Aerial cable broken strand defect restoration recognition positioning method and system based on semantic segmentation

Abstract

The application discloses a method and system for recognizing and positioning broken strand defects of a ground wire based on semantic segmentation, and the method comprises the following steps: acquiring an image of a broken strand defect area in a process of performing a broken strand defect recovery of a live-line work robot; inputting the image of the broken strand defect area into a pre-trained model for recognizing and positioning the broken strand defects of the ground wire to perform semantic segmentation and obtain a semantic segmentation result image, wherein the semantic segmentation result image comprises a whole ground wire, a single broken strand ground wire and a gap formed between the single broken strand ground wire and the whole ground wire; extracting the gap formed between the single broken strand ground wire and the whole ground wire from the semantic segmentation result image and calculating geometric features of the gap; and judging whether the single broken strand ground wire has been completely rolled back into the gap according to the geometric features of the gap. The application aims to realize high-precision and high-accuracy recognition and positioning of the broken strand defects of the ground wire of a high-voltage transmission line.

Description

Technical Field

[0001] This invention relates to the field of conductor and ground wire repair technology and image recognition and understanding technology in power systems, specifically to a method and system for identifying and locating conductor and ground wire strand breakage defects based on semantic segmentation. Background Technology

[0002] High-voltage transmission lines are mostly erected in the field, making them susceptible to natural weather conditions such as rain, snow, and lightning strikes. This can lead to broken or scattered strands in the transmission lines, jeopardizing the safe operation of power transmission. Traditional manual repairs require workers to climb towers or use gantry cranes to reach the broken strands, which is labor-intensive and dangerous. However, with the continuous improvement of production automation, robots are gradually replacing manual labor in various live-line operations, especially in high-risk and labor-intensive situations. The widespread application of live-line working robots can greatly improve the efficiency and quality of live-line work, and is one of the practical methods to solve the contradictory problem of power supply reliability and the safety of manual live-line maintenance. However, currently, live-line working robots cannot accurately identify and locate the broken strands in the conductor for repair. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide a method and system for identifying and locating conductor and ground wire strand breakage defects based on semantic segmentation, in order to address the above-mentioned problems in the prior art. The present invention aims to achieve high-precision and high-accuracy identification and location of conductor and ground wire strand breakage defects in high-voltage transmission lines.

[0004] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A semantic segmentation-based method for identifying and locating conductor strand defects during restoration includes the following steps: acquiring an image of the strand defect area during the restoration process of a live-line working robot; inputting the strand defect area image into a pre-trained conductor strand defect restoration identification and location model for semantic segmentation to obtain a semantic segmentation result image, wherein the semantic segmentation result image includes the segmented whole conductor, single strand conductor, and the gap formed between the single strand conductor and the whole conductor; extracting the gap formed between the single strand conductor and the whole conductor from the semantic segmentation result image and calculating the gap geometric features; and determining whether the single strand conductor has been completely pulled back into the gap based on the gap geometric features.

[0005] Optionally, the regions of the entire conductor, the single broken conductor, and the gap formed between the single broken conductor and the entire conductor in the semantic segmentation result image are filled with a different color other than black, while the remaining regions are filled with black.

[0006] Optionally, the step of extracting the gap formed between a single broken strand of wire and the entire wire from the semantic segmentation result image and calculating the gap geometric features includes: converting the semantic segmentation result image from RGB color space to HSV color space to obtain an HSV image; separating the three color channels of the HSV image; extracting the color components of the gap formed between the single broken strand of wire and the entire wire from the separated specified color channels; converting the extracted color components back to RGB color space to obtain an independent gap segmentation result image; binarizing the gap segmentation result image; performing contour detection on the binarized image; and calculating the area occupied by the pixels within the detected contour to obtain the gap geometric features, wherein the gap geometric features are part or all of the area, length, and width of the gap.

