Underground cable detection method and system based on feature mixing
By building a multi-layer network model and combining expert descriptions with image features, the problem of insufficient accuracy in underground cable X-ray image detection is solved, and more efficient underground cable location detection is achieved.
Patent Information
- Application Number
- CN202510496187.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-21
- Publication Date
- 2025-09-23
AI Technical Summary
Existing technologies make it difficult to effectively use expert knowledge to describe underground cable X-ray images, resulting in insufficient accuracy in underground cable detection, especially in complex environments where it is difficult to accurately extract image features.
A multi-layer network model was constructed, including an image description layer, an image feature extraction layer, a feature relationship interaction layer, and an underground cable detection layer. The text features of the expert description and the X-ray image features were combined to detect the location of underground cables through a deep learning network.
It improves the precision and accuracy of underground cable detection and provides rich contextual information to guide subsequent fault detection and automated inspections.
Smart Images
Figure CN120689266A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of underground cable X-ray image detection, and in particular to an underground cable detection method and system based on feature mixing. Background Art
[0002] As an essential component of urban infrastructure, the safe operation of underground cables is crucial for ensuring the city's power supply. However, due to the complexity of the underground environment, cables may develop various defects and failures during long-term use. X-ray inspection technology, as a non-destructive testing method, can effectively identify defects in the internal structure of underground cables, such as cracks and corrosion, thereby improving cable maintenance efficiency and safety. However, because X-ray images often contain a large amount of noise and interference information, and the boundaries between the cable area and other areas (such as soil, pipelines, etc.) may be unclear, this makes underground cable inspection more difficult.
[0003] In recent years, with the rapid development of computer vision and artificial intelligence technologies, image enhancement methods based on deep learning have gradually become a research hotspot. These methods can achieve efficient object detection in complex environments by training large amounts of image data and learning image feature representation strategies. Summary of the Invention
[0004] In view of the above-mentioned problems, the present invention is proposed.
[0005] Therefore, the technical problem addressed by this invention is: The key to detecting underground cable X-ray images is to cleverly leverage expert descriptions of underground cables in X-ray images to extract representative text features from them. Furthermore, representative image features can be extracted from X-ray images to meet the needs of subsequent applications such as underground cable detection. Currently, there is an urgent need to design a suitable deep learning network framework to address this problem, leverage high-performance computer processing power to train the network, and enhance the quality of underground cable X-ray images.
[0006] To solve the above technical problems, the present invention provides the following technical solutions: a method for detecting underground cables based on feature mixing, comprising: obtaining an X-ray image dataset of underground cables.
[0007] Build a multi-layer network model and optimize the model.
[0008] The image to be detected is input into the optimized model to complete the detection of the underground cable position in the X-ray image.
[0009] As a preferred solution of the underground cable detection method based on feature mixing described in the present invention, wherein: the X-ray image dataset of the underground cable is obtained, the dataset includes the underground cable X-ray image dataset, the underground cable location information and the expert description information of part of the image.
[0010] As a preferred solution of the underground cable detection method based on feature mixing described in the present invention, wherein: the construction of a multi-layer network model includes constructing a four-layer network model, including an image description layer, an image feature extraction layer, a feature relationship interaction layer and an underground cable detection layer.
[0011] As a preferred solution of the underground cable detection method based on feature mixing described in the present invention, the four-layer network model includes an image description layer, a general visual language model to describe the image, and uses the expert knowledge information in the underground cable X-ray image data set for feedback optimization, so that the general visual language model can describe the image from multiple dimensions and generate a text description. Then, a text encoder is used to encode the text description into text features to obtain the text feature F T .
[0012] The image feature extraction layer includes multiple residual convolution layers and two convolution layers C1. The image feature extraction module performs image encoding on the input underground cable X-ray image to obtain the image feature F I .
[0013] Feature relationship interaction layer for text feature F T and image feature F I Perform feature interaction to obtain the interactive feature map F f , including internal attention submodule, dual attention submodule, convolutional layer C1, activation function layer and 3 fully connected layers.
[0014] The underground cable detection layer includes 3 convolutional layers, 1 pooling layer, and 2 fully connected layers. The position detection module performs a deep learning operation on the image depth features {f l ,f m ,f s}Perform feature recombination and finally obtain the underground cable location information.
