An automatic classification method for transmission line defects based on deep visual learning

Abstract

The present invention belongs to the field of computer vision, and relates to a method for automatic classification of power transmission line defects based on deep visual learning, including the following steps: collecting sample image data of the power transmission line, and dividing the image in the sample image data into a number of segmented areas; calculating the maximum depth of the defect feature; presetting the size of the first depth threshold and the second depth threshold; if the maximum depth is less than or equal to the first depth threshold, analyzing whether the defect feature belongs to a corrosion spot for defect assessment level; if the maximum depth is greater than the first depth threshold and less than or equal to the second depth threshold, analyzing the position and area size of the defect feature for defect assessment level; if the maximum depth is greater than the second depth threshold, directly evaluating the defect feature for defect assessment level. The present invention solves the problem that the existing defect assessment compares the similarity of defect features to judge the severity, and the comparison process takes a lot of time.

Description

Technical Field

[0001] The present invention belongs to the field of computer vision and relates to an automatic classification method for power transmission line defects based on deep visual learning. Background Art

[0002] Transmission lines are key components of power transmission, including transmission lines erected on the ground and cable lines laid in underground pipelines. Due to their wide distribution, transmission lines are susceptible to various factors such as natural disasters and external damage, which lead to various defects in the lines. If these defects are not discovered and handled in time, they may cause faults such as short circuits and tripping, and even cause large-scale power outages, seriously affecting social production and residents' lives. Traditional transmission line defect detection methods mainly rely on manual inspections, which have problems such as long inspection cycles, high labor intensity, low efficiency, and susceptibility to subjective factors.

[0003] With the rapid development of computer vision and deep learning technology, it has become possible to automatically identify and classify defects by combining deep learning algorithms. This method can collect image data of cable lines in real time, automatically detect defects in the lines through advanced image processing and analysis technology, and classify and evaluate them, greatly improving the efficiency and accuracy of inspections.

[0004] Based on the above problems, the existing cable line defect assessment for underground pipelines judges the severity by comparing the similarity of defect features, and the comparison process takes a lot of time. Therefore, it is of great practical significance and application value to study a more accurate and efficient automatic classification method for transmission line defects based on deep visual learning. Summary of the invention

[0005] In order to solve the problem that the existing defect assessment judges the severity by comparing the similarity of defect features, and the comparison process takes a lot of time, the present invention provides an automatic classification method for transmission line defects based on deep visual learning.

[0006] A method for automatic classification of transmission line defects based on deep visual learning, comprising the following steps:

[0007] S1, collecting sample image data of the transmission line, dividing the image in the sample image data into a plurality of segmented areas, and encoding the plurality of segmented areas independently;

[0008] S2, extracting features from historical data to obtain defect features that characterize transmission line defects;

[0009] S3, collecting defect depth data of defect features, and calculating the maximum depth of the defect features;

[0010] S4, presetting the first depth threshold and the second depth threshold;

[0011] S5. If the maximum depth is less than or equal to the first depth threshold, construct a visual model of corrosion spots, select images with and without corrosion spots from historical records, and train the visual model of corrosion spots. During the training process of the visual model of corrosion spots, learn how to analyze whether defect features belong to corrosion spots based on the extracted features;

[0012] S6. If the maximum depth is greater than the first depth threshold and less than or equal to the second depth threshold, a visual model of the area and position of the defect feature is constructed based on historical data, and the position and area of ​​the defect feature are analyzed to evaluate the defect level;

[0013] S7. If the maximum depth is greater than a second depth threshold, directly perform defect assessment on the defect feature.

[0014] A further solution of the present invention, step S1, comprises the following steps:

[0015] The mobile monitoring equipment carries a high-definition camera to collect raw image data of the power transmission line and imports the raw image data into the processing system for pre-processing;

[0016] Using the superpixel segmentation algorithm, the image in the collected original image data is segmented into several areas with similar visual features;

[0017] Each segmented region is encoded independently.

[0018] A further solution of the present invention, step S2, comprises the following steps:

[0019] A visual model for defect feature recognition is constructed, images with defect features and without defect features in historical records are selected, and the visual model for defect feature recognition is trained. During the training process of the visual model for defect feature recognition, the model learns how to distinguish defect areas and non-defect areas based on the extracted features.

