A Transmission Line Monitoring Method and Device Based on Machine Vision

Abstract

The present invention discloses a transmission line monitoring method and device based on machine vision, which relates to the technical field of machine vision; obtain a line image containing the target transmission line, and preprocess the line image to obtain an initial image; perform edge enhancement on the initial image to obtain a first enhanced image; determine a target foreground image that only contains the target transmission line according to the line image and the first enhanced image; perform boundary enhancement on the target foreground image through a preset algorithm to obtain a second enhanced image; substitute the second enhanced image into a preset model to obtain target features, and search a preset database according to the target features to determine the line state of the target transmission line. By performing multiple enhancements on the image, the key features of the target transmission line can be highlighted, noise and interference can be reduced, thereby improving the accuracy of target recognition. Through image processing and a preset model, the transmission line can be identified in detail, external interference caused by snow accumulation and icing can be reduced, and the line monitoring efficiency is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of machine vision, and particularly relates to a transmission line monitoring method and device based on machine vision. Background Art

[0002] With the continuous expansion of the power grid scale, the safe operation of transmission lines is of crucial importance. The snowfall weather in winter poses a serious threat to the normal operation of transmission lines. Snow accumulation and icing may lead to problems such as line breakage, insulator damage, and conductor overload. Currently, the inspection of transmission lines mainly relies on manual inspection, but manual inspection has problems such as heavy workload, low efficiency, and susceptibility to environmental factors under adverse weather conditions, making it difficult to achieve efficient and intelligent monitoring.

[0003] Patent CN111402248B discloses a method for detecting conductor defects in transmission lines based on machine vision, including: obtaining on-site image data and making a training data set; constructing and training an instance segmentation network to obtain a prediction model; deriving the input picture by the model to obtain the rectangular area image and binary mask image of the conductor; using a skeleton algorithm to extract the conductor skeleton and calculate the average width of the conductor, and reconstructing the binary mask image; using a homomorphic filtering algorithm to eliminate the influence of uneven illumination in the rectangular area image, and combining the reconstructed binary mask image to extract the segmented conductor area image; generating a large number of rectangular frames on the conductor area for screening; making a classification training data set, constructing and training a shallow classification network to obtain a classification prediction model; inputting the conductor segment area picture into the classification prediction model to count the defect types and defect proportions of the conductor segment state. The present invention can accurately segment the conductor and segmentally detect the state of the conductor, and judge the conductor defect type and defect degree. However, this solution is not accurate enough when facing lines with ice or snow, resulting in low line monitoring efficiency. Summary of the Invention

[0004] The object of the present invention is to solve the problems, and to propose a transmission line monitoring method and device based on machine vision.

[0005] In the first aspect of the implementation of the present invention, a transmission line monitoring method based on machine vision is first proposed. The method includes:

[0006] Obtaining a line image including a target transmission line, and preprocessing the line image to obtain an initial image;

[0007] Performing edge enhancement on the initial image to obtain a first enhanced image;

[0008] Determining a target foreground image that only includes the target transmission line according to the line image and the first enhanced image;

[0009] Performing boundary enhancement on the target foreground image through a preset algorithm to obtain a second enhanced image;

[0010] Substituting the second enhanced image into a preset model to obtain target features, and determining the line state of the target transmission line by searching a preset database according to the target features.

[0011] Optionally, preprocessing the line image to obtain an initial image includes:

[0012] Performing non-local means denoising on the line image and then performing resolution reconstruction to obtain a first preset image;

[0013] Performing adaptive local contrast enhancement on the first preset image to obtain a second preset image;

[0014] Performing gray conversion on the second preset image and then performing multi-scale bilateral filtering to obtain an initial image.

[0015] Optionally, determining a target foreground image that only contains the target transmission line according to the line image and the first enhanced image includes:

[0016] Performing multi-Otsu threshold and connected graph analysis on the first enhanced image to obtain a target recognition region;

[0017] Cutting the line image according to the target recognition region to obtain a target foreground image.

[0018] Optionally, performing boundary enhancement on the target foreground image through a preset algorithm to obtain a second enhanced image includes:

[0019] Performing gray conversion on the target foreground image to obtain a target gray image, and performing gradient morphological operations on the target gray image to obtain a target edge image;

[0020] Performing non-maximum suppression on the target edge image to obtain a target fusion image;

[0021] Performing weighted fusion on the target foreground image and the target fusion image to obtain a second enhanced image.