[0015] As a preferred solution of the underground cable detection method based on feature mixing described in the present invention, the image feature extraction layer includes processing the input underground cable X-ray image to obtain the image feature F I The convolution kernel size of all convolution layers C1 in the image feature extraction layer is 1*1, and the convolution kernel size of all convolution layers C3 is 3*3. The synthesis mechanism is expressed as:
[0016]
[0017] in represents the i-th residual convolution layer, and C1() represents the convolution layer C1.
[0018] The feature relationship interaction layer includes the text feature F T Input to the internal attention submodule to obtain the filtered text feature F TT , the synthesis mechanism is expressed as:
[0019] F TT =FC2(FC1(Pool(C3(F T ))))×C3(F T )
[0020] Among them, FC1 represents the fully connected layer, FC2 represents the fully connected layer, and Pool() represents the pooling layer.
[0021] The text feature F T and image feature F I Input the dual attention submodule to extract image features and obtain the optimized image features F IT , the synthesis mechanism is expressed as:
[0022]
[0023] Among them, e represents the dot product operation, Indicates intermediate features.
[0024] The text feature F will be filtered TT and optimize image features F IT After passing through the fully connected layer, they are projected into a shared space and the features are added element by element to obtain the feature map.
[0025] The image feature F I After inputting into the convolution layer C1 and the activation function layer, the result is combined with the feature map Perform Hadamard product operation and pass through the fully connected layer to finally obtain the interaction feature map F f .
[0026] The underground cable detection layer includes: l ,f m ,f s} Input 3 convolutional layers and obtain the region of interest features {R1,...,R n}.
[0027] The obtained region of interest features are input into the pooling layer and two fully connected layers to obtain the underground cable location information {p i ,t i}, where pi represents the confidence that the predicted location box i contains underground cables, t i Indicates the offset required for the predicted location box i.
[0028] As a preferred embodiment of the underground cable detection method based on feature mixing of the present invention, the optimization of the model includes optimizing the loss function of the network model based on the prediction results and the manually annotated underground cable location information in the underground cable X-ray image dataset to achieve convergence of the network model and obtain a trained network model, which is expressed as:
[0029]
[0030] Among them, p i Indicates the confidence that the detection box i predicted by the position detection module contains defects, It represents the label of the prior frame i, 0 means no defect, 1 means defect, and the labeling method is obtained by calculating the intersection of the detection frame i and the real frame marked manually. If the intersection of the intersection of the detection frame i is greater than 0.7, it is marked as 1, and if it is less than 0.3, it is marked as 0. Assuming that the label of the prior frame i is 1, then t i Indicates the offset required by the detection box i predicted by the position detection module, N represents the offset between the detection box i and the annotated ground truth box. cls Indicates the total number of detection boxes marked as 1 and 0 in the detection box, N reg Represents the total number of detection boxes marked as 1 in the detection box, and λ represents the weight parameter. Classification loss L reg Using the cross entropy loss function, the regression loss L cls Using SmoothL1Loss, it is expressed as:
[0031]
[0032] By calculating the loss function results and performing backpropagation calculations using stochastic gradient descent, the network parameters of the network model are optimized. When the training epoch of the network model has reached the preset training epoch, the network model has achieved convergence and a trained network model is obtained.
[0033] As a preferred solution of the underground cable detection method based on feature mixing described in the present invention, the step of inputting the image to be detected into the optimized model to complete the detection of the underground cable position in the X-ray image includes taking X-ray images of underground cables and high-quality and high-definition X-ray images of underground cables, and manually annotating the underground cable position information of the corresponding images. At the same time, experts describe the underground cables in the X-ray images, including multiple dimensions such as size and length.
[0034] An underground cable detection system based on feature mixing, characterized by comprising:
[0035] The data acquisition module acquires X-ray image datasets of underground cables.
[0036] Optimization module, builds a multi-layer network model and optimizes the model.
[0037] The detection module inputs the image to be detected into the optimized model to complete the detection of the underground cable position in the X-ray image.
[0038] A computer device includes a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.
[0039] A computer-readable storage medium stores a computer program, which implements the steps of the method described above when executed by a processor.
[0040] The present invention demonstrates the following benefits: The proposed method and system for underground cable detection based on feature blending applies deep learning network technology to the problem of detecting underground cable locations in X-ray images. By combining the textual features of expert descriptions with the image features of X-ray images, underground cables can be detected more accurately. The expert descriptions provide rich contextual information, helping the model understand the complex structures and details in the image, thereby improving detection accuracy and providing guidance for subsequent underground cable fault detection and automated inspections. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive effort. Among them:
[0042] Figure 1 This is an overall flow chart of a method and system for underground cable detection based on feature mixing provided in the first embodiment of the present invention.
