Underground cable abnormal damage detection method
Through deep learning technology and multi-scale feature extraction, the problems of low efficiency and low accuracy in underground cable detection are solved, and the precise positioning of abnormal damage areas of underground cables is achieved, providing an effective detection method for underground cable safety inspection.
Patent Information
- Application Number
- CN202411968834.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-30
- Publication Date
- 2025-05-16
AI Technical Summary
The prior art is inefficient and has low accuracy when detecting abnormal damage to underground cables, especially in complex environments.
By using X-ray equipment to acquire underground cable images, combined with deep learning technology, a regional molecular network, background difference molecular network and feature fusion subnet are constructed, multi-scale features are extracted and loss functions are optimized to achieve accurate positioning of the abnormally damaged areas of underground cables.
It realizes effective extraction and precise positioning of the characteristics of abnormal damage areas of underground cables, improves detection accuracy, and meets the needs of underground cable safety inspection.
Smart Images
Figure CN120014318A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of underground cable detection, in particular to a method for detecting abnormal damage of underground cables. Background Art
[0002] Underground cables are an important part of modern urban infrastructure. They are responsible for transmitting electricity and data to ensure the normal operation of the city. However, due to the complexity and invisibility of the underground environment, underground cables are prone to abnormal damage when subjected to external forces and are prone to failure. For example, misoperation during piling may cause cable damage, which in turn affects the stability and safety of power supply. Traditional underground cable detection methods mainly rely on manual inspections and simple physical detection methods, which are inefficient and inaccurate. In recent years, although some detection methods based on advanced technologies such as electromagnetic methods and acoustic-magnetic synchronization methods have emerged, these methods still have certain limitations in their application in complex environments. Real-time monitoring of whether underground cable areas have suffered abnormal damage through X-ray images / videos has become an important means.
[0003] Unsupervised background difference method is a widely recognized video analysis method, but its application scenarios have great limitations and its adaptability to various environmental disturbances is relatively poor. After the background difference method is introduced and combined with deep learning technology, the performance and robustness of the algorithm have been greatly improved, but there is still a large degree of scene dependence. Summary of the invention
[0004] In view of the above problems or problems existing in the prior art, the present invention is proposed.
[0005] Therefore, the purpose of the present invention is to provide a method for detecting abnormal damage of underground cables, which can identify the characteristics of abnormal objects and background features that cause damage in X-ray images of underground cables, and promptly discover foreign body damage to meet the needs of underground cable maintenance.
[0006] To solve the above technical problems, the present invention provides the following technical solutions: a method for detecting abnormal damage of underground cables, which comprises using an X-ray device to collect images of underground cables to form a training data set;
[0007] Input the collected images into the network model to obtain the prediction results;
[0008] Optimize the model’s loss function based on manually labeled information and prediction results;
[0009] Through loss function optimization, accurate network model is trained;
[0010] Through the trained and accurate network model, abnormal damage to underground cables can be detected;
[0011] Improve detection accuracy through multi-scale feature extraction and fusion;
[0012] Output the inspection results, marking the area damaged by foreign matter or displaying a completely black image to indicate no damage.
[0013] As a preferred solution of the underground cable abnormal damage detection method of the present invention, the training data set includes an underground cable image data set of foreign matter damage and an underground cable background template image data set.
[0014] As a preferred solution of the underground cable abnormal damage detection method of the present invention, wherein: the network model includes a region division sub-network, a background difference sub-network and a feature fusion sub-network;
[0015] The purpose of the network model is to analyze X-ray images of underground cables through deep learning technology to detect and locate areas of foreign object damage.
[0016] As a preferred solution of the underground cable abnormal damage detection method of the present invention, the manual marking information includes underground cable area information and foreign body damage area Mask information.
[0017] As a preferred solution of the underground cable abnormal damage detection method of the present invention, the loss function includes a regional binary classification cross entropy loss function, a Mask binary classification cross entropy loss function, and a Dice loss function.
[0018] As a preferred solution of the underground cable abnormal damage detection method of the present invention, the regional binary classification cross entropy loss function is used to train the regional division subnetwork, optimize the positioning and extraction of the underground cable area, and calculate the difference between the true value and the output result of the regional division subnetwork. The formula is as follows:
[0019]
[0020] Among them, y is the true value, Output result of dividing the subnetwork into regions.
