Methods, devices, and computer equipment for identifying water seepage through cracks in cable tunnels

Abstract

This application relates to a method, apparatus, and computer device for identifying water seepage in cable tunnel cracks. The method includes: inverting a cable tunnel image to obtain an inverted image; processing the cable tunnel image and the inverted image using a fusion sub-model in an identification model to determine two-phase fusion features; performing data enhancement on the cable tunnel image based on the two-phase fusion features and an enhancement sub-model in the identification model to obtain an enhanced image; inputting the enhanced image and the two-phase fusion features into a water seepage segmentation marker sub-model in the identification model to obtain a tunnel water seepage segmentation marker map output by the water seepage segmentation marker sub-model; and determining the seepage area in the cable tunnel based on the tunnel water seepage segmentation marker map, thereby improving the accuracy of identifying water seepage in cable tunnels.

Description

Technical Field

[0001] This application relates to the field of image recognition technology, and in particular to a method, apparatus and computer equipment for identifying water seepage through cracks in cable tunnels. Background Technology

[0002] As a crucial infrastructure for power transmission, the structural safety of cable tunnels directly impacts the stable operation of power systems. However, due to long-term exposure to environmental factors such as groundwater, soil pressure, and temperature fluctuations, cable tunnels are prone to cracking, leading to water seepage. Water seepage not only accelerates the aging of the tunnel structure but can also cause cable short circuits, equipment damage, and even serious safety accidents. Therefore, timely identification and diagnosis of water seepage problems in cable tunnels are essential for ensuring the safe operation of power systems.

[0003] In traditional technologies, the identification of water seepage through cracks in cable tunnels mainly relies on manual inspections and traditional monitoring methods, including the use of humidity sensors, temperature sensors, and other equipment.

[0004] However, traditional methods for identifying water seepage through cracks in cable tunnels suffer from inaccurate identification. Summary of the Invention

[0005] Therefore, it is necessary to provide a method, device, and computer equipment for identifying cable tunnel cracks and seepage that can improve the accuracy of identification, addressing the aforementioned technical problems.

[0006] Firstly, this application provides a method for identifying water seepage through cracks in cable tunnels, including:

[0007] The cable tunnel image is inverted to obtain an inverted image;

[0008] The cable tunnel image and the inverted image are processed by identifying the fusion sub-model in the model to determine the two-phase fusion features;

[0009] Based on the dual-phase fusion features and the enhancement sub-model in the recognition model, the cable tunnel image is augmented to obtain an enhanced image.

[0010] The enhanced image and the dual-phase fusion features are input into the leakage segmentation marker sub-model in the recognition model to obtain the tunnel leakage segmentation marker map output by the leakage segmentation marker model;

[0011] The seepage areas in cable tunnels are determined based on the tunnel seepage segmentation marking map.

[0012] In one embodiment, the leakage segmentation marker sub-model includes an adaptive module and a fragment arbitrary module. The step of inputting the enhanced image and biphasic fusion features into the leakage segmentation marker sub-model in the recognition model to obtain the tunnel leakage segmentation marker map output by the leakage segmentation marker model includes:

[0013] The pixel data of the enhanced image is separated based on the pixel color to obtain the three-channel features corresponding to the enhanced image;

[0014] The three-channel features are input into the adaptive module for processing to obtain the first processed data;

[0015] By stacking the first processed data, the three-channel features, and the dual-phase fusion features using the arbitrary module of the segment, a tunnel seepage segmentation marker map output by the seepage segmentation marker sub-model is obtained.

[0016] In one embodiment, the step of inputting the three-channel features into the adaptive module for processing to obtain first processed data includes:

[0017] The three-channel features are input into the convolutional layer of the adaptive module for feature extraction to obtain the single-channel features corresponding to the three-channel features; the single-channel features include red channel data, green channel data, and blue channel data;

[0018] Max pooling is performed on the single-channel features to enhance the texture features associated with the detection of the leaking area edges;

[0019] The enhanced texture features and the downsampled enhanced image are superimposed to obtain the first processed data.

[0020] In one embodiment, the step of processing the cable tunnel image and the inverted image through a fusion sub-model in the identification model to determine the two-phase fusion features includes:

[0021] The cable tunnel image and the inverted image are input into the fusion sub-model; the fusion sub-model includes a convolution module and a fusion module.

[0022] The cable tunnel image is processed by multiple convolution modules to obtain a first feature map.

[0023] The inverted image is processed by multiple convolutions through the convolution module to obtain the second feature map;

[0024] The first feature map and the second feature map are fused to obtain a biphasic fused feature.

[0025] In one embodiment, the convolution module includes multiple convolutional layers, multiple activation function layers, and an attention mechanism layer, wherein the multiple activation function layers are disposed between some of the convolutional layers in the multiple convolutional layers, and the attention mechanism layer is disposed after the convolutional layers.

[0026] The multi-layer convolutional layer includes multiple first convolutional layers, multiple second convolutional layers, and multiple third convolutional layers. The third convolutional layer corresponds one-to-one with the first convolutional layer, and the third convolutional layer is a skip convolutional layer of the corresponding first convolutional layer.

[0027] In one embodiment, the training process of the recognition model includes:

[0028] Construct an initial identification model; the initial identification model includes an initial fusion sub-model, an initial enhancement sub-model, and an initial leakage segmentation marker sub-model;

[0029] A no-reference loss function and a detection performance evaluation index are constructed; the detection performance evaluation index is used to evaluate the detection performance of overlap and similarity between the ground truth label and the predicted mask;

[0030] Obtain a dark light dataset of cable tunnel leakage water, and train the initial recognition model based on the cable tunnel leakage water dark light dataset, the no-reference loss function, and the detection effect evaluation index to obtain the recognition model.

[0031] In one embodiment, the no-reference loss function is determined based on at least one of the spatial consistency loss function, exposure control loss function, color consistency loss function, total variation loss function, and channel consistency loss function.

[0032] Secondly, this application also provides a cable tunnel crack seepage identification device, comprising:

[0033] The phase inversion module is used to invert cable tunnel images to obtain inverted images;

[0034] The fusion module is used to process the cable tunnel image and the inverted image by identifying the fusion sub-model in the recognition model to determine the two-phase fusion features;

[0035] An enhancement module is used to perform data enhancement on the cable tunnel image based on the dual-phase fusion features and the enhancement sub-model in the recognition model to obtain an enhanced image;

[0036] The acquisition module is used to input the enhanced image and the dual-phase fusion features into the leakage segmentation marker sub-model in the recognition model, so as to obtain the tunnel leakage segmentation marker map output by the leakage segmentation marker model;

[0037] The determination module is used to identify seepage areas in cable tunnels based on tunnel seepage segmentation marking maps.

[0038] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0039] The cable tunnel image is inverted to obtain an inverted image;

[0040] The cable tunnel image and the inverted image are processed by identifying the fusion sub-model in the model to determine the two-phase fusion features;

[0041] Based on the dual-phase fusion features and the enhancement sub-model in the recognition model, the cable tunnel image is augmented to obtain an enhanced image.

