for safe tunnel construction. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0048] Figure 1 A flow chart of a method in an embodiment of the present invention;

[0049] Figure 2 It is a diagram of equipment layout in an embodiment of the present invention;

[0050] Figure 3 This is a framework diagram of the front-end display system in an embodiment of the present invention;

[0051] Figure 4 This is a framework diagram of a water inrush target detection model in an embodiment of the present invention;

[0052] Figure 5 This is a CA_A content-aware attention module framework diagram in an embodiment of the present invention;

[0053] Figure 6 Graph showing the detection results in an embodiment of the present invention. DETAILED DESCRIPTION

[0054] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0055] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0056] The present invention provides a method for automatically identifying sudden water inrush in a tunnel based on a sudden water inrush target detection model. The method uses a sudden water inrush video frame obtained in the tunnel to construct a sudden water inrush target detection model to obtain the location and amount of sudden water inrush; an early warning prompt is performed in the tunnel to realize real-time monitoring of sudden water in the tunnel; specifically, combined with Figure 4 The method of the present invention comprises the following steps:

[0057] S1, feature extraction through ResNet50 network;

[0058] S2, send the last layer feature F5 into the Transformer encoder module to obtain the new feature S5;

[0059] Q, K, V = Flatten (F5)

[0060] S5=Reshape(Attention(Q,K,V))

[0061] Among them, Q (Query): query vector, which represents the information that the current element needs to obtain from other elements.

[0062] K (Key): Key vector, used to match the query vector to determine which elements are related to the current element.

[0063] V (Value): A numerical vector containing the actual content of the selected elements, which will be weighted and summed according to the degree of match with the query vector.

[0064] Flatten means converting a feature from multidimensional to one-dimensional, and the Reshape operation represents the reshaping of the feature dimension;

[0065] S3, update the new features S5 of the 4th layer and the features of the first three layers. The specific structure is as follows Figure 5 As shown;

[0066] S31. Obtain an exclusive spatial importance map for each single channel of the input feature in a coarse-to-fine manner, fully mixing the channel attention weight and the spatial attention weight:

[0067]

[0068] W out =W s +W c +CS(C 1*1 (Y)

[0069] Among them, CS represents the channel shuffle operation; GAP represents the global average pooling; GMP represents the global maximum pooling; max(0,x) represents the RELU activation function; C k*k W represents the convolution with a kernel size of k*k; c represents the spatial attention weight, W s represents the channel attention weight.

[0070] S32. Based on the spatial attention output value and the channel attention output value, the four-layer feature will be converted into a new four-layer feature after passing through the content perception module:

[0071]

[0072] Among them, CA_A stands for content-aware attention.

[0073] S4, fuse the four layers of features after the content perception module to promote semantic communication between features at different levels, enrich the semantic information at each level, and enable accurate identification of small sudden water gushing points in the tunnel. The specific feature fusion formula is as follows:

[0074]

[0075] f l =f l-1 +f l+1 ,l=7

[0076] f k =f k-4 +DS(f k-1 ),k=8,9

[0077] The lower two layers are then downsampled to keep their dimensions consistent with the high-level features; they are fused after partial convolution PConv operations to obtain the final output feature F:

[0078] F = PConv(f 10 )+PConv(f9)+PConv(DS(f8))+PConv(DS(f7)

[0079] Among them, PConv represents the partial convolution operation, f i represents features, DS represents downsampling operation, and BL represents bilinear interpolation upsampling operation.

[0080] S5. Send the feature fused feature F to the transformer decoder module, and then pass it through the feedforward neural network to finally obtain the target box and category results.

[0081] The present invention also provides a tunnel water inrush automatic identification system based on a water inrush target detection model, Figure 4 , further explained:

[0082] The backbone module is used to extract the features of the video images in the tunnel;

[0083] Content-aware module, which mixes the spatial attention weights and channel attention weights of features and updates features;

[0084] Multi-layer feature fusion module, which fuses the updated features;

[0085] The recognition module decodes the fused features to obtain the target box and category results.

[0086] Combination Figure 1 The method and system of the present invention are further described with specific application examples:

[0087] Equipment layout, such as Figure 2 shown.

[0088] In this embodiment, as a preferred embodiment, an industrial camera with a field of view greater than 10m and a pixel of not less than 12 million is installed at the waist of the tunnel to capture clear videos of sudden water gushing in the tunnel in real time; and a fill light is provided above the industrial camera to increase the light intensity.

[0089] Collect training data sets for water inrush target detection models;

[0090] The video is transmitted to the video acquisition module of the front-end display system.

[0091] Transfer the video in the video acquisition module of the front-end display system to the local server;

[0092] The video frames with obvious water inrush are intercepted, and the wavelet denoising method is used to decompose the image into sub-bands of different frequencies. The sub-bands are thresholded, and the appropriate wavelet basis and decomposition layer number are selected to remove noise, thereby achieving data enhancement of the water inrush video frames.

[0093] The enhanced water surge video frames are annotated using the Labelme annotation tool. The annotations are divided into two categories: one is the water surge point and the other is the water surge. The training set and the test set of the data set are divided into 8:1 ratio.