[0020] A further solution of the present invention is to construct a visual model for defect feature recognition, comprising the following steps:

[0021] The original image data of the historical records are annotated by manual annotation, and a number of annotated original image data are combined into a data set, which includes images of normal and defective power transmission lines; among them, the images with defects in the original image data, the images of the same defect are suitable for different lighting conditions, angles, and backgrounds;

[0022] The labeled data set is divided into training set, validation set and test set to train, validate and test the visual model for defect feature recognition.

[0023] A further solution of the present invention, step S3, comprises the following steps:

[0024] The trained visual model for defect feature recognition extracts and classifies features of the image in the segmented area, compares the extracted features with the defect features learned by the visual model, and the visual model for defect feature recognition determines whether there is a defect in the segmented area;

[0025] If a defect feature existing in the segmented area is identified, defect depth data of the defect feature is collected, and the maximum depth of the defect feature is calculated.

[0026] A further solution of the present invention, step S4, comprises the following steps:

[0027] By comparing the defect depth data with the outer sheath thickness data of the transmission line, a relative value is obtained, and a first depth threshold and a second depth threshold are pre-set to determine within which threshold range the relative value falls.

[0028] A further solution of the present invention, step S5, comprises the following steps:

[0029] The maximum depth is less than or equal to the first depth threshold, and the defect feature is analyzed based on the trained visual model of corrosion spots to determine whether it belongs to a corrosion spot;

[0030] If the defect feature belongs to the defect feature of a corrosion spot, it is judged that the defect feature is moderate; if the defect feature does not belong to the defect feature of a corrosion spot, it is judged that the defect feature is mild.

[0031] A further solution of the present invention, step S6, comprises the following steps:

[0032] Construct a visual model of the area and position of defect features. The visual model of area and position learns how to adaptively focus on key areas in the image.

[0033] The defect area is divided into three concentric circles with the key position as the center. The three dynamically generated concentric circles are used as a reference frame. Three defect assessment levels are set. The three concentric circles are marked from the innermost circle to the outermost circle as: severe, moderate, and mild.

[0034] A further solution of the present invention is to construct a visual model of the area and position of the defect feature, comprising the following steps:

[0035] Use a Canny edge detector or a similar measurement method to determine the edge position of the defect feature close to the center of the circle. After the edge position close to the center of the circle is determined, calculate the distance from the position to the center of the circle to determine which concentric ring the edge of the defect feature falls on.

[0036] The relative area comparison method is used for comparison, that is, the defect feature area in each concentric ring is compared with the total area of ​​the concentric ring to obtain a relative value. According to the comparison of the relative values, the maximum relative value is selected. If the relative area of ​​a concentric ring is the largest, the defect assessment level of the defect feature can be evaluated to correspond to the concentric ring.

[0037] According to the defect characteristics in the defect area, the position and area of ​​the defect characteristics in the defect area are comprehensively analyzed. The two factors are quantified and integrated into a unified evaluation framework for calculation. The higher the comprehensive evaluation value, the higher the severity level of the defect, which satisfies the following formula:

[0038]

[0039] in, Indicates the comprehensive evaluation value of the defect level; represents the weight factor of the position, represents a weighting factor for the area; and ;

[0040] It represents the position coefficient, which is defined according to the concentric circles where the defect characteristics are located: the defect assessment level is severe P=1, the defect assessment level is moderate P=0.67, and the defect assessment level is mild P=0.33;

[0041] Indicates the area coefficient, which is defined according to the concentric rings where the defect features are located: the defect assessment level is severe =1, the defect assessment level is medium =0.67, the defect assessment level is mild =0.33.

[0042] A further solution of the present invention, step S7, comprises the following steps:

[0043] If the maximum depth is greater than the second depth threshold, the defect feature is directly evaluated as severe; the defect area is automatically classified into a corresponding severity level, and the defect evaluation level is directly output to the power system operation and maintenance personnel.

[0044] In summary, the present invention includes the following beneficial technical effects:

[0045] 1. Through inspections with cameras and lidar, sample image data and defect depth data of the transmission line are collected in real time; combined with a deep visual learning model, the collected data is automatically processed and analyzed to accurately identify defective areas in the transmission line, significantly improving the efficiency and accuracy of inspections and ensuring the safe and stable operation of the transmission line;

[0046] 2. By introducing the attention mechanism and dynamically generating concentric circles as a reference frame, it is possible to focus on the key positions of each segmented area and conduct a comprehensive analysis of the position, area, and depth of the defect features. This refined analysis method helps to more accurately assess the severity and potential risks of the defects, providing strong support for subsequent repair and replacement work;

[0047] 3. This systematic solution organically combines the steps of sample image data collection, defect feature extraction, segmentation area feature judgment, key position focusing, comprehensive analysis and automatic classification to improve the intelligent level of transmission line maintenance and make maintenance work more efficient, accurate and reliable. Combined with pre-set thresholds, defect areas can be automatically classified to further improve the accuracy and efficiency of classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 A flow chart of a method for automatic classification of transmission line defects based on deep visual learning is disclosed.