[0022] Optionally, substituting the second enhanced image into a preset model to obtain target features includes:

[0023] The second enhanced image passes through a first convolution module and a second convolution module to obtain a first convolution feature, and the first convolution feature passes through a first C2f module to obtain a first extraction feature;

[0024] The first extracted feature image passes through a third convolutional module to obtain a second convolutional feature, and the first convolutional feature and the second convolutional feature are weighted and fused to obtain a first fused feature. The first fused feature passes through a second C2f module to obtain a second extracted feature;

[0025] The second extracted feature passes through a first attention module to obtain a first attention feature;

[0026] The second extracted feature passes through a fourth convolutional module to obtain a third convolutional feature, and the second convolutional feature and the third convolutional feature are weighted and fused to obtain a second fused feature. The second fused feature passes through a third C2f module to obtain a third extracted feature;

[0027] The third extracted feature passes through a second attention module to obtain a second attention feature;

[0028] The third extracted feature passes through a feature enhancement module to obtain a first enhanced feature;

[0029] The first attention feature, the second attention feature, and the first enhanced feature are weighted and fused to obtain a target feature.

[0030] In the second aspect of the implementation of the present invention, a transmission line monitoring device based on machine vision is proposed, including:

[0031] A preprocessing module for acquiring a line image containing a target transmission line and preprocessing the line image to obtain an initial image;

[0032] An edge enhancement module for performing edge enhancement on the initial image to obtain a first enhanced image;

[0033] A target foreground image determination module for determining a target foreground image containing only the target transmission line according to the line image and the first enhanced image;

[0034] A boundary enhancement module for performing boundary enhancement on the target foreground image through a preset algorithm to obtain a second enhanced image;

[0035] A target feature determination module for substituting the second enhanced image into a preset model to obtain a target feature and determining the line state of the target transmission line by searching a preset database according to the target feature.

[0036] Optionally, the preprocessing module includes:

[0037] A resolution reconstruction module for performing non-local mean denoising on the line image and then performing resolution reconstruction to obtain a first preset image;

[0038] A contrast enhancement module for adaptively enhancing the local contrast of the first preset image to obtain a second preset image;

[0039] A grayscale conversion module for grayscale converting the second preset image and then performing multi-scale bilateral filtering to obtain an initial image.

[0040] Optionally, the target foreground image determination module includes:

[0041] A target recognition region determination module for performing multi-Otsu thresholding and connected graph analysis on the first enhanced image to obtain a target recognition region;

[0042] A region cutting module for cutting the line image according to the target recognition region to obtain a target foreground image.

[0043] Optionally, the boundary enhancement module includes:

[0044] A target edge image determination module for grayscale converting the target foreground image to obtain a target grayscale image and performing gradient morphological operations on the target grayscale image to obtain a target edge image;

[0045] A target fusion image determination module for performing non-maximum suppression on the target edge image to obtain a target fusion image;

[0046] A first weighted fusion module for weighted fusing the target foreground image and the target fusion image to obtain a second enhanced image.

[0047] Optionally, the target feature determination module includes:

[0048] A first extracted feature determination module for obtaining a first convolutional feature when the second enhanced image passes through a first convolutional module and a second convolutional module, and obtaining a first extracted feature when the first convolutional feature passes through a first C2f module;

[0049] A second extracted feature determination module for obtaining a second convolutional feature when the first extracted feature image passes through a third convolutional module, weighted fusing the first convolutional feature and the second convolutional feature to obtain a first fusion feature, and obtaining a second extracted feature when the first fusion feature passes through a second C2f module;

[0050] A first attention feature determination module for obtaining a first attention feature when the second extracted feature passes through a first attention module;

[0051] A third extracted feature determination module for obtaining a third convolutional feature when the second extracted feature passes through a fourth convolutional module, weighted fusing the second convolutional feature and the third convolutional feature to obtain a second fusion feature, and obtaining a third extracted feature when the second fusion feature passes through a third C2f module;

[0052] The second attention feature determination module is configured to obtain a second attention feature from the third extracted feature through a second attention module;

[0053] The first enhanced feature determination module is configured to obtain a first enhanced feature from the third extracted feature through a feature enhancement module;

[0054] The second weighted fusion module is configured to perform weighted fusion on the first attention feature, the second attention feature, and the first enhanced feature to obtain a target feature.