[0094] Pre-training a water surge target detection model based on a water surge training set, and using the pre-trained water surge target detection model to perform water surge detection;

[0095] The training results and target recognition results are output to the front-end display page for display and warning; the number of sudden water inrush points is counted to determine the current sudden water inrush situation in order to warn whether excavation can continue at this time and ensure the safety of construction workers.

[0096] In this embodiment, the output detection results are as follows: Figure 6 shown.

[0097] Combination Figure 3 As shown, the system of this embodiment is based on the operation and display of a computing device, including:

[0098] The system of the present invention performs registration and login of basic users and administrators through the system registration module;

[0099] Through the video acquisition and storage module in the computing device, the video stream of the tunnel face and side wall is obtained and displayed in real time; and the video frames within a period of time are saved to the local server.

[0100] The sudden water target detection model in the present invention is implemented based on the detection module, uploading real-time video and sudden water detection of the real-time video and counting the number of sudden water targets; and through the setting module, managing the saving path, saving format and saving quality of the target detection results of the detection module.

[0101] The present invention obtains real-time monitoring video by shooting locations such as the tunnel face and arch waist where sudden water may occur, and uses video frames with obvious sudden water to make a training set, and then performs sudden water detection through sudden water target detection; wherein the content-aware attention mechanism is coupled with contextual features to ensure accurate identification of sudden water in tunnels under various tunnel conditions, realize all-round high-precision detection and statistical analysis of sudden water in tunnels, and judge whether to issue early warning prompts in the tunnel, thereby predicting the evolution of sudden water, and providing guarantee for safe tunnel construction.

[0102] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for automatically identifying sudden water inrush in a tunnel based on a sudden water inrush target detection model, characterized in that: include: Construct a water inrush target detection model; The sudden water target detection model includes: Extract video image features in tunnels; Mix the spatial attention weights and channel attention weights of the features and update the features; Fusion of updated features; The fused features are decoded to obtain the target frame and category results.

2. According to claim 1, a method for automatically identifying sudden water inrush in a tunnel based on a sudden water inrush target detection model is characterized in that: The step of extracting features of the video image in the tunnel comprises: Feature extraction is performed through the ResNet50 network to obtain 4 layers of features; The encoder obtains new features of the fourth layer of features among the four layers of features.

3. According to claim 1, a method for automatically identifying sudden water inrush in a tunnel based on a sudden water inrush target detection model is characterized in that: The spatial attention weight and channel attention weight update features of the hybrid feature include: Obtaining new features of the fourth layer of features and channel attention weights of the first layer of features, the second layer of features, and the third layer of features in the four layers of features through convolution, global average pooling, and channel shuffling; Obtain new features of the fourth layer of features and spatial attention weights of the first layer features, the second layer features, and the third layer features in the four layers of features through global average pooling, global maximum pooling, and channel shuffling; Mixing the channel attention weight and the spatial attention weight; Update features via content-aware attention mechanism.

4. According to the method of automatic identification of sudden water in tunnels based on sudden water inrush target detection model according to claim 1, it is characterized in that: The updated features are fused through a multi-layer feature fusion module: in l =f l- 1+f l+1 ,l=7 f k =f k-4 +DS(f k-1 ),k=8,9 F=PConv(f 10 )+PConv(f9)+PConv(DS(f8))+PConv(DS(f7) Among them, PConv represents the partial convolution operation, f i represents features, DS represents downsampling operation, and BL represents bilinear interpolation upsampling operation.

5. According to the method of automatic identification of sudden water inrush in tunnels based on sudden water inrush target detection model of claim 1, it is characterized in that: The decoding of the fused features to obtain the target frame and category results includes: The fused feature F is passed through the transformer decoder module and the feedforward neural network to obtain the target box and category results.

6. The method for automatically identifying sudden water inrush in a tunnel based on a sudden water inrush target detection model according to claim 5, characterized in that: The category results include: sudden water point and sudden water.

7. The method for automatically identifying sudden water inrush in a tunnel based on a sudden water inrush target detection model according to claim 1, characterized in that: Before constructing the water inrush target detection model, the method further includes: constructing a water inrush target detection model training data set.

8. The method for automatically identifying sudden water inrush in a tunnel based on a sudden water inrush target detection model according to claim 7, characterized in that: The constructing of a water inrush target detection model training data set includes: Collect water inrush video frames and perform preprocessing; Annotate and divide the preprocessed video frames as training data sets for water inrush target detection models; The pre-processing comprises: Capture the video frame of sudden water inrush in the tunnel surveillance video; The water inrush video frames are decomposed into sub-bands of different frequencies by wavelet denoising; Threshold processing is performed on sub-bands of different frequencies to remove noise from the water inrush video frames.

9. A tunnel water inrush automatic identification system based on water inrush target detection model, characterized in that: include: The backbone module is used to extract the features of the video images in the tunnel; Content-aware module, which mixes the spatial attention weights and channel attention weights of features and updates features; Multi-layer feature fusion module, which fuses the updated features; The recognition module decodes the fused features to obtain the target box and category results.

Images