[0043] Figure 2 A structural diagram of an image description module of an underground cable detection method and system based on feature mixing provided in the first embodiment of the present invention.
[0044] Figure 3 A structural diagram of an image feature extraction module of an underground cable detection method and system based on feature mixing provided in the first embodiment of the present invention.
[0045] Figure 4 A structural diagram of an image feature extraction module of an underground cable detection method and system based on feature mixing provided in the first embodiment of the present invention.
[0046] Figure 5 A structural diagram of a feature relationship interaction module of a feature mixing-based underground cable detection method and system provided in the first embodiment of the present invention.
[0047] Figure 6 A structural diagram of a feature relationship interaction module of a feature mixing-based underground cable detection method and system provided in the first embodiment of the present invention.
[0048] Figure 7 A structural diagram of a feature relationship interaction module of a feature mixing-based underground cable detection method and system provided in the first embodiment of the present invention. DETAILED DESCRIPTION
[0049] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.
[0050] Example 1
[0051] Reference Figures 1 to 7 , as an embodiment of the present invention, provides an underground cable detection method based on feature mixing, comprising:
[0052] S1: Obtain an X-ray image dataset of underground cables.
[0053] The dataset includes underground cable X-ray image dataset, underground cable location information, and expert description information of some images.
[0054] S2: Build a multi-layer network model and optimize the model.
[0055] A four-layer network model is constructed, including image description layer, image feature extraction layer, feature relationship interaction layer and underground cable detection layer.
[0056] It should be noted that if Figure 2 As shown in Figure 2, the image description module uses a general visual language model to describe the image, and uses the expert knowledge information in the underground cable X-ray image dataset for feedback optimization to generate a text description. The text encoder is then used to encode the text features of the text description to obtain the text feature F T .
[0057] Furthermore, the general visual language model can use pre-trained visual language models such as GPT4 and Spark AI. Expert knowledge describes underground cables in X-ray images, including multiple dimensions such as size and length. Through interactive conversation, the general visual language model can provide a relatively accurate multi-dimensional description of underground cables. The text encoder can use a text sequence encoder or a Transformer model.
[0058] The image feature extraction layer includes multiple residual convolution layers and two convolution layers C1. The image feature extraction module performs image encoding on the input underground cable X-ray image to obtain the image feature F I .
[0059] Feature relationship interaction layer for text feature F T and image feature F I Perform feature interaction to obtain the interactive feature map F f , including internal attention submodule, dual attention submodule, convolutional layer C1, activation function layer and 3 fully connected layers.
[0060] The underground cable detection layer includes 3 convolutional layers, 1 pooling layer, and 2 fully connected layers. The location detection module uses image depth features {f l ,f m ,f s}Perform feature recombination and finally obtain the underground cable location information.
[0061] like Figure 3 As shown in the figure, the image feature extraction module processes the input underground cable X-ray image to obtain the image feature F I The image feature extraction module includes two convolutional layers C1 and multiple residual convolutional layers, where the residual convolutional layer includes two convolutional layers C1 and one convolutional layer C3. The convolution kernel size of all convolutional layers C1 is 1*1, and the convolution kernel size of all convolutional layers C3 is 3*3. The specific synthesis mechanism is as follows:
[0062]
[0063] in represents the i-th residual convolution layer, and C1() represents the convolution layer C1.
[0064] like Figure 4 、 Figure 5 、 Figure 6 As shown, the feature relationship interaction layer includes the text feature F T Input to the internal attention submodule to obtain the filtered text feature F TT, including the internal attention submodule, the dual attention module, and multiple fully connected layers. Specifically:
[0065] The text feature F T Input to the internal attention submodule to obtain the filtered text feature F TT , the synthesis mechanism is expressed as:
[0066] F TT =FC2(FC1(Pool(C3(F T ))))×C3(F T )
[0067] Among them, FC1 represents the fully connected layer, FC2 represents the fully connected layer, and Pool() represents the pooling layer.
[0068] The text feature F T and image feature F I Input the dual attention submodule to extract image features and obtain the optimized image features F IT , the synthesis mechanism is expressed as:
[0069]
[0070] The text feature F will be filtered TT and optimize image features F IT After passing through the fully connected layer, they are projected into a shared space and the features are added element by element to obtain the feature map.