[0055] Advantages of the present invention:

[0056] The present invention proposes a transmission line monitoring method based on machine vision, which acquires a line image containing a target transmission line, preprocesses the line image to obtain an initial image; performs edge enhancement on the initial image to obtain a first enhanced image; determines a target foreground image containing only the target transmission line according to the line image and the first enhanced image; performs boundary enhancement on the target foreground image through a preset algorithm to obtain a second enhanced image; substitutes the second enhanced image into a preset model to obtain a target feature, and determines the line state of the target transmission line by searching a preset database according to the target feature. By performing multiple enhancements on the image, the key features of the target transmission line can be highlighted, noise and interference can be reduced, thereby improving the accuracy of target recognition. Through image processing and a preset model, the transmission line can be identified in detail, external interference caused by snow and ice can be reduced, and the line monitoring efficiency can be improved. Description of the Drawings

[0057] The following further describes the present invention with reference to the drawings.

[0058] Figure 1 It is a flowchart of a transmission line monitoring method based on machine vision provided by an embodiment of the present invention;

[0059] Figure 2 It is a schematic structural diagram of a transmission line monitoring device based on machine vision provided by an embodiment of the present invention. Detailed Embodiments

[0060] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. The term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and B can represent: A exists alone, A and B exist simultaneously, and B exists alone. In addition, in the present invention, descriptions such as "first" and "second" are only for descriptive purposes, and cannot be understood as indicating or implying their relative importance or implicitly indicating the quantity of the indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include at least one of the features. In addition, the technical solutions between the various embodiments can be combined with each other, but it must be based on what can be achieved by those of ordinary skill in the art. When the combination of technical solutions is contradictory or cannot be achieved, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0061] All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0062] The embodiments of the present invention provide a transmission line monitoring method based on machine vision. See Figure 1 , Figure 1 which is a flowchart of a transmission line monitoring method based on machine vision provided by the embodiments of the present invention. The method includes the following steps:

[0063] S101, obtain a line image including the target transmission line, and preprocess the line image to obtain an initial image;

[0064] S102, perform edge enhancement on the initial image to obtain a first enhanced image;

[0065] S103, determine a target foreground image that only includes the target transmission line according to the line image and the first enhanced image;

[0066] S104, perform boundary enhancement on the target foreground image through a preset algorithm to obtain a second enhanced image;

[0067] S105, substitute the second enhanced image into a preset model to obtain target features, and find the line state of the target transmission line by querying a preset database.

[0068] A machine vision-based transmission line monitoring method provided in an embodiment of the present invention can highlight the key features of the target transmission line and reduce noise and interference by performing multiple enhancements on the image, thereby improving the accuracy of target recognition. Through image processing and preset models, the transmission line can be identified in detail, reducing external interference caused by snow and ice accumulation, thereby improving the line monitoring efficiency.

[0069] In one implementation, preprocessing can remove noise and interference in the image and improve the quality of the image. In complex weather conditions such as snow, fog, low light, etc., preprocessing can enhance the usability of the image and ensure that the lines and key features in the image are still clearly presented even in unideal environments.

[0070] In one implementation, the line image containing the target transmission line is an RGB image taken by a drone. The preset database is used to save different line states and corresponding features. If the similarity between two features is greater than a similarity threshold, the two features are considered to be the same. The similarity threshold is determined by a technician.

[0071] In one implementation, edge enhancement can improve the recognizability of details in an image, especially under visually complex conditions such as snow cover or ice, helping the system to accurately identify line anomalies; improving the clarity of important boundaries and contours in the image helps reduce the impact of background noise and irrelevant information on the results, thereby reducing the probability of misidentification.

[0072] In one implementation, by combining the initial image with the enhanced image, the target transmission line can be accurately extracted, eliminating background interference or non-target areas. After the target area is extracted, subsequent image processing can focus more on key content, reducing computational complexity and improving processing efficiency and accuracy.

[0073] In one implementation, the boundary information in the image is further strengthened by enhancing the boundaries of the target foreground image, making the precise shape of the line and structural anomalies such as fractures or damages clearer, and enabling accurate identification of line details. In complex environments such as snow and ice, the enhanced boundaries can better help the system identify parts of the line that may be covered or damaged, thereby improving its ability to adapt to severe weather.