[0042] The enhanced image and the dual-phase fusion features are input into the leakage segmentation marker sub-model in the recognition model to obtain the tunnel leakage segmentation marker map output by the leakage segmentation marker model;

[0043] The seepage areas in cable tunnels are determined based on the tunnel seepage segmentation marking map.

[0044] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:

[0045] The cable tunnel image is inverted to obtain an inverted image;

[0046] The cable tunnel image and the inverted image are processed by identifying the fusion sub-model in the model to determine the two-phase fusion features;

[0047] Based on the dual-phase fusion features and the enhancement sub-model in the recognition model, the cable tunnel image is augmented to obtain an enhanced image.

[0048] The enhanced image and the dual-phase fusion features are input into the leakage segmentation marker sub-model in the recognition model to obtain the tunnel leakage segmentation marker map output by the leakage segmentation marker model;

[0049] The seepage areas in cable tunnels are determined based on the tunnel seepage segmentation marking map.

[0050] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps:

[0051] The cable tunnel image is inverted to obtain an inverted image;

[0052] The cable tunnel image and the inverted image are processed by identifying the fusion sub-model in the model to determine the two-phase fusion features;

[0053] Based on the dual-phase fusion features and the enhancement sub-model in the recognition model, the cable tunnel image is augmented to obtain an enhanced image.

[0054] The enhanced image and the dual-phase fusion features are input into the leakage segmentation marker sub-model in the recognition model to obtain the tunnel leakage segmentation marker map output by the leakage segmentation marker model;

[0055] The seepage areas in cable tunnels are determined based on the tunnel seepage segmentation marking map.

[0056] The aforementioned method, apparatus, and computer equipment for identifying cable tunnel cracks and seepage involve inverting a cable tunnel image to obtain an inverted image; processing the cable tunnel image and the inverted image using a fusion sub-model in the identification model to determine the two-phase fusion features; performing data enhancement on the cable tunnel image based on the two-phase fusion features and the enhancement sub-model in the identification model to obtain an enhanced image; inputting the enhanced image and the two-phase fusion features into the leakage segmentation marker sub-model in the identification model to obtain a tunnel leakage segmentation marker map output by the leakage segmentation marker sub-model; and determining the seepage area in the cable tunnel based on the tunnel leakage segmentation marker map. By obtaining the corresponding inverted image of the cable tunnel image and utilizing the two-phase fusion features of the cable tunnel image and the inverted image to perform data enhancement on the cable tunnel image, the accuracy of enhanced data identification is improved. Furthermore, using the enhanced data and the two-phase fusion features as input to the leakage segmentation marker sub-model makes the input data more reliable and accurate, thus improving the accuracy of identifying cable tunnel leakage. Attached Figure Description

[0057] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0058] Figure 1 This is a diagram illustrating the application environment of a cable tunnel crack seepage identification method in one embodiment.

[0059] Figure 2 This is one of the flowcharts illustrating a method for identifying water seepage through cracks in a cable tunnel in one embodiment;

[0060] Figure 3This is a second schematic flowchart of a method for identifying water seepage through cracks in a cable tunnel in one embodiment;

[0061] Figure 4 This is a schematic diagram of any module of a segment in one embodiment;

[0062] Figure 5 This is the third flowchart of a method for identifying water seepage through cracks in a cable tunnel in one embodiment;

[0063] Figure 6 This is a schematic diagram of an adaptive module in one embodiment;

[0064] Figure 7 This is the fourth flowchart of a method for identifying water seepage through cracks in a cable tunnel in one embodiment;

[0065] Figure 8 This is a schematic diagram of a convolution module in one embodiment;

[0066] Figure 9 This is the fifth flowchart illustrating a method for identifying water seepage through cracks in a cable tunnel in one embodiment;

[0067] Figure 10 This is a schematic diagram of the framework of the identification model in one embodiment;

[0068] Figure 11 This is a structural block diagram of a cable tunnel crack seepage identification device in one embodiment. Detailed Implementation

[0069] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0070] As a crucial infrastructure for power transmission, the structural safety of cable tunnels directly impacts the stable operation of power systems. However, due to long-term exposure to environmental factors such as groundwater, soil pressure, and temperature fluctuations, cable tunnels are prone to cracking, leading to water seepage. Water seepage not only accelerates the aging of the tunnel structure but can also cause cable short circuits, equipment damage, and even serious safety accidents. Therefore, timely identification and diagnosis of water seepage problems in cable tunnels are essential for ensuring the safe operation of power systems.

[0071] In traditional technologies, the identification and diagnosis of water seepage through cracks in cable tunnels mainly relies on manual inspections and conventional monitoring methods. Manual inspections typically involve professionals periodically entering the tunnel to visually inspect and listen for cracks and seepage by tapping and tapping. While intuitive, this method suffers from low efficiency, limited coverage, and strong subjectivity, and poses a safety threat to inspection personnel in high-risk environments. Traditional monitoring methods include using humidity and temperature sensors to indirectly determine the presence of seepage by monitoring changes in humidity and temperature inside the tunnel. However, these methods often provide only limited information, cannot accurately locate cracks, and are susceptible to environmental interference, leading to false alarms or missed detections. Furthermore, the installation and maintenance costs of traditional monitoring equipment are high, making it difficult to promote and apply them in large-scale tunnel networks.

[0072] With the development of technology, some advanced detection methods have been gradually introduced, such as infrared thermal imaging and ultrasonic testing. Infrared thermal imaging can indirectly identify seepage areas by capturing the temperature distribution on the tunnel surface, but its resolution is limited and it is difficult to detect tiny cracks. Ultrasonic testing utilizes the propagation characteristics of sound waves in materials to detect the presence of cracks, but its operation is complex and requires a high level of professional skill from the inspectors.

[0073] The identification and diagnosis of water seepage through cracks in cable tunnels mainly relies on manual inspections and traditional monitoring methods, which suffer from low efficiency, poor accuracy, and high cost. Therefore, there is an urgent need for an efficient, accurate, and low-cost method for identifying and diagnosing water seepage through cracks in cable tunnels to meet the growing maintenance demands of power infrastructure.

[0074] The cable tunnel crack seepage identification method provided in this application embodiment can be applied to, for example... Figure 1 The application environment shown. The computer device can be a server, and its internal structure diagram can be as follows. Figure 1As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores data on cable tunnel crack seepage identification. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for identifying cable tunnel crack seepage.

[0075] Those skilled in the art will understand that Figure 1 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0076] In one embodiment, such as Figure 2 As shown, a method for identifying water seepage through cracks in cable tunnels is provided, which can be applied to... Figure 1 Taking the server in the example of this, the explanation includes:

[0077] S201, Invert the cable tunnel image to obtain an inverted image.

[0078] In this embodiment of the application, the cable tunnel image is a dark image. The inversion processing is to invert the colors of the cable tunnel image, convert the dark areas (low pixel values) in the cable tunnel image into bright areas (high pixel values), and change the value of each pixel channel in the cable tunnel image from the current value I to 255-I (for an 8-bit image). For example, the formula for inverting the cable tunnel image I to obtain the inverted image I' can be: I'=255-I.

[0079] S202, by identifying the fusion sub-model in the model, the cable tunnel image and the inverted image are processed to determine the two-phase fusion characteristics.