[0021] As a preferred solution of the underground cable abnormal damage detection method of the present invention, the Mask binary classification cross entropy loss function is used to optimize the network model, adjust the amplitude of the overall cross entropy loss and the loss size of the difficult and easy samples. The difference between the Mask value of the foreign body damage area and the true Mask value is predicted, and the formula is as follows:
[0022]
[0023] Among them, α is used to adjust the amplitude of the overall cross entropy loss, and β is used to adjust the loss size of difficult and easy samples. The larger the value of β, the smaller the loss value of easy samples. rIndicates the predicted foreign body damage area Mask value, parameter Indicates the real Mask value of foreign body damage.
[0024] As a preferred solution of the underground cable abnormal damage detection method of the present invention, the Dice loss function is used to optimize the network model and measure the similarity between the real Mask value and the predicted Mask value, and its formula is as follows:
[0025] Among them, |X∩Y| represents the result of element-by-element multiplication of the real Mask value of foreign body damage and the predicted Mask value of the foreign body damage area, and then element-by-element addition, |X| represents the sum of the elements in GT, and |Y| represents the sum of the elements in Mask.
[0026] As a preferred solution of the underground cable abnormal damage detection method of the present invention, the multi-scale feature extraction includes a regional multi-scale feature extraction module and a Mask multi-scale feature extraction module in the feature fusion subnetwork to extract features of different scales on the underground cable regional image and the differential Mask image respectively;
[0027] Feature fusion refers to the fusion of features of different scales by the multi-scale fusion feature module.
[0028] As a preferred solution of the underground cable abnormal damage detection method of the present invention, the foreign body damage area mark is included in the output Mask image, the foreign body damage area is marked in white, and the non-damaged area is marked in black.
[0029] The beneficial effects of the present invention are as follows: The present invention applies deep learning network technology to the problem of detecting abnormal damage of underground cables. By extracting the regional features of underground cables and the differential Mask features to assist in extracting the regional features of foreign objects, the features of the abnormal damage of underground cables can be effectively extracted, thereby achieving precise positioning of the abnormal damage of underground cables and providing guidance for the safety inspection of underground cables. BRIEF DESCRIPTION OF THE DRAWINGS
[0030] In order 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. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them:
[0031] Figure 1 This is a schematic diagram of the structure of the network model of the underground cable abnormal damage detection method.
[0032] Figure 2A flow chart of the regional division of sub-networks for underground cable abnormal damage detection method.
[0033] Figure 3 Schematic diagram of the structure of the background differential sub-network of the underground cable abnormal damage detection method.
[0034] Figure 4 Schematic diagram of the structure of the feature fusion sub-network of the underground cable abnormal damage detection method.
[0035] Figure 5 This is a schematic diagram of the structure of the regional multi-scale feature extraction module in the feature fusion subnetwork of the underground cable abnormal damage detection method.
[0036] Figure 6 This is a structural diagram of the Mask multi-scale feature extraction module in the feature fusion sub-network of the underground cable abnormal damage detection method.
[0037] Figure 7 This is a schematic diagram of the structure of the multi-scale fusion feature module in the feature fusion subnetwork of the underground cable abnormal damage detection method. DETAILED DESCRIPTION
[0038] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.
[0039] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.
[0040] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.
[0041] Example 1
[0042] Reference Figure 1 to Figure 7 , which is an embodiment of the present invention, provides an underground cable abnormal damage detection method, which can utilize SCD file information extraction and statistical analysis to realize the automatic establishment and update of the standard virtual circuit library.
[0043] Specifically, an X-ray device is used to collect underground cable images to form a training data set;
[0044] Input the collected images into the network model to obtain the prediction results;
[0045] Optimize the model’s loss function based on manually labeled information and prediction results;
[0046] Through loss function optimization, accurate network model is trained;
[0047] Through the trained and accurate network model, abnormal damage to underground cables can be detected;
[0048] Improve detection accuracy through multi-scale feature extraction and fusion;
[0049] Output the inspection results, marking the area damaged by foreign matter or displaying a completely black image to indicate no damage.
[0050] Furthermore, the training data set includes a data set of images of underground cables damaged by foreign objects and a data set of images of underground cables background templates.