[0049] Figure 2 A framework schematic diagram of an automatic classification system for transmission line defects based on deep visual learning is disclosed. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field without creative work belong to the scope of protection of the present invention.

[0051] The following is combined with Figure 1-2 The preferred embodiments of the present invention are described in detail.

[0052] See attached Figure 1 The present invention proposes a method for automatic classification of power transmission line defects based on deep visual learning. Based on deep visual learning, sample image data of power transmission lines are collected, defect features are extracted after segmentation and encoding, defect areas are compared and identified, an attention mechanism is introduced to focus on key positions and dynamically generate concentric ring frames, and the position, area and depth of defect features are comprehensively analyzed to automatically classify power transmission line defects. The method includes the following steps:

[0053] S1, collecting sample image data of the transmission line, dividing the image in the sample image data into a plurality of segmented areas, and encoding the plurality of segmented areas independently;

[0054] S2, extracting features from historical data to obtain defect features that characterize transmission line defects;

[0055] S3, collecting defect depth data of defect features, and calculating the maximum depth of the defect features;

[0056] S4, presetting the first depth threshold and the second depth threshold;

[0057] S5. If the maximum depth is less than or equal to the first depth threshold, construct a visual model of corrosion spots, select images with and without corrosion spots from historical records, and train the visual model of corrosion spots. During the training process of the visual model of corrosion spots, learn how to analyze whether defect features belong to corrosion spots based on the extracted features;

[0058] S6. If the maximum depth is greater than the first depth threshold and less than or equal to the second depth threshold, a visual model of the area and position of the defect feature is constructed based on historical data, and the position and area of ​​the defect feature are analyzed to evaluate the defect level;

[0059] S7. If the maximum depth is greater than a second depth threshold, directly perform defect assessment on the defect feature.

[0060] Specifically, the mobile monitoring device carries a laser radar and a camera to perform inspections along the path direction of the buried cable channel. The camera captures a sample image of the transmission line to collect sample image data of the transmission line; the image in the sample image data is divided into a plurality of segmented areas, and the plurality of segmented areas are independently encoded;

[0061] The sample image data of the segmented area is subjected to image preprocessing such as noise removal, contrast adjustment, and clarity adjustment to optimize the sample image data of the segmented area after encoding. Based on the feature extraction of historical data, the defect features that characterize the defects of the transmission line are obtained, and the defect features existing in the segmented area are identified. The first depth threshold and the second depth threshold are set in advance, and the laser radar collects the defect depth data of the defect feature to calculate the maximum depth of the defect feature. Compare the maximum depth of the defect feature with the first depth threshold and the second depth threshold. If the maximum depth is less than or equal to the first depth threshold, a visual model of the corrosion spot is constructed based on the historical data, and the defect feature is analyzed whether it belongs to the corrosion spot for defect assessment level; if the maximum depth is greater than the first depth threshold and less than or equal to the second depth threshold, a visual model of the area and position of the defect feature is constructed based on the historical data, and the position and area size of the defect feature are analyzed for defect assessment level; if the maximum depth is greater than the second depth threshold, the defect feature of the sample image data is directly evaluated for defect level.

[0062] In one embodiment of the present invention, step S1 includes the following steps:

[0063] The mobile monitoring equipment carries a high-definition camera to collect raw image data of the transmission line, and imports the raw image data into the processing system for image preprocessing such as noise removal, contrast adjustment, and clarity adjustment to optimize the raw image data of the transmission line.

[0064] The superpixel segmentation algorithm (SLIC) is used to segment the image in the collected original image data into several areas with similar visual features such as shape and brightness, which helps subsequent recognition to focus more on the specific details of the image; wherein the superpixel segmentation algorithm (SLIC) performs clustering based on local features such as shape, brightness, and color of the image.