[0007] Optionally, the binarization of the gap segmentation result image includes: converting the gap segmentation result image to grayscale to obtain a grayscale image; dividing the pixels in the grayscale image into target pixels and background pixels according to a given initial threshold T0; and calculating the proportion of the number of target pixels to the number of background pixels in the grayscale image. The proportion of background pixels to grayscale image : ; ; in, The proportion of the target pixel's pixels to the grayscale image. The proportion of background pixels to the grayscale image; N Less than the initial threshold T The number of background pixels is 0; M The number of target pixels that are greater than the initial threshold T0; A This represents the number of pixels in the horizontal direction of the grayscale image. B This represents the number of pixels in the vertical direction of the grayscale image. A × B Let be the total number of pixels in the grayscale image; calculate the average grayscale value of each grayscale image. Average grayscale value of target pixels and the average grayscale value of background pixels : ; ;

[0008] in, The grayscale value is i The number of pixels, i ∈[0,255]; based on the proportion of the number of pixels of the target pixel to the grayscale image. The proportion of background pixels to grayscale images Average gray level of a grayscale image Average grayscale value of target pixels and the average grayscale value of background pixels Calculate the segmentation threshold: ; in, The threshold is used for segmentation; based on the segmentation threshold, the color of each pixel in the grayscale image is assigned to either black or white, thus obtaining a binarized image.

[0009] Optionally, determining whether the single broken strand of the conductor has been completely pulled back into the gap based on the gap geometry features includes: determining whether the gap geometry features are less than a preset threshold; if they are less than the preset threshold, it is determined that the single broken strand of the conductor has been completely pulled back into the gap, and the conductor breakage defect restoration operation of the live-line working robot is terminated; otherwise, it is determined that the single broken strand of the conductor has not been completely pulled back into the gap, and the conductor breakage defect restoration operation is continued and the conductor breakage defect identification and positioning are performed.

[0010] Optionally, the conductor strand breakage defect restoration identification and localization model consists of an encoder and a decoder. The encoder includes a feature extraction module, a first attention mechanism module, a hole spatial pyramid pooling module, and a first 1×1 convolutional layer connected in sequence. The feature extraction module and the first 1×1 convolutional layer of the encoder output feature maps to the decoder, respectively. The decoder includes a second 1×1 convolutional layer, a connection module, a first upsampling module, a second attention mechanism module, a 3×3 convolutional layer, and a second upsampling module. The output feature map of the feature extraction module generates a first-path feature map after passing through the second 1×1 convolutional layer. The output feature map of the first 1×1 convolutional layer generates a second-path feature map after passing through the first upsampling module. The first-path feature map and the second-path feature map are concatenated along the dimension by the connection module and then sequentially passed through the second attention mechanism module, the 3×3 convolutional layer, and the second upsampling module to obtain the semantic segmentation result image.

[0011] Optionally, the feature extraction module includes a convolutional module, a bottleneck module, an average pooling module, and a fully connected layer module connected in sequence. The convolutional module includes a convolutional layer, a batch normalization layer, and an Hswish activation function layer connected in sequence. The bottleneck module includes a 1×1 upscaling convolutional layer, a 3×3 depthwise separable convolutional layer, a 1×1 downscaling convolutional layer, and an SE attention mechanism module connected in sequence. The fully connected layer module includes a fully connected layer, a batch normalization layer, and an Hswish activation function connected in sequence. The first attention mechanism module includes a global average pooling layer, an adaptive one-dimensional convolutional layer, and a normalization layer connected in sequence. The weight vector obtained by the normalization layer is multiplied by the input feature map of the first attention mechanism module to obtain the output feature map of the first attention mechanism module. The kernel size used by the adaptive one-dimensional convolutional layer is... The expression for the computation function is: ; in, Regarding channel dimension The mapping function, and These are preset parameters; To select the nearest odd number.

[0012] Optionally, the training of the conductor strand breakage defect restoration identification and localization model includes: acquiring image samples of the strand breakage defect area during the process of the live-line working robot performing conductor strand breakage defect restoration, to obtain an initial image dataset; labeling the strand breakage defect restoration image samples in the initial image dataset, labeling the entire conductor, a single strand breakage conductor, and the gap formed between the single strand breakage conductor and the entire conductor; expanding the labeled image dataset, including performing flipping, rotation, noise addition, brightness adjustment, contrast adjustment, random scaling according to a preset ratio range, and boundary filling with black pixels on the strand breakage defect restoration image samples in the initial image dataset to generate new strand breakage defect restoration image samples, thereby obtaining a training dataset composed of strand breakage defect image samples and their labels; constructing an initial conductor strand breakage defect restoration identification and localization model; and training the initial conductor strand breakage defect restoration identification and localization model using the training dataset and a preset loss function, thereby obtaining a trained conductor strand breakage defect restoration identification and localization model.