[0051] Furthermore, the network model includes a region partition sub-network, a background difference sub-network, and a feature fusion sub-network;
[0052] The purpose of the network model is to analyze X-ray images of underground cables through deep learning technology to detect and locate areas of foreign object damage.
[0053] It should be noted that if Figure 2 As shown, the regional division sub-network is used to extract features of the foreign body damaged underground cable images in the foreign body damaged underground cable image data set and to divide the underground cable regions to obtain underground cable region images; wherein the regional division sub-network model includes a convolution layer i and a downsampling layer i, where i = 1, 2, ..., n-1. In this embodiment, n = 5. The convolution kernel size of the convolution layer is 3, the step size defaults to 1, and the padding defaults to 1;
[0054] The sizes of the feature maps generated by downsampling layer 1 to downsampling layer n-1 (n=5) are:
[0055]
[0056] Among them, h and w represent the height and width of the input image.
[0057] The convolution kernel size of the convolution layer n is 3, the step size defaults to 1, and the padding defaults to 1 to obtain the deep features; the upsampling layer converts the size of the deep features into the size of the input foreign object damaging the underground cable image.
[0058] like Figure 3As shown, the background difference sub-network is used to perform a difference operation on the underground cable background template image in the underground cable background template image dataset and the underground cable image damaged by foreign objects in the underground cable image dataset, and a difference Mask image is obtained by distinguishing the foreground from the background; wherein, the specific operation steps of the background difference sub-network are as follows:
[0059] Subtract the grayscale of the underground cable image damaged by foreign matter from the underground cable background template image, and take the absolute value to obtain the difference result;
[0060] For the pixel point in the image of the underground cable damaged by foreign objects, if the difference result value is greater than the threshold T, the pixel point is the foreground value and its value is set to 1; otherwise, it is set to 0; and finally the difference Mask image is obtained. The specific synthesis mechanism is:
[0061]
[0062] Among them, J M Represents the difference Mask image, I r Indicates foreign matter damaging underground cables. B Represents an underground cables background template image.
[0063] like Figure 4 As shown, a feature fusion subnetwork is used to perform multi-scale feature extraction and feature fusion on the underground cable area image and the differential Mask image, and finally the foreign body damage area Mask result of the network model is obtained; according to the foreign body damage area Mask result and the manually annotated underground cable area information and foreign body damage area Mask information in the foreign body damage underground cable image data set, the loss function of the network model is optimized to obtain a trained network model; wherein, the feature fusion subnetwork includes a regional multi-scale feature extraction module, a Mask multi-scale feature extraction module and a multi-scale fusion feature module.
[0064] The underground cable image to be tested and the underground cable background template image are input into the trained network model to complete the detection of foreign body damage and Mask positioning.
[0065] Among them, the background difference sub-network is used to extract the difference Mask image J M ; The background difference subnetwork can use existing background difference methods according to different calculation needs, including but not limited to median method background modeling, mean method background modeling, single Gaussian distribution model, mixed Gaussian distribution model, etc.
[0066] In addition, the feature fusion sub-network is used to fuse the regional image with the differential mask image information to predict the mask result of the foreign body damage area. It includes:
[0067] like Figure 5 As shown in the figure, the regional multi-scale feature extraction module in the feature fusion sub-network performs multi-scale feature extraction on the regional image to obtain regional features of four scales, namely, F Q1 ,F Q2 ,F Q3 ,F Q4 ;
[0068] The multi-scale fusion feature module in the feature fusion sub-network fuses the above-mentioned regional features with the Mask features at different scales, and finally obtains the Mask result of the foreign body damage area.
[0069] Specifically, the regional multi-scale feature extraction module in the feature fusion subnetwork contains four groups of convolutional layers and downsampling layers, which are connected in sequence and represented as {CQi, DQi}, i = 1, 2, 3, 4; the output of each group of convolutional layers and downsampling layers is the regional feature F of the corresponding scale. Qi ;
[0070] The specific synthesis mechanism is:
[0071]
[0072] Where i=2,3,4.
[0073] like Figure 6 As shown in Figure 2, the Mask multi-scale feature extraction module in the feature fusion sub-network performs multi-scale feature extraction on the differential Mask image, and can obtain Mask features of four scales, namely, F M1 ,F M2 ,F M3 ,F M4 ;
[0074] The Mask multi-scale feature extraction module in the feature fusion subnetwork consists of four groups of convolutional layers and downsampling layers, which are connected in sequence and represented as {CMi, DMi}, i = 1, 2, 3, 4; the output of each group of convolutional layers and downsampling layers is the Mask feature F of the corresponding scale. Mi ;
[0075] The specific synthesis mechanism is:
[0076]
[0077] Where i=2,3,4.