[0065] The encoder uses the run length encoding (RLE) encoding technology. The encoder performs independent encoding processing on each segmented area to eliminate or reduce encoding redundancy, laying the foundation for subsequent feature extraction and defect identification.

[0066] In one embodiment of the present invention, step S2 comprises the following steps:

[0067] Construct a visual model for defect feature recognition, select images with defect features and without defect features from historical records, train the visual model for defect feature recognition, and during the training process of the visual model for defect feature recognition, learn how to distinguish defect and non-defect areas based on the extracted features.

[0068] In one embodiment of the present invention, constructing a visual model for defect feature recognition includes the following steps:

[0069] The original image data of the historical records are annotated by manual annotation methods, and several annotated original image data are combined into a data set, which includes images of normal and defective transmission lines. The types of defects include corrosion spots, line damage, and line bumps. For example, the staff uses professional annotation tools to draw a 2D bounding box to annotate the defect shape of the defect features;

[0070] Among them, the images with defects in the original image data and the images with the same defect should be suitable for different lighting conditions, angles, and backgrounds to ensure the generalization ability of the visual model for defect feature recognition.

[0071] The labeled data set is divided into a training set, a validation set, and a test set. The training set is used to train the visual model for defect feature recognition, the validation set is used to adjust the parameters of the visual model for defect feature recognition and verify the performance of the visual model for defect feature recognition during the training process, and the test set is used to finally evaluate the accuracy of the visual model for defect feature recognition;

[0072] Exemplarily, the training set, validation set, and test set are allocated in a ratio of 7:2:1, and the SSD target detection model is selected as the visual model for defect feature recognition. The training set allows the visual model for defect feature recognition to learn the features in the transmission line image and identify the defect features. The validation set is used to verify that the visual model for defect feature recognition can recognize defect features under different lighting conditions, angles, and backgrounds during the training process. The test set is used to ultimately evaluate the accuracy of the visual model for defect feature recognition.

[0073] In one embodiment of the present invention, step S3 includes the following steps:

[0074] The trained visual model for defect feature recognition extracts and classifies features of the image of the power transmission line, compares the extracted features with the defect features learned in the model, and the visual model for defect feature recognition determines whether there is a defect in the segmented area;

[0075] If defect features with obvious differences in the segmented areas are identified, the mobile monitoring equipment carries a laser radar to collect defect depth data of the transmission line to optimize the defect depth data of the defect features of the transmission line.

[0076] In one embodiment of the present invention, step S4 comprises the following steps:

[0077] Compare the defect depth data with the outer sheath thickness data of the transmission line to obtain a relative value, pre-set a first depth threshold and a second depth threshold, and determine in which threshold range the relative value falls;

[0078] Exemplarily, the first depth threshold is preset to 0.2, the second depth threshold is preset to 0.7, the laser radar measures the depth of the defect feature, calculates the maximum depth of the defect feature, and compares the maximum depth with the outer sheath thickness data of the transmission line to obtain a relative value;

[0079] Relative value range: relative value 0.2, it is judged that the maximum depth is much smaller than the outer sheath thickness of the transmission line, and the defect assessment level of step S5 is continued;

[0080] Relative value range: 0.2 Relative value 0.7, it is judged that the maximum depth is close to half of the thickness of the outer sheath of the transmission line, and the defect assessment level of step S6 is continued;

[0081] Relative value range: relative value 0.7, it is judged that the maximum depth is close to the outer sheath thickness of the transmission line, and the defect assessment level of step S7 is continued.

[0082] In one embodiment of the present invention, constructing a visual model of corrosion spots includes the following steps:

[0083] According to the original image data of historical records, manual annotation is performed, and several annotated original image data of corrosion spots are combined into a data set, which is divided into a training set, a validation set and a test set. The training set is used to train the visual model of corrosion spots, the validation set is used to adjust the visual model parameters of corrosion spots and verify the performance of the visual model of corrosion spots during the training process, and the test set is used to evaluate the accuracy of the visual model for defect feature recognition;

[0084] Exemplarily, the training set, validation set, and test set are allocated in a ratio of 7:2:1. The SSD target detection model is selected as the visual model of corrosion spots. The training set allows the visual model of corrosion spots to learn the corrosion spots in the transmission line image and identify the corrosion spots. The validation set is used to verify the visual model of corrosion spots in the training process for the recognition of corrosion spots under different lighting conditions, angles, and backgrounds. The test set is used to finally evaluate the accuracy of the visual model of corrosion spots.