[0013] Optionally, the preset loss function is expressed as follows: ; ; ; in, L The default loss function is... LDice The similarity coefficient loss function, L Focal For focus loss function; t i This is the target value for the i-th category, and its value can be 0 or 1. y i Let be the predicted value for the i-th category, with a value range of (0,1). γ For smoothing coefficients; α i Let be the weight factor for the i-th category; As a regulating factor; Let be the predicted probability of the i-th category.

[0014] Furthermore, the present invention also provides a semantic segmentation-based conductor strand breakage defect restoration identification and location system, comprising a microprocessor and a memory interconnected thereto, wherein the microprocessor is programmed or configured to execute the semantic segmentation-based conductor strand breakage defect restoration identification and location method.

[0015] Compared with existing technologies, the present invention mainly achieves the following beneficial effects: The method of the present invention includes acquiring an image of the broken strand defect area during the restoration process of a live-line working robot performing conductor-to-ground wire broken strand defect restoration; inputting the broken strand defect area image into a pre-trained conductor-to-ground wire broken strand defect restoration identification and positioning model for semantic segmentation to obtain a semantic segmentation result image, wherein the semantic segmentation result image includes the segmented whole conductor, single broken strand conductor, and the gap formed between the single broken strand conductor and the whole conductor; extracting the gap formed between the single broken strand conductor and the whole conductor from the semantic segmentation result image and calculating the gap geometric features; and determining whether the single broken strand conductor has been completely pulled back into the gap based on the gap geometric features. The present invention, by using a conductor-to-ground wire broken strand defect restoration identification and positioning model to perform semantic segmentation of the broken strand defect area image, extracting the gap formed between the single broken strand conductor and the whole conductor from the semantic segmentation result image and calculating the gap geometric features, and determining whether the single broken strand conductor has been completely pulled back into the gap based on the gap geometric features, can achieve high-precision and high-accuracy conductor-to-ground wire broken strand defect identification and positioning for high-voltage transmission lines. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the basic process of the method in an embodiment of the present invention.

[0017] Figure 2 Examples of images and processing results of the broken-feet defect region in embodiments of the present invention are shown, where (a) is the original broken-feet defect region image, (b) is the semantic segmentation result image, and (c) is the binarized image.

[0018] Figure 3This is a schematic diagram of the network structure of the conductor strand breakage defect restoration identification and positioning model in an embodiment of the present invention.

[0019] Figure 4 This is a schematic diagram of the training process of the conductor strand breakage defect restoration identification and positioning model in an embodiment of the present invention.

[0020] Figure 5 This is an example of the experimental results of the conductor strand breakage defect restoration identification and localization model in an embodiment of the present invention, wherein (a1) is the original first image of the strand breakage defect area, (a2) is the first semantic segmentation result image, and (a3) ​​is the first binarized image. (b1) is the original second image of the strand breakage defect area, (b2) is the second semantic segmentation result image, and (b3) is the second binarized image. (c1) is the original third image of the strand breakage defect area, (c2) is the third semantic segmentation result image, and (c3) is the third binarized image. Detailed Implementation

[0021] To enable those skilled in the art to better understand the technical solutions of the present invention, the technical solutions of the present invention will be further described in detail below with reference to the accompanying drawings in the embodiments of the present invention.

[0022] like Figure 1 As shown, the method for identifying and locating conductor strand breakage defects based on semantic segmentation in this embodiment includes the following steps: acquiring an image of the strand breakage defect area during the conductor strand breakage defect restoration process performed by a live-line working robot; inputting the strand breakage defect area image into a pre-trained conductor strand breakage defect restoration identification and location model for semantic segmentation to obtain a semantic segmentation result image, wherein the semantic segmentation result image includes the segmented whole conductor, single strand breakage conductor, and the gap formed between the single strand breakage conductor and the whole conductor; extracting the gap formed between the single strand breakage conductor and the whole conductor from the semantic segmentation result image and calculating the gap geometric features; and determining whether the single strand breakage conductor has been completely pulled back into the gap based on the gap geometric features.