[0080] In this embodiment of the application, the identification model includes a fusion sub-model, an enhancement sub-model, and a marker sub-model. The cable tunnel image and the inverted image are input into the fusion sub-model for feature extraction to obtain a first feature of the cable tunnel image and a second feature of the inverted image. Furthermore, the first feature and the second feature are fused to obtain a two-phase fusion feature.

[0081] For example, the fusion sub-model can merge the first feature and the second feature to obtain a biphasic fusion feature, wherein the merging operation can include any of the following:

[0082] (1) Concatenation: directly concatenate two feature vectors or feature maps along the channel dimension;

[0083] (2) Element-wise operations: addition, multiplication, averaging, etc.;

[0084] (3) Weighted fusion: The importance weights of different channels or spatial locations are learned using the attention mechanism, and then weighted fusion is performed;

[0085] (4) Learning-based fusion: Using small neural networks to learn how to optimally combine two feature streams.

[0086] Optionally, the dual-phase fusion feature can be used as a curve parameter mapping for the adaptive light enhancement curve. ,in Represents pixel coordinates.

[0087] S203, based on the dual-phase fusion features and the enhancement sub-model in the recognition model, performs data enhancement on the cable tunnel image to obtain an enhanced image.

[0088] In this embodiment, the two-phase fusion features and the cable tunnel image are input into the enhancement sub-model to perform data enhancement on the cable tunnel image to obtain an enhanced image. For example, the cable tunnel image can be input into an amplitude iteration controller, based on... The brightness is enhanced by the illumination. and number of iterations Furthermore, the adaptive light enhancement curve (ALE) is iteratively adjusted to obtain an enhanced image.

[0089] Optionally, select low-light image datasets such as SID and DPED, extract a certain number of images from each dataset, record the enhancement and suppression amplitude values ​​of the images at different exposure levels, and select the best-fit amplitude curve. This is used to calculate the optimal enhancement or suppression amplitude required for the input image at different exposure levels. The amplitude curve is represented as follows: ,in This represents the normalized average pixel value of the image. Further, this can be achieved through iterative curve calculations. The iterative scheme required to fit images with different exposures, and then through... Determine the number of iterations. Iteration curve It can be shown in Equation 1:

[0090] (Equation 1)

[0091] Where n is the number of iterations. This is for rounding operations.

[0092] Furthermore, the curve parameters are mapped. Amplitude intensity Number of iterations n, brightness of input graph The adaptive light enhancement (ALE) curve of the design is incorporated so that the illumination level of the image is iteratively enhanced or suppressed to a specified interval. After multiple iterations, a dark-light enhanced image is obtained. The ALE curve is shown in Equation 2.

[0093] (Equation 2)

[0094] in, for The enhanced result For the initial input image , =0.6 is the appropriate exposure level for the target image.

[0095] In this embodiment, by simultaneously considering the input image and its inverted version, the model can learn more comprehensive illumination information, thereby significantly improving its processing capability for images at different exposure levels. The method employs an adaptive light enhancement curve and an amplitude iteration controller, which can precisely control the intensity of image enhancement or suppression, effectively avoiding noise amplification and achieving high-quality enhancement of low-light images. This not only improves the model's versatility but also enhances its application advantages on real-world devices, providing reliable technical support for image processing in low-light environments.

[0096] S204, the enhanced image and biphasic fusion features are input into the leakage segmentation marker sub-model in the recognition model to obtain the tunnel leakage segmentation marker map output by the leakage segmentation marker sub-model.

[0097] In this embodiment, the cable tunnel image after image enhancement processing and its corresponding dual-phase fusion features are synchronously input into the leakage segmentation marker sub-model in the recognition model. The leakage segmentation marker sub-model can jointly analyze the two types of input data through a convolutional neural network architecture and output a tunnel leakage segmentation marker map with pixel-level prediction accuracy.

[0098] Optionally, the tunnel seepage segmentation marker map can be a binary matrix or a probability heatmap. When the tunnel seepage segmentation marker map is a binary matrix, a specific pixel value represents the spatial location where seepage occurs; when the tunnel seepage segmentation marker map is a probability heatmap, a high-probability region represents the spatial location where seepage occurs.

[0099] S205, Determine the seepage area in the cable tunnel based on the tunnel seepage segmentation marking map.

[0100] In this embodiment, based on the spatial coordinate mapping relationship of the tunnel seepage segmentation marker map, the activated pixel regions (or connected regions with a probability exceeding a preset threshold) in the marker map are mapped to the corresponding positions in the original cable tunnel image, accurately delineating the geometric boundaries of the seepage area, and finally outputting a cable tunnel structure map with spatial annotations of the seepage area and quantitative data of the seepage range. For example, the quantitative data of the seepage range includes area, location coordinates, etc.

[0101] In the above-described embodiments, the cable tunnel image is inverted to obtain an inverted image; the cable tunnel image and the inverted image are processed by a fusion sub-model in the recognition model to determine the two-phase fusion features; based on the two-phase fusion features and the enhancement sub-model in the recognition model, the cable tunnel image is augmented to obtain an enhanced image; the enhanced image and the two-phase fusion features are input into the leakage segmentation marker sub-model in the recognition model to obtain the tunnel leakage segmentation marker map output by the leakage segmentation marker sub-model; the leakage area in the cable tunnel is determined based on the tunnel leakage segmentation marker map. Obtaining the inverted image corresponding to the cable tunnel image, and thus using the two-phase fusion features of the cable tunnel image and the inverted image to perform data augmentation on the cable tunnel image, improves the accuracy of enhanced data recognition. Furthermore, using the enhanced data and the two-phase fusion features as input to the leakage segmentation marker sub-model makes the input data more reliable and accurate, thus improving the accuracy of identifying leakage in cable tunnels.

[0102] In one embodiment, one implementation of the above-described S203 is provided, such as... Figure 3 As shown, the leakage segmentation marker sub-model includes an adaptive module and a segment arbitrary module. The aforementioned "inputting the enhanced image and biphasic fusion features into the leakage segmentation marker sub-model in the recognition model to obtain the tunnel leakage segmentation marker map output by the leakage segmentation marker model" includes:

[0103] S301, the pixel data of the enhanced image is separated according to the pixel color to obtain the three-channel features corresponding to the enhanced image.

[0104] In this embodiment, color component analysis can be performed on each pixel of the enhanced image based on the RGB color space model or a preset color separation algorithm. Optionally, a pixel-level color channel decoupling algorithm can be used to decompose the composite color value of each pixel into three independent spectral components: red (R), green (G), and blue (B). Further, two-dimensional feature matrices with the same size as the original image are constructed respectively.

[0105] (1) Red channel feature matrix: stores the R component values ​​of all pixels;

[0106] (2) Green channel feature matrix: stores the G component values ​​of all pixels;

[0107] (3) Blue channel feature matrix: stores the B component values ​​of all pixels.

[0108] Furthermore, the output constitutes a three-channel separation feature set of the enhanced image.

[0109] S302, the three-channel features are input to the adaptive module for processing to obtain the first processed data.