[0071] Furthermore, the text feature F TT and optimize image features F IT After passing through the fully connected layer, they are projected into a shared space, and then the above features are added element by element to obtain the feature map
[0072] The image feature F I After inputting into the convolution layer C1 and the activation function layer, the result is combined with the feature map Perform Hadamard product operation and pass through the fully connected layer to finally obtain the interaction feature map F f .
[0073] like Figure 7 As shown, the underground cable detection module is used to detect the interactive feature map F f Perform feature extraction and use pooling layer and full connection layer to obtain underground cable location information {p i ,t i}. The fully connected layer 1 is used to process underground cable features and has a length of 1024. The fully connected layer 2 has a length of 5 and is used to extract the confidence of the candidate box of the underground cable position and to further adjust the orientation of the candidate box. The underground cable detection layer converts the image depth feature {f l ,f m ,f s} Input 3 convolutional layers and obtain the region of interest features {R1,...,R n}.
[0074] The obtained region of interest features are input into the pooling layer and two fully connected layers to obtain the underground cable location information {p i ,t i}, where p i represents the confidence that the predicted location box i contains underground cables, t i Indicates the offset required for the predicted location box i.
[0075] S3: Input the image to be detected into the optimized model to complete the detection of the underground cable position in the X-ray image.
[0076] According to the prediction results and the manually marked underground cable location information in the underground cable X-ray image dataset, the loss function of the network model is optimized to achieve the convergence of the network model and obtain the trained network model, which is expressed as:
[0077]
[0078] Among them, p i Indicates the confidence that the detection box i predicted by the position detection module contains defects, It represents the label of the prior frame i, 0 means no defect, 1 means defect, and the labeling method is obtained by calculating the intersection of the detection frame i and the real frame marked manually. If the intersection of the intersection of the detection frame i is greater than 0.7, it is marked as 1, and if it is less than 0.3, it is marked as 0. Assuming that the label of the prior frame i is 1, then t i Indicates the offset required by the detection box i predicted by the position detection module, N represents the offset between the detection box i and the annotated ground truth box. cls Indicates the total number of detection boxes marked as 1 and 0 in the detection box, N reg Represents the total number of detection boxes marked as 1 in the detection box, and λ represents the weight parameter. Classification loss L reg Using the cross entropy loss function, the regression loss L cls Using SmoothL1Loss, it is expressed as:
[0079]
[0080] By calculating the loss function results and performing backpropagation calculations using stochastic gradient descent, the network parameters of the network model are optimized. When the training epoch of the network model has reached the preset training epoch, the network model has achieved convergence and a trained network model is obtained.
[0081] The detection of underground cable locations in X-ray images is completed by taking X-ray images of underground cables, as well as high-quality and high-definition X-ray images of underground cables, and manually labeling the underground cable location information of the corresponding images. At the same time, experts describe the underground cables in the X-ray images, including multiple dimensions such as size and length.
[0082] The above embodiments also include an underground cable detection system based on feature mixing, specifically:
[0083] The data acquisition module acquires X-ray image datasets of underground cables.
[0084] Optimization module, builds a multi-layer network model and optimizes the model.
[0085] The detection module inputs the image to be detected into the optimized model to complete the detection of the underground cable position in the X-ray image.
[0086] The computer device may be a server. The computer device includes a processor, a memory, an input / output interface (I / O) and a communication interface. The processor, the memory and the input / output interface are connected via a system bus, and the communication interface is connected to the system bus via the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data cluster data of the power monitoring system. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a feature-mixing-based underground cable detection method is implemented.
[0087] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may 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 may include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in the various embodiments provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processor involved in the various embodiments provided in this application may be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic unit, a data processing logic unit based on quantum computing, etc., but are not limited to these.
[0088] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.
Claims
1. A method for underground cable detection based on feature mixing, characterized in that: include: Obtain X-ray image datasets of underground cables; Build a multi-layer network model and optimize the model; The image to be detected is input into the optimized model to complete the detection of the underground cable position in the X-ray image.
2. The underground cable detection method based on feature mixing according to claim 1, characterized in that: The method of obtaining an X-ray image dataset of underground cables includes: the dataset includes an X-ray image dataset of underground cables, location information of underground cables, and description information of a portion of the images by experts.
3. The underground cable detection method based on feature mixing according to claim 2, characterized in that: The construction of the multi-layer network model includes constructing a four-layer network model, namely an image description layer, an image feature extraction layer, a feature relationship interaction layer and an underground cable detection layer.