[0023] In this embodiment, the original image of the broken strand defect area is as follows: Figure 2 As shown in (a) in the figure, Figure 2 The image of the broken wire defect region shown in (a) is input into a pre-trained conductor broken wire defect restoration recognition and localization model for semantic segmentation to obtain the semantic segmentation result image as shown in Figure 1. Figure 2 As shown in (b) in this embodiment, the regions of the whole conductor, the single broken conductor, and the gap formed between the single broken conductor and the whole conductor in the semantic segmentation result image are filled with different colors other than black, while the remaining regions are filled with black.

[0024] In this embodiment, the step of extracting the gap formed between a single broken strand of wire and the entire wire from the semantic segmentation result image and calculating the gap geometric features includes: converting the semantic segmentation result image from the RGB color space to the HSV color space to obtain an HSV image; separating the three color channels of the HSV image; extracting the color components of the gap formed between the single broken strand of wire and the entire wire from the separated specified color channels; converting the extracted color components back to the RGB color space to obtain an independent gap segmentation result image; binarizing the gap segmentation result image; performing contour detection on the binarized image; and calculating the area occupied by the pixels within the detected contour to obtain the gap geometric features, wherein the gap geometric features are part or all of the area, length, and width of the gap.

[0025] In this embodiment, binarizing the gap segmentation result image includes: converting the gap segmentation result image to grayscale to obtain a grayscale image; dividing the pixels in the grayscale image into target pixels and background pixels according to a given initial threshold T0; and calculating the proportion of the number of target pixels to the number of background pixels in the grayscale image. The proportion of background pixels to grayscale image : ; ; in, The proportion of the target pixel's pixels to the grayscale image. The proportion of background pixels to the grayscale image; N Less than the initial threshold T The number of background pixels is 0; M The number of target pixels that are greater than the initial threshold T0; A This represents the number of pixels in the horizontal direction of the grayscale image. B This represents the number of pixels in the vertical direction of the grayscale image. A × B Let be the total number of pixels in the grayscale image; calculate the average grayscale value of each grayscale image. Average grayscale value of target pixels and the average grayscale value of background pixels : ; ;

[0026] in, The grayscale value is i The number of pixels, i ∈[0,255]; based on the proportion of the number of pixels of the target pixel to the grayscale image. The proportion of background pixels to grayscale images Average gray level of a grayscale image Average grayscale value of target pixels and the average grayscale value of background pixels Calculate the segmentation threshold: ; in, The segmentation threshold is used to assign the color of each pixel in the grayscale image to either black or white, thus obtaining a binarized image, such as... Figure 2 As shown in (c) in the figure.

[0027] In this embodiment, determining whether the single broken strand of the conductor has been completely pulled back into the gap based on the gap geometry features includes: determining whether the gap geometry features are less than a preset threshold. If they are less than the preset threshold, it is determined that the single broken strand of the conductor has been completely pulled back into the gap, and the live-line working robot's conductor broken strand defect restoration operation ends; otherwise, it is determined that the single broken strand of the conductor has not yet been completely pulled back into the gap, and the conductor broken strand defect restoration operation continues, along with conductor broken strand defect identification and positioning. Specifically, in this embodiment, the gap geometry features are the gap area. If the area is less than the preset threshold, it is determined that the single broken strand of the conductor has been completely pulled back into the gap, and the live-line working robot's conductor broken strand defect restoration operation ends; otherwise, it is determined that the single broken strand of the conductor has not yet been completely pulled back into the gap, and the conductor broken strand defect restoration operation continues, along with conductor broken strand defect identification and positioning.