[0074] In one implementation, by comparing historical data or standardized models in the database, the health of the transmission line can be accurately determined and potential problems can be quickly identified.

[0075] In one embodiment, preprocessing the line image to obtain the initial image includes:

[0076] Performing non-local mean denoising on the line image and then reconstructing the resolution to obtain a first preset image;

[0077] Adaptive local contrast enhancement is performed on the first preset image to obtain the second preset image;

[0078] After grayscale conversion of the second preset image, multi-scale bilateral filtering is performed to obtain the initial image.

[0079] In one implementation, non-local means denoising is an information denoising method based on the overall similarity of the image, which can effectively reduce the noise in the image while retaining the details of the image. By processing the entire image, the noise impact of a single local area is avoided.

[0080] In one implementation, through resolution reconstruction, the clarity of the image can be improved. Especially for fine line structures and potential problems such as breaks and snow accumulation, it helps to observe details more precisely and enhance the effect of subsequent processing.

[0081] In one implementation, the non-local means denoising of the line image is specifically as follows: First, a search window (an area of size S×S) is defined. All pixel points similar to pixel i are found in this window. Taking pixel i as the center, a small area Patch of size f×f (the size of the small area is smaller than the search window) is taken. For each candidate pixel i in the search window, a Patch of the same size is also taken for comparison. The Euclidean distance between the two Patches is calculated, and the Gaussian kernel function is used to convert the Euclidean distance into a weight. Finally, pixel i is denoised through weighted averaging.

[0082] In one implementation, the specific method of adaptive local contrast enhancement is as follows: The first preset image is converted to the LAB color space, the L channel is enhanced, the image is divided into multiple small sub-blocks (such as 8×8), histogram equalization is performed on each area separately, the pixel intensity distribution in each small area is statistically calculated, its cumulative distribution function is calculated, and then the transition between different areas is smoothed through bilinear interpolation. Then, the processed image is converted back to the RGB image to obtain the second preset image; by enhancing the L channel and then enhancing the contrast for different areas, local details will not be lost as in global histogram equalization. Using bilinear interpolation smoothing can be used for the transition between different areas to make the image more natural.

[0083] In one implementation, the resolution reconstruction is performed through super-resolution reconstruction (such as bicubic interpolation, deep learning SRGAN, ESRGAN, etc.).

[0084] In one implementation, local contrast enhancement can improve the detail contrast within a local area. Especially for low-contrast areas such as snow-covered wires, it helps to highlight the features of these areas and improve the visibility of details. The adaptive method can dynamically adjust according to the contrast requirements of different regions in the image, avoiding over-enhancement or under-enhancement, and ensuring a balanced improvement in image quality. After enhancing the contrast, the key elements in the image, namely the wires and ice and snow accumulation, become more prominent, facilitating subsequent analysis and problem detection.

[0085] In one implementation, grayscale conversion helps to remove the color information of the image and simplify the processing process. Especially when the color information is not a key factor, the grayscale image can highlight the brightness and texture details, making the lines and structures clearer. The grayscale image can strengthen the line structure through brightness contrast, avoiding color interference, and helping to focus on the shape and structure features of the image during subsequent processing.

[0086] In one embodiment, determining the target foreground image that only contains the target transmission line according to the line image and the first enhanced image includes:

[0087] Performing multi-Otsu thresholding and connected graph analysis on the first enhanced image to obtain the target recognition region;

[0088] Cutting the line image according to the target recognition region to obtain the target foreground image.

[0089] In one implementation, under variable weather conditions such as ice and snow, low light, haze, etc., the gray distribution of the image may be relatively complex. The multiple segmentation of the Otsu threshold can be automatically adjusted according to the different gray levels of the image, improving the adaptability and accuracy of target recognition. Using multiple thresholds for segmentation can more accurately identify the target region. Especially in complex images, there may be multiple target regions at different levels such as the background, wires, snow accumulation, etc. This method can enhance the segmentation effect of the image and avoid information loss caused by a single threshold.

[0090] In one implementation, connected graph analysis can help identify the connected regions in the image and exclude irrelevant scattered noise. For the transmission line image, connectivity analysis can effectively identify the line regions that are connected in a line, ignoring irrelevant debris or background information. By cutting the image according to the target recognition region, the region that only contains the target transmission line can be accurately extracted, removing the background and other irrelevant parts. In this way, subsequent image processing and analysis can focus on the key content, improving the processing efficiency and accuracy.