4. The underground cable detection method based on feature mixing according to claim 3, wherein: The four-layer network model includes an image description layer, a general visual language model to describe the image, and uses the expert knowledge information in the underground cable X-ray image dataset for feedback optimization, so that the general visual language model can describe the image from multiple dimensions and generate a text description; then the text description is encoded using a text encoder to obtain text features F T ; The image feature extraction layer includes multiple residual convolution layers and two convolution layers C1; the image feature extraction module performs image encoding on the input underground cable X-ray image to obtain the image feature F I ; Feature relationship interaction layer for text feature F T and image feature F I Perform feature interaction to obtain the interactive feature map F f , including internal attention submodule, dual attention submodule, convolution layer C1, activation function layer and 3 fully connected layers; The underground cable detection layer consists of 3 convolutional layers, 1 pooling layer, and 2 fully connected layers; The position detection module detects the image depth feature {f l ,f m ,f s }Perform feature recombination and finally obtain the underground cable location information.
5. The underground cable detection method based on feature mixing according to claim 4, characterized in that: The image feature extraction layer includes processing the input underground cable X-ray image to obtain the image feature F I ; The convolution kernel size of all convolution layers C1 in the image feature extraction layer is 1*1, and the convolution kernel size of all convolution layers C3 is 3*3; the synthesis mechanism is expressed as: in, represents the i-th residual convolution layer, represents the i-1th residual convolution layer, C1 represents the convolution layer; The feature relationship interaction layer includes the text feature F T Input to the internal attention submodule to obtain the filtered text feature F TT , the synthesis mechanism is expressed as: F TT =FC2(FC1(Pool(C3(F T ))))×C3(F T ) Among them, FC1 represents the fully connected layer, FC2 represents the fully connected layer, and Pool represents the pooling layer; The text feature F T and image feature F I Input the dual attention submodule to extract image features and obtain the optimized image features F IT , the synthesis mechanism is expressed as: Among them, e represents the dot product operation, Indicates intermediate features; The text feature F will be filtered TT and optimize image features F IT After passing through the fully connected layer, they are projected into a shared space and the features are added element by element to obtain the feature map. The image feature F I After inputting into the convolution layer C1 and the activation function layer, the result is combined with the feature map Perform Hadamard product operation and pass through the fully connected layer to finally obtain the interaction feature map F f ; The underground cable detection layer includes: l ,f m ,f s } Input 3 convolutional layers and obtain the region of interest features {R1,...,R n }; The obtained region of interest features are input into the pooling layer and two fully connected layers to obtain the underground cable location information {p i ,t i }, where p i represents the confidence that the predicted location box i contains underground cables, t i Indicates the offset required for the predicted location box i.
6. The underground cable detection method based on feature mixing according to claim 5, characterized in that: The optimization of the model includes optimizing the loss function of the network model based on the prediction results and the manually marked underground cable location information in the underground cable X-ray image dataset to achieve convergence of the network model and obtain a trained network model, which is expressed as: Among them, p i Indicates the confidence that the detection box i predicted by the position detection module contains defects, Represents the label of the prior box i; Set the label of the prior frame i to 1, then t i Indicates the offset required by the detection box i predicted by the position detection module, Represents the offset between the detection box i and the annotated real box; N cls Indicates the total number of detection boxes marked as 1 and 0 in the detection box, N reg Represents the total number of detection boxes marked as 1 in the detection box, λ represents the weight parameter; classification loss L reg Using the cross entropy loss function, the regression loss L cls Using SmoothL1 Loss, it is expressed as: By calculating the loss function results and using the stochastic gradient descent method for backpropagation calculation, the network parameters of the network model are optimized; when the training epoch of the network model has reached the preset training epoch, the network model has achieved convergence and a trained network model is obtained.
7. The underground cable detection method based on feature mixing according to claim 6, characterized in that: The method of inputting the image to be detected into the optimized model to complete the detection of the underground cable position in the X-ray image includes taking X-ray images of the underground cables, as well as high-quality and high-definition X-ray images of the underground cables, and manually labeling the underground cable position information of the corresponding images. At the same time, experts describe the underground cables in the X-ray images, including multiple dimensions such as size and length.
8. An underground cable detection system based on feature mixing using the method according to any one of claims 1 to 7, characterized in that: Including data acquisition module, optimization module, and detection module; The data acquisition module acquires an X-ray image dataset of underground cables; The optimization module constructs a multi-layer network model and optimizes the model; The detection module inputs the image to be detected into the optimized model to complete the detection of the underground cable position in the X-ray image.
9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.