[0028] like Figure 3 As shown, the conductor strand breakage defect restoration identification and localization model of this embodiment consists of an encoder and a decoder. The encoder includes a feature extraction module, a first attention mechanism module, a hole space pyramid pooling module, and a first 1×1 convolutional layer connected in sequence. The feature extraction module and the first 1×1 convolutional layer of the encoder output feature maps to the decoder respectively. The decoder includes a second 1×1 convolutional layer, a connection module, a first upsampling module, a second attention mechanism module, a 3×3 convolutional layer, and a second upsampling module. The output feature map of the feature extraction module generates a first-path feature map after passing through the second 1×1 convolutional layer. The output feature map of the first 1×1 convolutional layer generates a second-path feature map after passing through the first upsampling module. The first-path feature map and the second-path feature map are concatenated along the dimension by the connection module and then sequentially passed through the second attention mechanism module, the 3×3 convolutional layer, and the second upsampling module to obtain the semantic segmentation result image.

[0029] In this embodiment, the feature extraction module includes a convolutional module, a bottleneck module, an average pooling module, and a fully connected layer module connected in sequence. The convolutional module includes a convolutional layer, a batch normalization layer, and an Hswish activation function layer connected in sequence. The bottleneck module includes a 1×1 upscaling convolutional layer, a 3×3 depthwise separable convolutional layer, a 1×1 downscaling convolutional layer, and an SE attention mechanism module connected in sequence. Feature extraction is completed through the depthwise separable convolutional layer, reducing the number of parameters and computational cost. The fully connected layer module includes a fully connected layer, a batch normalization layer, and an Hswish activation function connected in sequence. The first attention mechanism module includes a global average pooling layer, an adaptive one-dimensional convolutional layer, and a normalization layer connected in sequence. The weight vector obtained from the normalization layer is multiplied by the input feature map of the first attention mechanism module to obtain the output feature map of the first attention mechanism module. The kernel size used in the adaptive one-dimensional convolutional layer is... The expression for the computation function is: ; in, Regarding channel dimension The mapping function, and These are preset parameters; To select the closest odd number, the first attention mechanism module consists of a series of interconnected average pooling layers, adaptive one-dimensional convolutional layers, and normalization layers. The first attention mechanism module first obtains the global average value for each channel through a global average pooling layer, then calculates the weight vector for each channel through an adaptive one-dimensional convolutional kernel. Finally, the weight vector for each channel is passed through a normalization layer and weighted and multiplied by the input feature map to obtain the output of the first attention mechanism module. The second attention mechanism module has the same structure as the first attention mechanism module, so it will not be described further here.

[0030] In this embodiment, the Spatial Pyramid Pooling Module (ASPP module) includes a 1×1 convolutional module, a first composite convolutional module, a second composite convolutional module, a third composite convolutional module, and an image pooling layer. Input data is processed through each of these modules, and then the outputs are fused to obtain the output feature map of the Spatial Pyramid Pooling Module. The first to third composite convolutional modules have identical structures, consisting of sequentially cascaded 3×1 and 1×3 convolutions. The parameters of the convolutional layers in the first composite convolutional module are padding = 6 and dilation = 6. The parameters of the convolutional layers in the second composite convolutional module are padding = 12 and dilation = 12. The parameters of the convolutional layers in the third composite convolutional module are padding = 18 and dilation = 18.

[0031] In the decoder, the second 1×1 convolutional layer takes the output of the encoder's feature extraction module as input; the first upsampling module takes the output of the encoder's first 1×1 convolutional layer as input; the concat module takes the outputs of the second 1×1 convolutional layer and the first upsampling module as input; the second attention mechanism module takes the output of the concat module as input; the second attention mechanism module, the 3×3 convolutional layer, and the second upsampling module are connected in series; the output of the second upsampling module is the final output of the conductor strand breakage defect restoration identification and localization model.