[0085] In one embodiment of the present invention, step S5 comprises the following steps:

[0086] If the maximum depth is less than or equal to the first depth threshold, the defect feature is analyzed based on the trained visual model of corrosion spots to determine whether it belongs to a corrosion spot. If the defect feature belongs to a corrosion spot, it is judged that the defect feature belongs to a medium defect assessment level and requires relevant professionals to clean and repair it. If the defect feature does not belong to a corrosion spot, it is judged that the defect feature belongs to a mild defect assessment level and requires frequent inspection.

[0087] In one embodiment of the present invention, step S6 comprises the following steps:

[0088] Construct a visual model of the area and position of defect features. The visual model of the area and position of defect features introduces an attention mechanism. Based on historical data and experience, the visual model of the area and position learns how to adaptively focus on key areas in the image. For example, the weak locations of the transmission line or the locations where the transmission line is prone to short circuits are set as key areas.

[0089] The defect area is divided into three concentric rings with the key position as the center. The three dynamically generated concentric rings are used as the reference frame to set three defect assessment levels. The three concentric rings are marked as severe, moderate, and mild from the innermost ring to the outermost ring, laying the foundation for subsequent automatic classification. For example, based on historical data and experience, it is assumed that a certain position in a defect area is prone to short circuit problems. The visual model of area and position divides the defect area into three concentric rings with the position as the center.

[0090] Based on the data and experience of the transmission line history records, the key positions of the defective area are intelligently focused to highlight the characteristics of the key key positions. The key position of each defective area is the center of the circle. Combined with the reference frame of three concentric rings, the position and area of ​​the defect characteristics of the defective area are comprehensively analyzed, and the defective area is classified based on the pre-set threshold.

[0091] In one embodiment of the present invention, constructing a visual model of the area and position of defect features includes the following steps:

[0092] 1. Location of features

[0093] The process of determining the feature position requires clarifying the specific division criteria of the concentric rings. The concentric ring where the edge of the defect feature close to the center of the circle falls is the defect assessment level, that is, the closer the defect feature is to the key position, the more dangerous it is.

[0094] Specifically, a Canny edge detector or a similar measurement method is used to determine the edge position of the defect feature close to the center of the circle. After the edge position close to the center of the circle is determined, the distance from the position to the center of the circle is calculated to determine which concentric ring the edge of the defect feature falls on;

[0095] In order to improve the accuracy of the judgment, a reasonable error range is set. The position of the feature edge falls within a certain concentric ring within this error range, and the defect assessment level of the defect feature can be evaluated to correspond to the concentric ring.

[0096] For example, assuming that a crack defect feature appears in the image in the original image data, the edge of the crack is identified by the Canny edge detector, and the distance from the crack edge to the center of the circle is measured to determine that it falls within the "medium" circle and the error is within an acceptable range. Therefore, the defect assessment level of the defect feature is determined to be medium.

[0097] 2. Area of ​​the feature

[0098] The process of determining the feature position requires clarifying the specific division criteria of the concentric rings. The concentric ring where the defect feature falls on the largest area is the defect assessment level.

[0099] Specifically, with the help of the segmentation algorithm in image processing technology, the defect features are accurately segmented and extracted to ensure that the division of concentric rings and the segmentation of feature areas are aligned on the image. The segmented feature areas will be processed as independent objects, and the measurement tool of pixel counting will be used to calculate the area of ​​the feature area in each concentric ring.

[0100] The relative area comparison method is used for comparison, that is, the defect characteristic area in each concentric ring is compared with the total area of ​​the concentric ring to obtain a relative value. According to the comparison of the relative values, the maximum relative value is selected. If the relative area of ​​a concentric ring is the largest, the defect assessment level of the defect characteristic can be evaluated to correspond to the concentric ring.

[0101] For example, by using the segmentation algorithm in image processing technology, the defect features are accurately segmented and extracted, and the area of ​​the defect feature segmentation region in each concentric ring is calculated. By using the relative area comparison method, it is found that the relative area of ​​the defect feature segmentation region in the "mild" ring is the largest. Therefore, the defect assessment level of the defect feature is determined to be mild.