[0078] like Figure 7As shown in Fig. 1, the multi-scale fusion feature module in the feature fusion subnetwork contains four groups of convolutional layers and upsampling layers, which are connected in sequence and represented as {CUi, DUi}, i = 1, 2, 3, 4; each group of convolutional layers and upsampling layers fuses and upsamples the regional features and Mask features of the corresponding scale to obtain the corresponding fusion feature F Ri ;
[0079] The specific synthesis mechanism is:
[0080]
[0081] Furthermore, the manually labeled information includes underground cable area information and foreign object damage area Mask information.
[0082] Furthermore, the loss functions include regional binary classification cross entropy loss function, Mask binary classification cross entropy loss function, and Dice loss function.
[0083] Furthermore, the regional binary classification cross entropy loss function is used to train the regional division subnetwork, optimize the positioning and extraction of underground cable areas, and calculate the difference between the true value and the output result of the regional division subnetwork. The formula is as follows:
[0084]
[0085] Among them, y is the true value, Output of the regional subnetwork.
[0086] It should be noted that the region division subnetwork is used to extract the underground cable area J in the image. r , including n convolutional layers, n-1 downsampling layers and one upsampling layer.
[0087] Specifically, the number of convolutional layers and downsampling layers can be set according to actual conditions.
[0088] Furthermore, the Mask binary classification cross entropy loss function is used to optimize the network model and adjust the magnitude of the overall cross entropy loss and the loss size of the difficult and easy samples. The difference between the Mask value of the foreign body damage area and the true Mask value is predicted, and the formula is as follows:
[0089]
[0090] Among them, α is used to adjust the amplitude of the overall cross entropy loss, and β is used to adjust the loss size of difficult and easy samples. The larger the value of β, the smaller the loss value of easy samples. r Indicates the predicted foreign body damage area Mask value, parameter Indicates the real Mask value of foreign body damage.
[0091] Furthermore, the Dice loss function is used to optimize the network model and measure the similarity between the true Mask value and the predicted Mask value. The formula is as follows:
[0092] Among them, |X∩Y| represents the result of element-by-element multiplication of the real Mask value of foreign body damage and the predicted Mask value of the foreign body damage area, and then element-by-element addition, |X| represents the sum of the elements in GT, and |Y| represents the sum of the elements in Mask.
[0093] Furthermore, the multi-scale feature extraction includes a regional multi-scale feature extraction module and a Mask multi-scale feature extraction module in the feature fusion sub-network, which respectively extract features of different scales from the underground cable regional image and the differential Mask image;
[0094] Feature fusion refers to the fusion of features of different scales by the multi-scale fusion feature module.
[0095] Furthermore, the foreign body damaged area mark is included in the output Mask image, the foreign body damaged area is marked as white, and the non-damaged area is marked as black.
[0096] It should be noted that the underground cable image to be tested is input into the trained network model, and the output result is used to determine whether there is foreign object damage; if so, the foreign object area is marked and the output result is a black and white binary image; if not, a completely black image is output.
[0097] In summary, the present invention applies deep learning network technology to the problem of abnormal damage detection of underground cables. By extracting underground cable area features and differential Mask features to assist in extracting foreign body area features, the effective extraction of underground cable abnormal damage area features is achieved, and then the accurate positioning of underground cable abnormal damage areas is achieved, providing guidance for underground cable safety inspections.
[0098] Importantly, it should be noted that the construction and arrangement of the present application shown in a plurality of different exemplary embodiments are only exemplary. Although only a few embodiments are described in detail in this disclosure, it should be readily understood by those who refer to this disclosure that many modifications are possible without substantially departing from the novel teachings and advantages of the subject matter described in the application (e.g., mounting arrangements, use of materials, color, changes in orientation, etc.). For example, the element shown as integrally formed may be composed of a plurality of parts or elements, the position of the element may be inverted or otherwise changed, and the nature or number or position of the discrete element may be altered or changed. Therefore, all such modifications are intended to be included within the scope of the present invention. The order or sequence of any process or method steps may be changed or reordered according to an alternative embodiment. In the claims, any "bracket plus function" clause is intended to cover the structure of the execution function described herein, and is not only structurally equivalent but also equivalent structure. Without departing from the scope of the present invention, other replacements, modifications, changes and omissions may be made in the design, operating conditions and arrangement of the exemplary embodiments. Therefore, the present invention is not limited to a specific embodiment, but extends to a variety of modifications that still fall within the scope of the appended claims.