[0091] In one embodiment, performing boundary enhancement on the target foreground image through a preset algorithm to obtain the second enhanced image includes:

[0092] The target foreground image is converted to grayscale to obtain the target grayscale image, and the target grayscale image is subjected to gradient morphological operations to obtain the target edge image;

[0093] The target edge image is subjected to non-maximum suppression to obtain the target fusion image;

[0094] The target foreground image and the target fusion image are weighted and fused to obtain the second enhanced image.

[0095] In one implementation, gradient morphological operations can emphasize the edge information of the image. Especially for structures with obvious edges such as transmission lines, it can highlight the details of the edge part and help subsequent detection and analysis; non-maximum suppression can remove non-edge pixels from the edge image, making the edges in the image more detailed and clear, which can help to make the edge lines of the transmission line more prominent and provide a clearer contour for subsequent defect analysis and fault diagnosis.

[0096] In one implementation, the specific operation of non-maximum suppression is to calculate the gradient using the Sobel operator, calculate the gradient direction, suppress non-maximum values along the gradient direction, only retain the local maximum gradient value, remove adjacent non-maximum pixels, and finally output the target fusion image; the target foreground image and the target fusion image are added and divided by 2 to obtain the second enhanced image.

[0097] In one implementation, through non-maximum suppression, redundant information in the edge region can be removed, making the edges of the image clearer and more accurate, and avoiding the influence of errors and noise on subsequent analysis.

[0098] In one implementation, weighted fusion combines the target foreground image and the target fusion image, taking into account the advantages of both. The foreground image provides the overall information of the line, and the fusion image provides more detailed edge information, generating a clearer and more complete image, which helps subsequent tasks such as line status judgment and fault detection.

[0099] In one embodiment, substituting the second enhanced image into a preset model to obtain the target features includes:

[0100] The second enhanced image passes through the first convolutional module and the second convolutional module to obtain the first convolutional feature, and the first convolutional feature passes through the first C2f module to obtain the first extraction feature;

[0101] The first extraction feature passes through the third convolutional module to obtain the second convolutional feature, the first convolutional feature and the second convolutional feature are weighted and fused to obtain the first fusion feature, and the first fusion feature passes through the second C2f module to obtain the second extraction feature;

[0102] The second extraction feature passes through the first attention module to obtain the first attention feature;

[0103] The second extracted feature passes through the fourth convolutional module to obtain the third convolutional feature. The second convolutional feature and the third convolutional feature are weighted and fused to obtain the second fused feature. The second fused feature passes through the third C2f module to obtain the third extracted feature;

[0104] The third extracted feature passes through the second attention module to obtain the second attention feature;

[0105] The third extracted feature passes through the feature enhancement module to obtain the first enhanced feature;

[0106] The first attention feature, the second attention feature, and the first enhanced feature are weighted and fused to obtain the target feature.

[0107] In one implementation, the convolution operation extracts local features to help obtain effective information from the original image, which can effectively enhance the local features in the input image, enabling the model to better capture details; through multiple convolutional modules, the model can gradually extract multi-level features in the image, from low-level features to high-level features, providing richer information for subsequent analysis.

[0108] In one implementation, the introduction of the C2f (Convolutional to Feature) module enables the features extracted by convolution to be further optimized and extracted, which can reduce the interference of irrelevant information and improve the expression ability of the features. The first C2f module further refines the features to ensure that the model can focus on the most important parts when processing images, remove redundant information, and make the features more compact and effective.

[0109] In one implementation, when the features are weighted and fused to obtain the fused feature, they are all aligned based on the maximum size and then fused, and the fusion is to take the average.