[0110] In this embodiment, the adaptive module can be a trained deep learning network model. The three-channel features are input into the adaptive module for processing to enhance the texture features necessary for detecting the edge of the leaking area, and the first processing data is determined based on the enhanced texture features.

[0111] As an optional implementation, the three-channel separated feature set is input into the adaptive module, and the first processed data is generated through the following processing flow:

[0112] (1) Channel importance assessment: A learnable weight generator is used to analyze the statistical characteristics of each channel feature and dynamically generate channel weight vectors;

[0113] (2) Weighted feature fusion: The three-channel features are linearly fused according to the weight vector to generate an enhanced feature matrix;

[0114] (3) Spatial adaptive enhancement: Multi-scale contextual information is extracted using dilated convolutional layers, and the first processed data is output by combining spatial attention mechanism.

[0115] S303, by stacking the first processed data, three-channel features and two-phase fusion features through arbitrary segment modules, the tunnel seepage segmentation marker map output by the water leakage segmentation marker sub-model is obtained.

[0116] In this embodiment, the first processed data output by the adaptive module is used as input, and the position information in the feature sequence is marked by the position encoding module PE. Here, a biphasic fusion feature output is introduced for multi-layer stacking of original features, such as... Figure 4As shown, in the leakage segmentation and labeling sub-model, adaptive modules are inserted before and after each layer of the self-attention feedforward Transformer block in the Spatial Attention Multi-scale Aggregation (SAMA) module. The input of each adaptive module is the parameter adjustment output of the previous adaptive module. Multiple adaptive modules are used to effectively adjust the network parameters. The self-attention feedforward Transformer block consists of a multi-head attention layer (MHA) and a multilayer perceptron (MLP) layer with added normalization. A pair of MLP layers is added between each Transformer block as an adaptive module, as shown in Equation 3:

[0117] (Equation 3)

[0118] in Through adaptive modules The obtained prompt, Pi, is attached to each Transformer block of the SAM encoder, with GELU as the activation function. Embedded features for patches High-frequency component characteristics The sum can be expressed as: .

[0119] In this embodiment, the first processed data, three-channel features, and biphasic fusion features are input into any trained segment module for stacking processing to obtain processed feature data, and the processed feature data is converted into a tunnel seepage water segmentation marker map.

[0120] In the above-mentioned application embodiments, the three-channel features are first processed by an adaptive module, and then the first processed data is used as the input of any segment module and stacked with the three-channel features and the biphasic fusion features, thereby improving the accuracy of the tunnel seepage water segmentation and marking map.

[0121] In one embodiment, one implementation of the above-described S302 is provided, such as... Figure 5 As shown, the above-mentioned "inputting the three-channel features into the adaptive module for processing to obtain the first processed data" includes:

[0122] S401, the three-channel features are input into the convolutional layer of the adaptive module for feature extraction, and the single-channel features corresponding to the three-channel features are obtained; the single-channel features include red channel data, green channel data and blue channel data.

[0123] S402 performs max pooling on single-channel features to enhance texture features associated with leak area edge detection.

[0124] S403, the enhanced texture features and the downsampled enhanced image are superimposed to obtain the first processed data.

[0125] In this embodiment of the application, the output image will be enhanced. An adaptive input module is implemented for R, G, and B channel segmentation. For example... Figure 6 As shown, the adaptive module first extracts single-channel features through a 1×1 convolutional layer, and then enhances the texture features necessary for detecting the edges of the leaking area through a max-pooling layer. The enhanced texture features are then repeatedly superimposed with the downsampled image, channel by channel and element by element, to preserve the texture features extracted from the convolutional layer across channels, serving as the input features for the segment-arbitrary model adaptive module. The design of the adaptive module can be represented as: .in, It is a low-light enhanced output detection image. This is for downsampling. It is a convolutional layer with a kernel size of 1×1. It is a max pooling layer with a kernel size of 2×2.

[0126] In the above-mentioned embodiments, the introduction of an adaptive module within the model enables any fragment model to adapt to a specific inspection task without retraining or fine-tuning the entire model, effectively reducing training time and memory consumption.

[0127] In one embodiment, one implementation of S202 above is provided, such as... Figure 7 As shown, the above-mentioned "processing cable tunnel images and inverted images by identifying the fusion sub-model in the model to determine the two-phase fusion features" includes:

[0128] S501, input the cable tunnel image and the inverted image into the fusion sub-model; the fusion sub-model includes a convolution module and a fusion module.

[0129] In this embodiment, cable tunnel images and inverted images are used as inputs to the fusion sub-model, thereby utilizing the convolution and fusion modules in the fusion sub-model to process the input data.

[0130] Optionally, the convolutional module includes multiple convolutional layers, multiple activation function layers, and an attention mechanism layer. The multiple activation function layers are placed between some of the convolutional layers in the multiple convolutional layers, and the attention mechanism layer is placed after the convolutional layers. Among them, the multiple convolutional layers include multiple first convolutional layers, multiple second convolutional layers, and multiple third convolutional layers. The first convolutional layer is an input convolutional layer with skip connections, the second convolutional layer is a convolutional layer that does not participate in skip connections, and the third convolutional layer is a convolutional layer formed by skip connections. The third convolutional layer corresponds one-to-one with the first convolutional layer, and the third convolutional layer is a convolutional layer formed by skip connections of the corresponding first convolutional layer.

[0131] S502, the cable tunnel image is processed by a convolution module to obtain the first feature map.

[0132] In this embodiment of the application, a first feature map is obtained by performing a 9-layer side-by-side ordered convolution on the cable tunnel image. For example... Figure 8 As shown, the convolutional module consists of 16 3×3 convolutional kernels with a stride of 1. ReLU activation functions are added after layers 2, 3, 5, and 7. A coordinated attention mechanism (CA) is added after the activation function in layer 8 to compress global spatial information into two dimensions and embed positional information into the spatial information, including inter-channel mapping relationships and enhancing the network's expressive power. Layers 5, 7, and 9 are Skip-Connection convolutional layers (layers 2, 3, and 1), respectively, preserving and propagating low-level features such as details and textures from the original input image. This allows low-level features to be directly passed to later classification stages, enabling low-level feature reuse, increasing feature diversity, providing an unobstructed path for gradients, effectively alleviating the gradient vanishing problem, and accelerating the training process. A Ghost module and Tanh activation function are added after the last layer of the convolutional module. The Ghost module first uses a 1×1 convolution to compress the input feature map, achieving cross-channel feature extraction. After obtaining the compressed features, a 3×3 convolutional kernel is used for layer-by-layer convolution to obtain an additional feature map. Finally, the 1×1 convolution result is superimposed with the layer-by-layer convolution result to obtain the final feature map, which greatly improves the utilization efficiency of the feature map, reduces the amount of computation, and achieves richer feature fusion.

[0133] Optionally, multiple first convolutional layers are Figure 8 The 2nd, 3rd, and 1st layers, and the multi-layer third convolutional layer are... Figure 8 The 5th, 7th, and 9th layers correspond one-to-one with the 2nd, 3rd, and 1st layers, respectively, and the remaining convolutional layers are multiple second convolutional layers.