[0032] like Figure 4 As shown, the training of the conductor / ground wire broken strand defect restoration identification and location model in this embodiment includes: S1, acquire image samples of the broken strand defect area during the process of the live-line working robot performing conductor broken strand defect restoration, and obtain the initial image dataset; S2, label the broken strand defect restoration image samples in the initial image dataset, labeling the entire conductor, a single broken strand conductor, and the gap formed between a single broken strand conductor and the entire conductor; in this embodiment, specifically, the Labelme annotation tool is used to label the entire conductor, a single broken strand conductor, and the gap formed between a single broken strand conductor and the entire conductor. S3. The labeled image dataset is augmented with additional samples. This includes performing flipping, rotation, noise addition, brightness adjustment, contrast adjustment, random scaling within a preset range, and boundary filling with black pixels on the initial image dataset's fragmented defect reconstruction image samples to generate new fragmented defect reconstruction image samples. This enhances the diversity of fragmented defect image samples and improves the accuracy of the conductor fragmented defect reconstruction identification and localization model, resulting in a training dataset composed of fragmented defect image samples and their labels; thus, the initial conductor fragmented defect reconstruction identification and localization model is constructed. S4. The initial conductor-ground wire broken strand defect restoration identification and localization model is trained using the training dataset and a preset loss function, thereby obtaining the trained conductor-ground wire broken strand defect restoration identification and localization model.

[0033] In this embodiment, the function expression of the preset loss function is: ; ; ; in, L The default loss function is... L Dice The similarity coefficient loss function, L Focal For focus loss function;t i This is the target value for the i-th category, and its value can be 0 or 1. y i Let be the predicted value for the i-th category, with a value range of (0,1). γ For smoothing coefficients; α i Let be the weight factor for the i-th category; As a regulating factor; Let be the predicted probability of the i-th category.

[0034] To verify the effectiveness of the semantic segmentation-based method for identifying and locating conductor strand defects in this embodiment, 3200 images of conductor strand defects in transmission lines were collected as a dataset. The labels included wire, dwire, and gap, representing the entire conductor, a single strand of broken conductor, and the gap between a single strand of broken conductor and the entire conductor, respectively. To improve the model's applicability and robustness, the dataset was expanded based on the original dataset. This embodiment selected four image expansion measures: (1) flipping / rotating the collected conductor strand defect images; (2) adding a certain amount of noise to the original images; (3) adjusting the brightness / contrast of the original dataset to improve the model's applicability under strong and low light conditions; (4) randomly scaling the conductor strand defect images according to a preset ratio range and filling the boundaries with uniform black pixels. The preprocessed and expanded dataset contained 16000 images, of which 90% were used as the training set and 10% were used as the test set, and the method of this invention was used to process them. A frozen training method was adopted to improve training efficiency and accelerate convergence. The total number of iterations was set to 500. Mosaic data augmentation was used for the first 90% of iterations, and the Adam optimizer was used throughout, employing a transfer learning strategy. The frozen phase consisted of 200 iterations with a batch size of 8 and a learning rate of 0.001. The unfrozen phase consisted of 300 iterations with a batch size of 4 and a learning rate of 0.0001. After training, the system automatically saved the neural network parameters. The model performance was tested using 1600 images from the test set. The test results are as follows: Figure 5 As shown, (a1) is the original first image of the fractured femur defect region, (a2) is the first semantic segmentation result image, and (a3) ​​is the first binarized image. (b1) is the original second image of the fractured femur defect region, (b2) is the second semantic segmentation result image, and (b3) is the second binarized image. (c1) is the original third image of the fractured femur defect region, (c2) is the third semantic segmentation result image, and (c3) is the third binarized image. The final detected gap area of ​​the first fractured femur defect region image is 148.3 cm². 2The gap area in the second image of the fractured region is 139.2 cm². 2 The gap area in the third image of the fractured region is 85.9 cm². 2 According to the test results, the semantic segmentation-based conductor strand breakage defect restoration identification and positioning method of this embodiment can accurately identify the gap at the conductor strand breakage defect and accurately output the size of the gap area at the conductor strand breakage.

[0035] Furthermore, this embodiment also provides a semantic segmentation-based conductor strand breakage defect restoration identification and location system, including a microprocessor and a memory interconnected, wherein the microprocessor is programmed or configured to execute the semantic segmentation-based conductor strand breakage defect restoration identification and location method.