[0110] In one implementation, the first convolution module in the solution is (kernel size: 3×3, stride: 2, number of output channels: 64), the second convolution module (kernel size: 3×3, stride: 2, number of output channels: 128), the third convolution module (kernel size: 3×3, stride: 2, number of output channels: 256), and the fourth convolution module (kernel size: 3×3, stride: 2, number of output channels: 512); the first C2f module, the second C2f module, and the third C2f module are common modules in YOLOv8; the first attention module and the second attention module are the same. The attention module consists of three branches, and each branch is responsible for calculating the weighted feature maps in the three dimensions of width, length, and channels respectively. In each branch, the feature map is first rotated, then processed through a pooling layer and a convolutional layer, and then the attention weights are obtained through the Sigmoid activation function. Then, the rotated feature map is weighted using the attention weights obtained from the three branches and restored to the original shape of the input feature map. Finally, the final output feature map is obtained by averaging the weighted feature maps of the three branches; the feature enhancement module includes a backbone network (ResNet + deformable convolution), a C2f module, a convolutional layer, a pooling layer, a fully connected layer, and an output layer, where (ResNet + deformable convolution), the convolutional layer, the pooling layer, the fully connected layer, and the output layer are the model parameters of the deformable convolutional network.

[0111] In one implementation, the second convolution processing further enhances the feature expression ability, especially in complex images, it can capture deeper feature information and enhance the accuracy of subsequent processing; through multiple convolutions, the model can more comprehensively understand the details of the image, thereby improving the robustness of the model and its adaptability to complex situations.

[0112] In one implementation, weighted fusion of the features of different convolution modules can combine their respective advantages, fuse low-level and high-level features, and enhance the expressiveness of information. Such fusion can ensure that the useful information extracted by each module can be effectively retained; by weighted fusion of different features, the biases or deficiencies that may exist in a single module can be eliminated, making the final features more comprehensive and accurate.

[0113] In one implementation, the second C2f module further processes the fused features, further removing redundant information and enhancing the discriminative ability of the features, which helps to extract more accurate and key features in subsequent analysis.

[0114] In one implementation, the attention mechanism can dynamically adjust the attention to features according to the importance of different regions, enabling the model to focus on more important parts of the image, such as power lines, snow-covered areas, etc., and improving the recognition accuracy of the model. By introducing the attention module, the model can more accurately process images in complex environments, such as weather interference like snow and frost, and enhance its sensitivity to key features.

[0115] In one implementation, the fourth convolutional module continues to strengthen the feature extraction of the image, especially in more complex scenarios, further capturing details and enhancing the expression ability of the model. Through continuous convolutional modules, the model can extract features at different levels, enabling complex images to be fully analyzed and helping subsequent analysis to be more accurate.

[0116] In one implementation, by weighted fusion of the second convolutional feature and the third convolutional feature, the model can integrate the advantages of both, eliminate the deficiencies, improve the accuracy and expression ability of the features. The third C2f module further refines the fused features, removes redundant information, retains the most representative parts, and improves the discriminative ability and expressiveness of the features.

[0117] In one implementation, through the second attention module, the model can further focus on the most important features in the image, especially for minor changes or key regions of the target, improving the accuracy of target detection. Through the processing of the enhancement module, the model can maintain high accuracy in various complex situations and improve the detection ability for complex faults.

[0118] In one implementation, by weighted fusion of multiple attention features and enhancement features, the model can integrate all the extracted useful information to ensure that the final target features contain all the key contents in the image. The weighted method of fusing different features can effectively improve the comprehensive performance of the model, enhance the robustness of the system in complex environments, reduce misjudgment, and improve the reliability of target detection.

[0119] Based on the same inventive concept, the embodiments of the present invention also provide a transmission line monitoring device based on machine vision. Refer to Figure 2 , Figure 2 which is a schematic structural diagram of a transmission line monitoring device based on machine vision provided by the embodiments of the present invention, including:

[0120] A preprocessing module, configured to obtain a line image including a target transmission line and preprocess the line image to obtain an initial image;

[0121] An edge enhancement module, configured to perform edge enhancement on the initial image to obtain a first enhanced image;

[0122] A target foreground image determination module, configured to determine a target foreground image that only includes a target transmission line according to a line image and a first enhanced image;

[0123] A boundary enhancement module, configured to perform boundary enhancement on the target foreground image through a preset algorithm to obtain a second enhanced image;

[0124] A target feature determination module, configured to substitute the second enhanced image into a preset model to obtain target features, and search a preset database according to the target features to determine the line state of the target transmission line.

[0125] Based on a transmission line monitoring device based on machine vision provided by an embodiment of the present invention, by performing multiple enhancements on an image, key features of the target transmission line can be highlighted, noise and interference can be reduced, thereby improving the accuracy of target recognition. Through image processing and a preset model, the transmission line can be identified in detail, external interference caused by snow and ice can be reduced, and the line monitoring efficiency is improved.