[0134] S503 performs multi-layer convolution processing on the inverted image through a convolution module to obtain the second feature map.

[0135] In this embodiment of the application, a second feature map is obtained by performing nine layers of side-by-side ordered convolution on the inverted input image.

[0136] S504, the first feature map and the second feature map are fused to obtain the biphasic fused feature.

[0137] In this embodiment, the three-channel (R, G, B) curve parameters of the two outputs are merged to obtain the curve parameter mapping of the adaptive light enhancement curve for each iteration. ,in Represents pixel coordinates.

[0138] In the above application embodiments, the first feature map corresponding to the cable tunnel image and the second feature map corresponding to the inverted image are fused to obtain a two-phase fused feature, making the data more comprehensive and reliable.

[0139] In one embodiment, such as Figure 9 As shown, the training process of the above recognition model includes:

[0140] S601, Construct the initial identification model; the initial identification model includes the initial fusion sub-model, the initial enhancement sub-model, and the initial leakage segmentation marker sub-model.

[0141] Optionally, the overall network structure in this application embodiment is as follows: Figure 10 As shown, the fusion sub-model and the enhancement sub-model can be Zero-Reference Deep Curve Estimation (Zero-DDCE). The fusion sub-model is a double-phase of Deep Curve EstimationNet (DDCE-Net), and the water leakage segmentation labeling sub-model is a Spatial Attention Multi-scale Aggregation (SAMA) module.

[0142] S602, construct the no-reference loss function and the detection performance evaluation index; the detection performance evaluation index is used to evaluate the detection performance of overlap and similarity between the ground truth label and the predicted mask.

[0143] In the embodiments of this application, a no-reference loss function can be designed with the goals of spatial consistency, color constancy, and illumination smoothness, and an evaluation index for detection effect can be constructed based on historical data.

[0144] Optionally, in this embodiment, the no-reference loss function is determined based on at least one of the following: spatial consistency loss function, exposure control loss function, color consistency loss function, total variation loss function, and channel consistency loss function. Wherein:

[0145] (1) Spatial consistency loss function This can promote and enhance the spatial consistency of images, as shown in Equation 4:

[0146] (Equation 4)

[0147] Where K is the number of local regions. The four adjacent regions centered on region i are: top, bottom, left, and right. A set of. and To enhance the average pixel values ​​of local regions in the image compared to the original image, the local region size is set to 4×4.

[0148] (2) Exposure control loss function The exposure level can be controlled, as shown in Equation 5:

[0149] (Equation 5)

[0150] Where M is the number of non-overlapping local regions of size 16×16, Y is the average pixel value of the enhanced local region, and E is the median image brightness of 0.6.

[0151] (3) Color consistency loss function This can reduce color deviation in enhanced images, as shown in Equation 6:

[0152] (Equation 6)

[0153] in, This represents the average pixel value of the p channel in the enhanced image. The values ​​p and q represent the pixel average of the q channel in the enhanced image, and the variables p and q iterate through all possible combinations of the enhanced color channels. , This represents a pair of channels.

[0154] (4) Total variation loss function ,Can As shown in Equation 7:

[0155] (Equation 7)

[0156] Where C, H, and W represent the number of image channels, height, and width, respectively. ∇x and ∇y represent the horizontal and vertical gradient operations, respectively. To enhance the channels in the image High value Width value The value of the pixel.

[0157] (5) Channel consistency loss function KL divergence can enhance the consistency between the original and enhanced images in terms of channel pixel difference, suppress the generation of noise information and invalid features, and improve the image enhancement effect, as shown in Equation 8:

[0158] (Equation 8)

[0159] Where R, G, and B represent the color channels of the original image. , , This represents the three color channels of the enhanced image. KL divergence represents the difference between two distributions. If the difference between the two distributions is small, the KL divergence is small. When the two distributions are identical, the KL divergence value is 0.

[0160] Furthermore, the total loss function It can be shown in Equation 9:

[0161] (Equation 9)

[0162] in, , , The weights for spatial consistency loss, color consistency loss, total variation loss, and channel consistency loss are respectively determined based on the emphasis placed on the actual detection effect.

[0163] Optionally, in this embodiment of the application, the average intersection / union is used. Average dice coefficient It serves as an evaluation metric for measuring the overlap and similarity between the predicted and the actual mask in detection performance. This can be expressed as Equation 10. This can be expressed as Equation 11:

[0164] (Equation 10)

[0165] (Equation 11)

[0166] in, Indicates the number of classes during the partitioning process. and These represent the intersection and union of the actual ground mask and the predicted water seepage mask, respectively.

[0167] S603. Obtain the dark light dataset of cable tunnel leakage water, and train the initial recognition model based on the cable tunnel leakage water dark light dataset, the no-reference loss function, and the detection effect evaluation index to obtain the recognition model.

[0168] In this embodiment of the application, a dark light dataset of cable tunnel leakage water is collected, and the data is analyzed based on the cable tunnel leakage water dark light dataset, a no-reference loss function, and detection performance evaluation metrics. Figure 10The Zero-DDCE-SAMA (Zero-Reference Double-phase of Deep Curve Estimation Segment Anything ModelAdapetr) initial recognition model is trained until the loss of the initial recognition model reaches the preset loss and meets the detection effect evaluation index, so as to realize the identification of cable tunnel leakage under low light conditions.

[0169] In the above-mentioned embodiments, by constructing a loss function with multiple dimensions, the recognition accuracy of the trained recognition model is improved, and the recognition model is evaluated according to the detection effect evaluation index, which further improves the recognition accuracy.

[0170] In one embodiment, a complete method for identifying water seepage through cracks in cable tunnels is provided, including:

[0171] S1, Construct the initial identification model; the initial identification model includes the initial fusion sub-model, the initial enhancement sub-model, and the initial leakage segmentation marker sub-model.

[0172] S2, construct the no-reference loss function and the detection performance evaluation index; the detection performance evaluation index is used to evaluate the detection performance of overlap and similarity between the ground truth label and the predicted mask.

[0173] S3. Obtain the dark light dataset of cable tunnel leakage water, and train the initial recognition model based on the cable tunnel leakage water dark light dataset, the no-reference loss function, and the detection effect evaluation index to obtain the recognition model.

[0174] S4. Invert the cable tunnel image to obtain an inverted image.

[0175] S5 inputs the cable tunnel image and the inverted image into the fusion sub-model; the fusion sub-model includes a convolution module and a fusion module.

[0176] S6, the cable tunnel image is processed by multiple convolution modules to obtain the first feature map.

[0177] S7 performs multi-layer convolution processing on the inverted image through the convolution module to obtain the second feature map.

[0178] S8, fuse the first feature map and the second feature map to obtain the biphasic fused feature.

[0179] S9, based on the dual-phase fusion features and the enhancement sub-model in the recognition model, performs data enhancement on the cable tunnel image to obtain an enhanced image.

[0180] S10: Separate the pixel data of the enhanced image according to the pixel color to obtain the three-channel features corresponding to the enhanced image.

[0181] S11, the three-channel features are input into the convolutional layer of the adaptive module for feature extraction, and the single-channel features corresponding to the three-channel features are obtained; the single-channel features include red channel data, green channel data and blue channel data.