[0102] 3. Comprehensive evaluation

[0103] For the defect characteristics in the defect area, it is necessary to comprehensively analyze the location and area of ​​the defect characteristics in the defect area. The two factors are quantified and integrated into a unified evaluation framework for calculation. The higher the comprehensive evaluation value, the higher the severity level of the defect, satisfying the following formula:

[0104]

[0105] in, Indicates the comprehensive evaluation value of the defect level; represents the weight factor of the position, represents a weighting factor for the area; and ;

[0106] It represents the position coefficient, which is defined according to the concentric circles where the defect characteristics are located: the defect assessment level is severe P=1, the defect assessment level is moderate P=0.67, and the defect assessment level is mild P=0.33;

[0107] Indicates the area coefficient, which is defined according to the concentric rings where the defect features are located: the defect assessment level is severe =1, the defect assessment level is medium =0.67, the defect assessment level is mild =0.33;

[0108] For example, it is assumed that the visual model of area and position determines that there is a defect in the segmented area, that is, there is a defect feature in the segmented area; at this time, P=0.67, =0.33; weight factor determined based on expert experience: =0.5, =0.3;

[0109] Combine historical data and experience to set the threshold in advance: D 0.15 means the defect assessment level is safe, 0.15 D 0.35 means the defect assessment level is mild, 0.35 D 0.55 means the defect assessment level is moderate, D 0.55 means the defect assessment level is severe;

[0110] Substituting into the formula, we can find that: 434;

[0111] According to the preset level threshold, 434 falls on 0.35 D 0.55, indicating that the defect assessment level is moderate.

[0112] In one embodiment of the present invention, step S7 includes the following steps:

[0113] If the maximum depth is greater than the second depth threshold, the defect feature is directly evaluated for defect grade.

[0114] Exemplary, relative value range: relative value 0.7, the maximum depth is judged to be close to the outer sheath thickness of the transmission line, and the defect assessment level is directly judged to be severe;

[0115] The defect areas are automatically classified into corresponding severity levels, and then the defect assessment levels are directly output to the power system operation and maintenance personnel, so that the power system operation and maintenance personnel can take timely maintenance or replacement measures to ensure the safe and stable operation of the transmission lines.

[0116] See attached Figure 2 The present invention also proposes an automatic classification system for power transmission line defects based on deep visual learning, which includes: a sample image data acquisition and segmentation module, a defect feature extraction module, a depth data calculation module, and a comprehensive analysis and automatic classification module.

[0117] The sample image data acquisition and segmentation module is used to acquire sample image data of the power transmission line in real time, segment the image in the sample image data into a plurality of segmented areas, and perform independent encoding processing on the plurality of segmented areas;

[0118] A defect feature extraction module extracts features from defect areas of a transmission line to obtain defect features for characterizing defects of the transmission line;

[0119] A depth data calculation module is used to collect defect depth data of defect features and calculate the maximum depth of the defect features;

[0120] The comprehensive analysis and automatic classification module is used to comprehensively analyze the location, area, and depth of the defect characteristics of the defect area, and automatically classify the defect area based on the pre-set threshold.

[0121] The modules described above may be implemented in whole or in part through software, hardware, or a combination thereof, supporting processors embedded in or independent of a computer device in hardware form, and also supporting memories stored in a computer device in software form, so that the processor can call and execute operations corresponding to the modules described above.

[0122] It should be noted that the user information (including but not limited to user device information and personal information, etc.) and data (including but not limited to data used for analysis, stored data and displayed data, etc.) involved in the present invention are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards.