[0036] Those skilled in the art will understand that the technical solutions provided by this invention may take the form of a method, system, or computer program product. Therefore, this invention may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this invention may take the form of a computer program product embodied on one or more computer-readable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, produce an implementation of the flowchart... Figure 1 One or more processes and / or boxes Figure 1 The computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable apparatus for implementing the process. Figure 1 One or more processes and / or boxes Figure 1The steps of the function specified in one or more boxes.

[0037] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for identifying and locating conductor / ground wire strand breakage defects based on semantic segmentation, characterized in that, The process includes the following steps: acquiring an image of the broken strand defect area during the restoration process of the conductor and ground wire broken strand defect by the live-line working robot; inputting the broken strand defect area image into a pre-trained conductor and ground wire broken strand defect restoration recognition and localization model for semantic segmentation to obtain a semantic segmentation result image, wherein the semantic segmentation result image includes the segmented whole conductor, single broken strand conductor, and the gap formed between the single broken strand conductor and the whole conductor; Extract the gap between the single broken strand of the conductor and the entire conductor from the semantic segmentation result image and calculate the geometric features of the gap. Based on the geometric features of the gap, determine whether the single broken strand of the conductor has been completely pulled back into the gap.

2. The method for identifying and locating conductor strand breakage defects based on semantic segmentation according to claim 1, characterized in that, The regions of the entire conductor, the single broken conductor, and the gap between the single broken conductor and the entire conductor in the semantic segmentation result image are filled with a different color other than black, while the remaining regions are filled with black.

3. The method for identifying and locating conductor strand breakage defects based on semantic segmentation according to claim 1, characterized in that, The step of extracting the gap formed between a single broken strand of wire and the entire wire from the semantic segmentation result image and calculating the gap geometric features includes: converting the semantic segmentation result image from RGB color space to HSV color space to obtain an HSV image; separating the three color channels of the HSV image; extracting the color components of the gap formed between the single broken strand of wire and the entire wire from the separated specified color channels; converting the extracted color components back to RGB color space to obtain an independent gap segmentation result image; binarizing the gap segmentation result image; performing contour detection on the binarized image; and calculating the gap geometric features based on the pixels within the contour, wherein the gap geometric features are part or all of the area, length, and width of the gap.

4. The method for identifying and locating conductor strand breakage defects based on semantic segmentation according to claim 3, characterized in that, The binarization of the gap segmentation result image includes: converting the gap segmentation result image to grayscale to obtain a grayscale image; dividing the pixels in the grayscale image into target pixels and background pixels according to a given initial threshold T0; and calculating the proportion of the number of target pixels to the number of background pixels in the grayscale image. The proportion of background pixels to grayscale image : ； ； in, The proportion of the target pixel's pixels to the grayscale image. The proportion of background pixels to the grayscale image; N Less than the initial threshold T The number of background pixels is 0; M The number of target pixels that are greater than the initial threshold T0; A This represents the number of pixels in the horizontal direction of the grayscale image. B This represents the number of pixels in the vertical direction of the grayscale image. A × B Let be the total number of pixels in the grayscale image; calculate the average grayscale value of each grayscale image. Average grayscale value of target pixels and the average grayscale value of background pixels : ； ； in, The grayscale value is i The number of pixels, i ∈[0,255]; based on the proportion of the number of pixels of the target pixel to the grayscale image. The proportion of background pixels to grayscale images Average gray level of a grayscale image Average grayscale value of target pixels and the average grayscale value of background pixels Calculate the segmentation threshold: ； in, The threshold is used for segmentation; based on the segmentation threshold, the color of each pixel in the grayscale image is assigned to either black or white, thus obtaining a binarized image.

5. The method for identifying and locating conductor strand breakage defects based on semantic segmentation according to claim 1, characterized in that, The step of determining whether the single broken strand of the conductor has been completely pulled back into the gap based on the gap geometry features includes: determining whether the gap geometry features are less than a preset threshold; if they are less than the preset threshold, it is determined that the single broken strand of the conductor has been completely pulled back into the gap, and the live-line working robot's conductor broken strand defect restoration operation ends; otherwise, it is determined that the single broken strand of the conductor has not been completely pulled back into the gap, and the conductor broken strand defect restoration operation continues and the conductor broken strand defect restoration identification and positioning are performed.