[0182] S12 performs max pooling on the single-channel features to enhance texture features associated with the detection of leaking area edges.

[0183] S13, the enhanced texture features and the downsampled enhanced image are superimposed to obtain the first processed data.

[0184] S14, the first processed data, three-channel features and two-phase fusion features are stacked through the arbitrary module of the segment to obtain the tunnel seepage segmentation marker map output by the water leakage segmentation marker sub-model.

[0185] S15, Determine the seepage area in the cable tunnel based on the tunnel seepage segmentation marking map.

[0186] In the above-described embodiments, the cable tunnel image is inverted to obtain an inverted image; the cable tunnel image and the inverted image are processed by a fusion sub-model in the recognition model to determine the two-phase fusion features; based on the two-phase fusion features and the enhancement sub-model in the recognition model, the cable tunnel image is augmented to obtain an enhanced image; the enhanced image and the two-phase fusion features are input into the leakage segmentation marker sub-model in the recognition model to obtain the tunnel leakage segmentation marker map output by the leakage segmentation marker sub-model; the leakage area in the cable tunnel is determined based on the tunnel leakage segmentation marker map. Obtaining the inverted image corresponding to the cable tunnel image, and thus using the two-phase fusion features of the cable tunnel image and the inverted image to perform data augmentation on the cable tunnel image, improves the accuracy of enhanced data recognition. Furthermore, using the enhanced data and the two-phase fusion features as input to the leakage segmentation marker sub-model makes the input data more reliable and accurate, thus improving the accuracy of identifying leakage in cable tunnels.

[0187] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0188] Based on the same inventive concept, this application also provides a cable tunnel crack seepage identification device for implementing the above-described cable tunnel crack seepage identification method. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations of one or more cable tunnel crack seepage identification device embodiments provided below can be found in the limitations of the cable tunnel crack seepage identification method described above, and will not be repeated here.

[0189] In one embodiment, such as Figure 11 As shown, a cable tunnel crack seepage identification device is provided, comprising: an inversion module 10, a fusion module 11, an enhancement module 12, an acquisition module 13, and a determination module 14, wherein:

[0190] The inversion module 10 is used to invert the cable tunnel image to obtain an inverted image;

[0191] The fusion module 11 is used to process the cable tunnel image and the inverted image by identifying the fusion sub-model in the model to determine the two-phase fusion features;

[0192] Enhancement module 12 is used to perform data enhancement on cable tunnel images based on dual-phase fusion features and enhancement sub-models in the recognition model to obtain enhanced images;

[0193] The acquisition module 13 is used to input the enhanced image and the dual-phase fusion features into the leakage segmentation marker sub-model in the recognition model, so as to obtain the tunnel leakage segmentation marker map output by the leakage segmentation marker sub-model;

[0194] Module 14 is used to determine the seepage area in the cable tunnel based on the tunnel seepage segmentation marking map.

[0195] In one embodiment, the acquisition module 13 includes: a separation unit, a processing unit, and a stacking unit, wherein:

[0196] The separation unit is used to separate the pixel data of the enhanced image according to the pixel color to obtain the three-channel features corresponding to the enhanced image.

[0197] The processing unit is used to input the three-channel features into the adaptive module for processing to obtain the first processed data.

[0198] The stacking unit is used to stack the first processed data, three-channel features and two-phase fusion features through arbitrary segment modules to obtain the tunnel seepage segmentation marker map output by the water leakage segmentation marker sub-model.

[0199] In one embodiment, the above processing unit is specifically used to input the three-channel features into the convolutional layer of the adaptive module for feature extraction to obtain single-channel features corresponding to the three-channel features; the single-channel features include red channel data, green channel data and blue channel data; perform max pooling operation on the single-channel features to enhance the texture features related to the edge detection of the leaking area; and superimpose the enhanced texture features and the downsampled enhanced image to obtain the first processed data.

[0200] In one embodiment, the fusion module 11 includes: an input unit, a first convolutional unit, a second convolutional unit, and a fusion unit, wherein:

[0201] The input unit is used to input cable tunnel images and inverted images into the fusion sub-model; the fusion sub-model includes a convolution module and a fusion module.

[0202] The first convolutional unit is used to perform multi-layer convolution processing on the cable tunnel image through the convolution module to obtain the first feature map.

[0203] The second convolutional unit is used to perform multi-layer convolution processing on the inverted image through the convolution module to obtain the second feature map;

[0204] The fusion unit is used to fuse the first feature map and the second feature map to obtain a biphasic fused feature.

[0205] In one embodiment, the convolutional module includes multiple convolutional layers, multiple activation function layers, and an attention mechanism layer. The multiple activation function layers are disposed between some of the convolutional layers in the multiple convolutional layers, and the attention mechanism layer is disposed after the convolutional layers. The multiple convolutional layers include multiple first convolutional layers, multiple second convolutional layers, and multiple third convolutional layers. The third convolutional layer corresponds one-to-one with the first convolutional layer, and the third convolutional layer is a skip convolutional layer of the corresponding first convolutional layer.

[0206] In one embodiment, the above-mentioned cable tunnel crack seepage identification device further includes: a first construction module, a second construction module, and a training module, wherein:

[0207] The first building module is used to build the initial identification model; the initial identification model includes the initial fusion sub-model, the initial enhancement sub-model, and the initial leakage segmentation marker sub-model.

[0208] The second building module is used to construct the no-reference loss function and the detection performance evaluation index; the detection performance evaluation index is used to evaluate the detection performance of overlap and similarity between the ground truth label and the predicted mask.

[0209] The training module is used to acquire a dataset of dark light from cable tunnel leakage, and to train the initial recognition model based on the dataset, the no-reference loss function, and the detection performance evaluation index, thus obtaining the recognition model.

[0210] In one embodiment, the aforementioned no-reference loss function is determined based on at least one of the following: spatial consistency loss function, exposure control loss function, color consistency loss function, total variation loss function, and channel consistency loss function.

[0211] The modules in the aforementioned cable tunnel crack seepage identification device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0212] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0213] The cable tunnel image is inverted to obtain an inverted image;

[0214] By identifying the fusion sub-model in the model, the cable tunnel image and the inverted image are processed to determine the two-phase fusion features;

[0215] Based on the dual-phase fusion features and the enhancement sub-model in the recognition model, data augmentation is performed on cable tunnel images to obtain enhanced images;

[0216] The enhanced image and the dual-phase fusion features are input into the leakage segmentation marker sub-model in the recognition model to obtain the tunnel leakage segmentation marker map output by the leakage segmentation marker sub-model.

[0217] The seepage areas in cable tunnels are determined based on the tunnel seepage segmentation marking map.

[0218] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0219] The pixel data of the enhanced image is separated based on the pixel color to obtain the three-channel features corresponding to the enhanced image;

[0220] The three-channel features are input into the adaptive module for processing to obtain the first processed data;

[0221] By stacking the first processed data, three-channel features, and two-phase fusion features using arbitrary modules of the fragment, a tunnel seepage segmentation and marking map output by the water leakage segmentation and marking sub-model is obtained.