[0123] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention is described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A method for automatic classification of transmission line defects based on deep visual learning, characterized in that: The following steps are involved: S1, collecting sample image data of the transmission line, dividing the image in the sample image data into a plurality of segmented areas, and encoding the plurality of segmented areas independently; S2, extracting features from historical data to obtain defect features that characterize transmission line defects; S3, collecting defect depth data of defect features, and calculating the maximum depth of the defect features; S4, presetting the first depth threshold and the second depth threshold; A relative value is obtained by comparing the defect depth data with the outer sheath thickness data of the transmission line, and a first depth threshold and a second depth threshold are pre-set to determine in which threshold range the relative value falls; S5. If the maximum depth is less than or equal to the first depth threshold, construct a visual model of corrosion spots, select images with and without corrosion spots from historical records, and train the visual model of corrosion spots. During the training process of the visual model of corrosion spots, learn how to analyze whether defect features belong to corrosion spots based on the extracted features; The maximum depth is less than or equal to the first depth threshold, and the defect feature is analyzed based on the trained visual model of corrosion spots to determine whether it belongs to a corrosion spot; If the defect feature belongs to the defect feature of a corrosion spot, it is judged that the defect feature belongs to moderate; if the defect feature does not belong to the defect feature of a corrosion spot, it is judged that the defect feature belongs to mild; S6. If the maximum depth is greater than the first depth threshold and less than or equal to the second depth threshold, a visual model of the area and position of the defect feature is constructed based on historical data, and the position and area of ​​the defect feature are analyzed to evaluate the defect level; Construct a visual model of the area and position of defect features. The visual model of area and position learns how to adaptively focus on key areas in the image. The defect area is divided into three concentric circles with the key position as the center. The three dynamically generated concentric circles are used as the reference frame. Three defect assessment levels are set. The three concentric circles are marked as severe, moderate, and mild from the innermost circle to the outermost circle. Use a Canny edge detector or a similar measurement method to determine the edge position of the defect feature close to the center of the circle. After the edge position close to the center of the circle is determined, calculate the distance from the position to the center of the circle to determine which concentric ring the edge of the defect feature falls on. The relative area comparison method is used for comparison, that is, the defect feature area in each concentric ring is compared with the total area of ​​the concentric ring to obtain a relative value. According to the comparison of the relative values, the maximum relative value is selected. If the relative area of ​​a concentric ring is the largest, the defect assessment level of the defect feature can be evaluated to correspond to the concentric ring. According to the defect characteristics in the defect area, the position and area of ​​the defect characteristics in the defect area are comprehensively analyzed. The two factors are quantified and integrated into a unified evaluation framework for calculation. The higher the comprehensive evaluation value, the higher the severity level of the defect, which satisfies the following formula: ； in, Indicates the comprehensive evaluation value of the defect level; represents the weight factor of the position, represents a weighting factor for the area; and ; It represents the position coefficient, which is defined according to the concentric circles where the defect characteristics are located: the defect assessment level is severe P=1, the defect assessment level is moderate P=0.67, and the defect assessment level is mild P=0.33; Indicates the area coefficient, which is defined according to the concentric rings where the defect features are located: the defect assessment level is severe =1, the defect assessment level is medium =0.67, the defect assessment level is mild =0.33; S7. If the maximum depth is greater than a second depth threshold, directly perform defect assessment on the defect feature.

2. According to claim 1, a method for automatic classification of power transmission line defects based on deep visual learning is characterized in that: Step S1 includes the following steps: The mobile monitoring equipment carries a high-definition camera to collect raw image data of the power transmission line and imports the raw image data into the processing system for pre-processing; Using the superpixel segmentation algorithm, the image in the collected original image data is segmented into several areas with similar visual features; Each segmented region is encoded independently.

3. According to claim 1, a method for automatic classification of power transmission line defects based on deep visual learning is characterized in that: Step S2 includes the following steps: A visual model for defect feature recognition is constructed, images with defect features and without defect features in historical records are selected, and the visual model for defect feature recognition is trained. During the training process of the visual model for defect feature recognition, the model learns how to distinguish defect areas and non-defect areas based on the extracted features.

4. The method for automatic classification of power transmission line defects based on deep visual learning according to claim 3 is characterized in that: Constructing the visual model for defect feature recognition includes the following steps: The original image data of the historical records are annotated by manual annotation, and a number of annotated original image data are combined into a data set, which includes images of normal and defective power transmission lines; among them, the images with defects in the original image data, the images of the same defect are suitable for different lighting conditions, angles, and backgrounds; The labeled data set is divided into training set, validation set and test set to train, validate and test the visual model for defect feature recognition.

5. The method for automatic classification of power transmission line defects based on deep visual learning according to claim 4 is characterized in that: Step S3 includes the following steps: The trained visual model for defect feature recognition extracts and classifies features of the image in the segmented area, compares the extracted features with the defect features learned by the visual model, and the visual model for defect feature recognition determines whether there is a defect in the segmented area; If a defect feature existing in the segmented area is identified, defect depth data of the defect feature is collected, and the maximum depth of the defect feature is calculated.

6. The method for automatic classification of power transmission line defects based on deep visual learning according to claim 5 is characterized in that: Step S7 includes the following steps: If the maximum depth is greater than the second depth threshold, the defect feature is directly evaluated as severe; the defect area is automatically classified into a corresponding severity level, and the defect evaluation level is directly output to the power system operation and maintenance personnel.

Images