[0126] In one embodiment, the preprocessing module includes:

[0127] A resolution reconstruction module, configured to perform non-local means denoising on the line image and then perform resolution reconstruction to obtain a first preset image;

[0128] A contrast enhancement module, configured to perform adaptive local contrast enhancement on the first preset image to obtain a second preset image;

[0129] A grayscale conversion module, configured to perform grayscale conversion on the second preset image and then perform multi-scale bilateral filtering to obtain an initial image.

[0130] In one embodiment, the target foreground image determination module includes:

[0131] A target recognition area determination module, configured to perform multi-Otsu threshold and connected graph analysis on the first enhanced image to obtain a target recognition area;

[0132] An area cutting module, configured to cut the line image according to the target recognition area to obtain a target foreground image.

[0133] In one embodiment, the boundary enhancement module includes:

[0134] A target edge image determination module, configured to perform grayscale conversion on the target foreground image to obtain a target grayscale image, and perform gradient morphological operations on the target grayscale image to obtain a target edge image;

[0135] A target fusion image determination module, configured to perform non-maximum suppression on the target edge image to obtain a target fusion image;

[0136] The first weighted fusion module is used to perform weighted fusion on the target foreground image and the target fusion image to obtain a second enhanced image.

[0137] In one embodiment, the target feature determination module includes:

[0138] The first extracted feature determination module is used to obtain a first convolutional feature when the second enhanced image passes through the first convolutional module and the second convolutional module, and the first convolutional feature passes through the first C2f module to obtain a first extracted feature;

[0139] The second extracted feature determination module is used to obtain a second convolutional feature when the first extracted feature passes through the third convolutional module, the first convolutional feature and the second convolutional feature are weighted and fused to obtain a first fused feature, and the first fused feature passes through the second C2f module to obtain a second extracted feature;

[0140] The first attention feature determination module is used to obtain a first attention feature when the second extracted feature passes through the first attention module;

[0141] The third extracted feature determination module is used to obtain a third convolutional feature when the second extracted feature passes through the fourth convolutional module, the second convolutional feature and the third convolutional feature are weighted and fused to obtain a second fused feature, and the second fused feature passes through the third C2f module to obtain a third extracted feature;

[0142] The second attention feature determination module is used to obtain a second attention feature when the third extracted feature passes through the second attention module;

[0143] The first enhanced feature determination module is used to obtain a first enhanced feature when the third extracted feature passes through the feature enhancement module;

[0144] The second weighted fusion module is used to perform weighted fusion on the first attention feature, the second attention feature and the first enhanced feature to obtain the target feature.

[0145] The above has described an embodiment of the present invention in detail, but the content is only a preferred embodiment of the present invention and cannot be considered as limiting the scope of implementation of the present invention. All equivalent changes and improvements made according to the scope of the present invention application shall still fall within the scope covered by the patent of the present invention.

Claims

1. A transmission line monitoring method based on machine vision, characterized in that The method includes: Obtain a line image including the target transmission line, and preprocess the line image to obtain an initial image; the target transmission line includes the line itself and the ice and snow covering it; Perform edge enhancement on the initial image to obtain a first enhanced image; Determine a target foreground image that only includes the target transmission line according to the line image and the first enhanced image; Perform boundary enhancement on the target foreground image through a preset algorithm to obtain a second enhanced image; Substitute the second enhanced image into a preset model to obtain target features, and search a preset database according to the target features to determine the line state of the target transmission line; Performing boundary enhancement on the target foreground image through a preset algorithm to obtain a second enhanced image includes: Perform gray conversion on the target foreground image to obtain a target gray image, and perform gradient morphological operations on the target gray image to obtain a target edge image; Perform non-maximum suppression on the target edge image to obtain a target fusion image; Perform weighted fusion on the target foreground image and the target fusion image to obtain a second enhanced image; Substituting the second enhanced image into a preset model to obtain target features includes: The second enhanced image passes through a first convolutional module and a second convolutional module to obtain a first convolutional feature, and the first convolutional feature passes through a first C2f module to obtain a first extracted feature; The first extracted feature passes through a third convolutional module to obtain a second convolutional feature, the first convolutional feature and the second convolutional feature are weighted and fused to obtain a first fusion feature, and the first fusion feature passes through a second C2f module to obtain a second extracted feature; The second extracted feature passes through a first attention module to obtain a first attention feature; The second extracted feature passes through a fourth convolutional module to obtain a third convolutional feature, the second convolutional feature and the third convolutional feature are weighted and fused to obtain a second fusion feature, and the second fusion feature passes through a third C2f module to obtain a third extracted feature; The third extracted feature passes through a second attention module to obtain a second attention feature; The third extracted feature passes through a feature enhancement module to obtain a first enhanced feature; Perform weighted fusion on the first attention feature, the second attention feature and the first enhanced feature to obtain target features.