[0222] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0223] The three-channel features are input into the convolutional layer of the adaptive module for feature extraction, resulting in single-channel features corresponding to the three-channel features. The single-channel features include red channel data, green channel data, and blue channel data.

[0224] Max pooling is performed on single-channel features to enhance texture features associated with leak area edge detection;

[0225] The enhanced texture features and the downsampled enhanced image are superimposed to obtain the first processed data.

[0226] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0227] The cable tunnel image and its inverted image are input into the fusion sub-model; the fusion sub-model includes a convolution module and a fusion module.

[0228] The cable tunnel image is processed by multiple convolution modules to obtain the first feature map.

[0229] The inverted image is processed by multiple convolution modules to obtain the second feature map;

[0230] The first feature map and the second feature map are fused to obtain the biphasic fused feature.

[0231] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0232] The convolution module includes multiple convolutional layers, multiple activation function layers, and an attention mechanism layer. The multiple activation function layers are placed between some of the convolutional layers in the multiple convolutional layers, and the attention mechanism layer is placed after each convolutional layer.

[0233] The multi-layer convolutional layer includes multiple first convolutional layers, multiple second convolutional layers, and multiple third convolutional layers. The third convolutional layer corresponds one-to-one with the first convolutional layer, and the third convolutional layer is a skip convolutional layer of the corresponding first convolutional layer.

[0234] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0235] Construct an initial identification model; the initial identification model includes an initial fusion sub-model, an initial enhancement sub-model, and an initial leakage segmentation and labeling sub-model;

[0236] Construct a no-reference loss function and a detection performance evaluation index; the detection performance evaluation index is used to evaluate the detection performance of overlap and similarity between the ground truth label and the predicted mask;

[0237] A dataset of dark light from cable tunnel leakage was obtained, and the initial recognition model was trained based on the dataset, the no-reference loss function, and the detection performance evaluation index to obtain the recognition model.

[0238] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0239] The no-reference loss function is determined based on at least one of the following: spatial consistency loss function, exposure control loss function, color consistency loss function, total variation loss function, and channel consistency loss function.

[0240] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0241] The cable tunnel image is inverted to obtain an inverted image;

[0242] By identifying the fusion sub-model in the model, the cable tunnel image and the inverted image are processed to determine the two-phase fusion features;

[0243] Based on the dual-phase fusion features and the enhancement sub-model in the recognition model, data augmentation is performed on cable tunnel images to obtain enhanced images;

[0244] The enhanced image and the dual-phase fusion features are input into the leakage segmentation marker sub-model in the recognition model to obtain the tunnel leakage segmentation marker map output by the leakage segmentation marker sub-model.

[0245] The seepage areas in cable tunnels are determined based on the tunnel seepage segmentation marking map.

[0246] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0247] The leakage segmentation marker sub-model includes an adaptive module and a segment arbitrary module. Enhanced images and biphasic fusion features are input into the leakage segmentation marker sub-model within the recognition model to obtain the tunnel leakage segmentation marker map output by the leakage segmentation marker model, including:

[0248] The pixel data of the enhanced image is separated based on the pixel color to obtain the three-channel features corresponding to the enhanced image;

[0249] The three-channel features are input into the adaptive module for processing to obtain the first processed data;

[0250] By stacking the first processed data, three-channel features, and two-phase fusion features using arbitrary modules of the fragment, a tunnel seepage segmentation and marking map output by the water leakage segmentation and marking sub-model is obtained.

[0251] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0252] The three-channel features are input into the convolutional layer of the adaptive module for feature extraction, resulting in single-channel features corresponding to the three-channel features. The single-channel features include red channel data, green channel data, and blue channel data.

[0253] Max pooling is performed on single-channel features to enhance texture features associated with leak area edge detection;

[0254] The enhanced texture features and the downsampled enhanced image are superimposed to obtain the first processed data.

[0255] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0256] The cable tunnel image and its inverted image are input into the fusion sub-model; the fusion sub-model includes a convolution module and a fusion module.

[0257] The cable tunnel image is processed by multiple convolution modules to obtain the first feature map.

[0258] The inverted image is processed by multiple convolution modules to obtain the second feature map;

[0259] The first feature map and the second feature map are fused to obtain the biphasic fused feature.

[0260] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0261] The convolution module includes multiple convolutional layers, multiple activation function layers, and an attention mechanism layer. The multiple activation function layers are placed between some of the convolutional layers in the multiple convolutional layers, and the attention mechanism layer is placed after each convolutional layer.

[0262] The multi-layer convolutional layer includes multiple first convolutional layers, multiple second convolutional layers, and multiple third convolutional layers. The third convolutional layer corresponds one-to-one with the first convolutional layer, and the third convolutional layer is a skip convolutional layer of the corresponding first convolutional layer.

[0263] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0264] Construct an initial identification model; the initial identification model includes an initial fusion sub-model, an initial enhancement sub-model, and an initial leakage segmentation and labeling sub-model;

[0265] Construct a no-reference loss function and a detection performance evaluation index; the detection performance evaluation index is used to evaluate the detection performance of overlap and similarity between the ground truth label and the predicted mask;

[0266] A dataset of dark light from cable tunnel leakage was obtained, and the initial recognition model was trained based on the dataset, the no-reference loss function, and the detection performance evaluation index to obtain the recognition model.

[0267] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0268] The no-reference loss function is determined based on at least one of the following: spatial consistency loss function, exposure control loss function, color consistency loss function, total variation loss function, and channel consistency loss function.

[0269] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, performs the following steps:

[0270] The cable tunnel image is inverted to obtain an inverted image;

[0271] By identifying the fusion sub-model in the model, the cable tunnel image and the inverted image are processed to determine the two-phase fusion features;

[0272] Based on the dual-phase fusion features and the enhancement sub-model in the recognition model, data augmentation is performed on cable tunnel images to obtain enhanced images;

[0273] The enhanced image and the dual-phase fusion features are input into the leakage segmentation marker sub-model in the recognition model to obtain the tunnel leakage segmentation marker map output by the leakage segmentation marker sub-model.

[0274] The seepage areas in cable tunnels are determined based on the tunnel seepage segmentation marking map.

[0275] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0276] The pixel data of the enhanced image is separated based on the pixel color to obtain the three-channel features corresponding to the enhanced image;

[0277] The three-channel features are input into the adaptive module for processing to obtain the first processed data;

[0278] By stacking the first processed data, three-channel features, and two-phase fusion features using arbitrary modules of the fragment, a tunnel seepage segmentation and marking map output by the water leakage segmentation and marking sub-model is obtained.

[0279] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0280] The three-channel features are input into the convolutional layer of the adaptive module for feature extraction, resulting in single-channel features corresponding to the three-channel features. The single-channel features include red channel data, green channel data, and blue channel data.

[0281] Max pooling is performed on single-channel features to enhance texture features associated with leak area edge detection;

[0282] The enhanced texture features and the downsampled enhanced image are superimposed to obtain the first processed data.

[0283] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0284] The cable tunnel image and its inverted image are input into the fusion sub-model; the fusion sub-model includes a convolution module and a fusion module.

[0285] The cable tunnel image is processed by multiple convolution modules to obtain the first feature map.