[0099] Additionally, in an effort to provide a concise description of example embodiments, all features of an actual implementation may not be described.
[0100] It should be understood that in the development of any actual implementation, as in any engineering or design project, numerous implementation-specific decisions may be made. Such a development effort may be complex and time-consuming, but for those of ordinary skill having the benefit of this disclosure, the development effort will be a routine task of design, fabrication, and production without undue experimentation.
[0101] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the 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 detecting abnormal damage of underground cables, characterized in that: include, Use X-ray equipment to collect underground cable images to form a training data set; Input the collected images into the network model to obtain the prediction results; Optimize the model’s loss function based on manually labeled information and prediction results; Through loss function optimization, accurate network model is trained; Through the trained and accurate network model, abnormal damage to underground cables can be detected; Improve detection accuracy through multi-scale feature extraction and fusion; Output the inspection results, marking the area damaged by foreign matter or displaying a completely black image to indicate no damage.
2. The method for detecting abnormal damage of underground cables according to claim 1, characterized in that: The training data set includes a foreign body damaging underground cable image data set and an underground cable background template image data set.
3. The method for detecting abnormal damage of underground cables as claimed in claim 2, wherein: The network model includes a region division sub-network, a background difference sub-network and a feature fusion sub-network; The purpose of the network model is to analyze X-ray images of underground cables through deep learning technology to detect and locate areas of foreign object damage.
4. The method for detecting abnormal damage of underground cables according to any one of claims 2 or 3, characterized in that: The manual marking information includes underground cable area information and foreign matter damage area Mask information.
5. The method for detecting abnormal damage of underground cables according to claim 4, characterized in that: The loss functions include a region binary classification cross entropy loss function, a Mask binary classification cross entropy loss function, and a Dice loss function.
6. The method for detecting abnormal damage of underground cables according to claim 5, characterized in that: The regional binary classification cross entropy loss function is used to train the regional division subnetwork, optimize the positioning and extraction of underground cable areas, and calculate the difference between the true value and the output result of the regional division subnetwork. The formula is as follows: Among them, y is the true value, Output result of dividing the subnetwork into regions.
7. The method for detecting abnormal damage of underground cables according to claim 5, characterized in that: The Mask binary classification cross entropy loss function is used to optimize the network model and adjust the magnitude of the overall cross entropy loss and the loss size of the difficult and easy samples. The difference between the Mask value of the foreign body damage area and the true Mask value is predicted, and the formula is as follows: Among them, α is used to adjust the amplitude of the overall cross entropy loss, and β is used to adjust the loss size of difficult and easy samples. The larger the value of β, the smaller the loss value of easy samples. r Indicates the predicted foreign body damage area Mask value, parameter Indicates the real Mask value of foreign body damage.
8. The method for detecting abnormal damage of underground cables according to claim 5, characterized in that: The Dice loss function is used to optimize the network model and measure the similarity between the true Mask value and the predicted Mask value. The formula is as follows: in, | X∩Y | It represents the result of element-by-element multiplication of the real Mask value of foreign body damage and the predicted Mask value of foreign body damage area, and then element-by-element addition. | X | represents the sum of the elements in GT, | Y | Represents the sum of the elements in Mask.
9. The method for detecting abnormal damage of underground cables according to any one of claims 5 to 8, characterized in that: The multi-scale feature extraction includes a regional multi-scale feature extraction module and a Mask multi-scale feature extraction module in the feature fusion sub-network, respectively extracting features of different scales from the underground cable regional image and the differential Mask image; The feature fusion refers to the multi-scale fusion feature module fusing features of different scales.
10. The method for detecting abnormal damage of underground cables according to claim 9, characterized in that: The foreign body damage area mark is included in the output Mask image, the foreign body damage area is marked in white, and the non-damaged area is marked in black.