2. The method for monitoring a transmission line based on machine vision according to claim 1, wherein Preprocessing the line image to obtain an initial image includes: Perform non-local mean denoising on the line image and then perform resolution reconstruction to obtain a first preset image; Perform adaptive local contrast enhancement on the first preset image to obtain a second preset image; Perform gray conversion on the second preset image and then perform multi-scale bilateral filtering to obtain an initial image.

3. The method for monitoring a transmission line based on machine vision according to claim 1, characterized in that, Determining a target foreground image that only includes the target transmission line according to the line image and the first enhanced image includes: Perform multi-Otsu threshold and connected graph analysis on the first enhanced image to obtain a target recognition region; Cut the line image according to the target recognition region to obtain a target foreground image.

4. A transmission line monitoring device based on machine vision, characterized in that, The device includes: A preprocessing module for obtaining a line image including a target transmission line and preprocessing the line image to obtain an initial image; the target transmission line includes the line itself and the ice and snow covering it. An edge enhancement module for performing edge enhancement on the initial image to obtain a first enhanced image. A target foreground image determination module for determining a target foreground image that only includes the target transmission line according to the line image and the first enhanced image. A boundary enhancement module for performing boundary enhancement on the target foreground image through a preset algorithm to obtain a second enhanced image. A target feature determination module for substituting the second enhanced image into a preset model to obtain target features, and searching a preset database according to the target features to determine the line state of the target transmission line. The boundary enhancement module includes: A target edge image determination module for performing gray conversion on the target foreground image to obtain a target gray image, and performing gradient morphological operations on the target gray image to obtain a target edge image. A target fusion image determination module for performing non-maximum suppression on the target edge image to obtain a target fusion image. A first weighted fusion module for performing weighted fusion on the target foreground image and the target fusion image to obtain a second enhanced image. The target feature determination module includes: A first extraction feature determination module for obtaining a first convolution feature after the second enhanced image passes through a first convolution module and a second convolution module, and obtaining a first extraction feature after the first convolution feature passes through a first C2f module. A second extraction feature determination module for obtaining a second convolution feature after the first extraction feature passes through a third convolution module, performing weighted fusion on the first convolution feature and the second convolution feature to obtain a first fusion feature, and obtaining a second extraction feature after the first fusion feature passes through a second C2f module. A first attention feature determination module for obtaining a first attention feature after the second extraction feature passes through a first attention module. A third extraction feature determination module for obtaining a third convolution feature after the second extraction feature passes through a fourth convolution module, performing weighted fusion on the second convolution feature and the third convolution feature to obtain a second fusion feature, and obtaining a third extraction feature after the second fusion feature passes through a third C2f module. A second attention feature determination module for obtaining a second attention feature after the third extraction feature passes through a second attention module. A first enhanced feature determination module for obtaining a first enhanced feature after the third extraction feature passes through a feature enhancement module. A second weighted fusion module for performing weighted fusion on the first attention feature, the second attention feature and the first enhanced feature to obtain target features.

5. The monitoring device for transmission lines based on machine vision according to claim 4, characterized in that, The preprocessing module includes: A resolution reconstruction module for performing non-local mean denoising on the line image and then performing resolution reconstruction to obtain a first preset image. A contrast enhancement module for performing adaptive local contrast enhancement on the first preset image to obtain a second preset image. A gray conversion module for performing gray conversion on the second preset image and then performing multi-scale bilateral filtering to obtain an initial image.

6. The monitoring device for transmission lines based on machine vision according to claim 4, characterized in that, The target foreground image determination module includes: A target recognition area determination module, configured to perform multi-Otsu thresholding and connected graph analysis on the first enhanced image to obtain a target recognition area; An area cutting module, configured to cut the line image according to the target recognition area to obtain a target foreground image.

Images