[0286] The inverted image is processed by multiple convolution modules to obtain the second feature map;

[0287] The first feature map and the second feature map are fused to obtain the biphasic fused feature.

[0288] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0289] The convolution module includes multiple convolutional layers, multiple activation function layers, and an attention mechanism layer. The multiple activation function layers are placed between some of the convolutional layers in the multiple convolutional layers, and the attention mechanism layer is placed after each convolutional layer.

[0290] The multi-layer convolutional layer includes multiple first convolutional layers, multiple second convolutional layers, and multiple third convolutional layers. The third convolutional layer corresponds one-to-one with the first convolutional layer, and the third convolutional layer is a skip convolutional layer of the corresponding first convolutional layer.

[0291] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0292] Construct an initial identification model; the initial identification model includes an initial fusion sub-model, an initial enhancement sub-model, and an initial leakage segmentation and labeling sub-model;

[0293] Construct a no-reference loss function and a detection performance evaluation index; the detection performance evaluation index is used to evaluate the detection performance of overlap and similarity between the ground truth label and the predicted mask;

[0294] A dataset of dark light from cable tunnel leakage was obtained, and the initial recognition model was trained based on the dataset, the no-reference loss function, and the detection performance evaluation index to obtain the recognition model.

[0295] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0296] The no-reference loss function is determined based on at least one of the following: spatial consistency loss function, exposure control loss function, color consistency loss function, total variation loss function, and channel consistency loss function.

[0297] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0298] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0299] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for identifying water seepage through cracks in cable tunnels, characterized in that, The method includes: The cable tunnel image is inverted to obtain an inverted image; The cable tunnel image and the inverted image are processed by identifying the fusion sub-model in the model to determine the two-phase fusion features; Based on the dual-phase fusion features and the enhancement sub-model in the recognition model, the cable tunnel image is augmented to obtain an enhanced image. The enhanced image and the dual-phase fusion features are input into the leakage segmentation marker sub-model in the recognition model to obtain the tunnel leakage segmentation marker map output by the leakage segmentation marker model; Determine the seepage areas in the cable tunnel based on the tunnel seepage segmentation marking map; The leakage segmentation marker sub-model includes an adaptive module and a segment arbitrary module. The step of inputting the enhanced image and biphasic fusion features into the leakage segmentation marker sub-model in the recognition model to obtain the tunnel leakage segmentation marker map output by the leakage segmentation marker model includes: The pixel data of the enhanced image is separated based on the pixel color to obtain the three-channel features corresponding to the enhanced image; The three-channel features are input into the adaptive module for processing to obtain the first processed data; By stacking the first processed data, the three-channel features, and the dual-phase fusion features using the arbitrary module of the segment, a tunnel seepage segmentation marker map output by the seepage segmentation marker sub-model is obtained.

2. The method according to claim 1, characterized in that, The step of inputting the three-channel features into the adaptive module for processing to obtain the first processed data includes: The three-channel features are input into the convolutional layer of the adaptive module for feature extraction to obtain the single-channel features corresponding to the three-channel features; the single-channel features include red channel data, green channel data, and blue channel data; Max pooling is performed on the single-channel features to enhance the texture features associated with the detection of the leaking area edges; The enhanced texture features and the downsampled enhanced image are superimposed to obtain the first processed data.

3. The method according to claim 1, characterized in that, The process of processing the cable tunnel image and the inverted image through a fusion sub-model in the identification model to determine the two-phase fusion features includes: The cable tunnel image and the inverted image are input into the fusion sub-model; the fusion sub-model includes a convolution module and a fusion module. The cable tunnel image is processed by multiple convolution modules to obtain a first feature map. The inverted image is processed by multiple convolutions through the convolution module to obtain the second feature map; The first feature map and the second feature map are fused to obtain a biphasic fused feature.

4. The method according to claim 3, characterized in that, The convolution module includes multiple convolutional layers, multiple activation function layers, and an attention mechanism layer. The multiple activation function layers are disposed between some of the multiple convolutional layers, and the attention mechanism layer is disposed after the multiple convolutional layers. The multi-layer convolutional layer includes multiple first convolutional layers, multiple second convolutional layers, and multiple third convolutional layers. The third convolutional layer corresponds one-to-one with the first convolutional layer, and the third convolutional layer is a skip convolutional layer of the corresponding first convolutional layer.

5. The method according to any one of claims 1-4, characterized in that, The training process of the recognition model includes: Construct an initial identification model; the initial identification model includes an initial fusion sub-model, an initial enhancement sub-model, and an initial leakage segmentation marker sub-model; A no-reference loss function and a detection performance evaluation index are constructed; the detection performance evaluation index is used to evaluate the detection performance of overlap and similarity between the ground truth label and the predicted mask; Obtain a dark light dataset of cable tunnel leakage water, and train the initial recognition model based on the cable tunnel leakage water dark light dataset, the no-reference loss function, and the detection effect evaluation index to obtain the recognition model.

6. The method according to claim 5, characterized in that, The no-reference loss function is determined based on at least one of the following: spatial consistency loss function, exposure control loss function, color consistency loss function, total variation loss function, and channel consistency loss function.

7. A device for identifying water seepage through cracks in cable tunnels, characterized in that, The device includes: The phase inversion module is used to invert cable tunnel images to obtain inverted images; The fusion module is used to process the cable tunnel image and the inverted image by identifying the fusion sub-model in the recognition model to determine the two-phase fusion features; An enhancement module is used to perform data enhancement on the cable tunnel image based on the dual-phase fusion features and the enhancement sub-model in the recognition model to obtain an enhanced image; The acquisition module is used to input the enhanced image and the dual-phase fusion features into the leakage segmentation marker sub-model in the recognition model, so as to obtain the tunnel leakage segmentation marker map output by the leakage segmentation marker model; The determination module is used to identify seepage areas in cable tunnels based on tunnel seepage segmentation marking maps; The leakage segmentation marker sub-model includes an adaptive module and a segment arbitrary module. The step of inputting the enhanced image and biphasic fusion features into the leakage segmentation marker sub-model in the recognition model to obtain the tunnel leakage segmentation marker map output by the leakage segmentation marker model includes: The pixel data of the enhanced image is separated based on the pixel color to obtain the three-channel features corresponding to the enhanced image; The three-channel features are input into the adaptive module for processing to obtain the first processed data; By stacking the first processed data, the three-channel features, and the dual-phase fusion features using the arbitrary module of the segment, a tunnel seepage segmentation marker map output by the seepage segmentation marker sub-model is obtained.

8. The apparatus according to claim 7, characterized in that, The device further includes: The first construction module is used to construct an initial identification model; the initial identification model includes an initial fusion sub-model, an initial enhancement sub-model, and an initial leakage segmentation marker sub-model. The second construction module is used to construct a no-reference loss function and a detection performance evaluation index; the detection performance evaluation index is used to evaluate the detection performance of overlap and similarity between the ground truth label and the predicted mask; The training module is used to acquire a dark light dataset of cable tunnel leakage water, and to train the initial recognition model based on the cable tunnel leakage water dark light dataset, the no-reference loss function, and the detection effect evaluation index to obtain the recognition model.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images