6. The method for identifying and locating conductor strand breakage defects based on semantic segmentation according to claim 1, characterized in that, The conductor strand breakage defect restoration and identification model consists of an encoder and a decoder. The encoder includes a feature extraction module, a first attention mechanism module, a hole spatial pyramid pooling module, and a first 1×1 convolutional layer connected in sequence. The feature extraction module and the first 1×1 convolutional layer of the encoder output feature maps to the decoder, respectively. The decoder includes a second 1×1 convolutional layer, a connection module, a first upsampling module, a second attention mechanism module, a 3×3 convolutional layer, and a second upsampling module. The output feature map of the feature extraction module is processed by the second 1×1 convolutional layer to generate a first-path feature map. The output feature map of the first 1×1 convolutional layer is processed by the first upsampling module to generate a second-path feature map. The first-path feature map and the second-path feature map are concatenated along the dimension by the connection module and then passed sequentially through the second attention mechanism module, the 3×3 convolutional layer, and the second upsampling module to obtain the semantic segmentation result image.

7. The method for identifying and locating conductor strand breakage defects based on semantic segmentation according to claim 6, characterized in that, The feature extraction module includes a convolutional module, a bottleneck module, an average pooling module, and a fully connected layer module connected in sequence. The convolutional module includes a convolutional layer, a batch normalization layer, and an Hswish activation function layer connected in sequence. The bottleneck module includes a 1×1 upscaling convolutional layer, a 3×3 depthwise separable convolutional layer, a 1×1 downscaling convolutional layer, and an SE attention mechanism module connected in sequence. The fully connected layer module includes a fully connected layer, a batch normalization layer, and an Hswish activation function connected in sequence. The first attention mechanism module includes a global average pooling layer, an adaptive one-dimensional convolutional layer, and a normalization layer connected in sequence. The weight vector obtained from the normalization layer is multiplied by the input feature map of the first attention mechanism module to obtain the output feature map of the first attention mechanism module. The adaptive one-dimensional convolutional layer uses a specific kernel size. The expression for the computation function is: ； in, Regarding channel dimension The mapping function, and These are preset parameters; To select the nearest odd number.

8. The method for restoring, identifying, and locating conductor strand breakage defects based on semantic segmentation according to claim 1, characterized in that, The training of the conductor strand breakage defect restoration identification and localization model includes: acquiring image samples of the strand breakage defect area during the restoration process of the conductor strand breakage defect by the live-line working robot, obtaining an initial image dataset; labeling the strand breakage defect restoration image samples in the initial image dataset, labeling the entire conductor, a single strand breakage conductor, and the gap formed between the single strand breakage conductor and the entire conductor; expanding the labeled image dataset, including performing flipping, rotation, noise addition, brightness adjustment, contrast adjustment, random scaling according to a preset ratio range, and boundary filling with black pixels on the strand breakage defect restoration image samples in the initial image dataset to generate new strand breakage defect restoration image samples, thereby obtaining a training dataset composed of strand breakage defect image samples and their labels; constructing the initial conductor strand breakage defect restoration identification and localization model; and training the initial conductor strand breakage defect restoration identification and localization model using the training dataset and a preset loss function, thereby obtaining the trained conductor strand breakage defect restoration identification and localization model.

9. The method for restoring, identifying, and locating conductor strand breakage defects based on semantic segmentation according to claim 8, characterized in that, The function expression of the preset loss function is: ； ； ； in, L The default loss function is... L Dice The similarity coefficient loss function, L Focal For focus loss function; t i This is the target value for the i-th category, and its value can be 0 or 1. y i Let be the predicted value for the i-th category, with a value range of (0,1). γ For smoothing coefficients; α i Let be the weight factor for the i-th category; As a regulating factor; Let be the predicted probability of the i-th category.

10. A system for identifying and locating broken strand defects in conductors and ground wires based on semantic segmentation, comprising a microprocessor and a memory interconnected, characterized in that, The microprocessor is programmed or configured to execute the semantic segmentation-based conductor strand breakage defect restoration identification and location method according to any one of claims 1 to 9.

Images