Power transmission line strain clamp defect detection method based on unmanned aerial vehicle and X-ray imaging

Abstract

The invention discloses a power transmission line strain clamp defect detection method based on an unmanned aerial vehicle and X-ray imaging, and belongs to the field of intelligent defect detection, and the method comprises the following steps: optimizing an imaging parameter combination of X-ray imaging equipment based on the material and thickness information of a strain clamp to obtain optimized X-ray imaging equipment; the optimized X-ray imaging equipment carries the unmanned aerial vehicle to obtain an internal structure image of the strain clamp; obtaining a defect image and a defect identification result based on the internal structure image of the strain clamp; constructing a distributed database based on the defect image and the defect identification result; and performing comparative analysis on the newly collected strain clamp image and the information of the distributed database to obtain a health assessment report of the strain clamp. According to the invention, automatic nondestructive testing of the internal structure of the strain clamp is realized, potential defects can be quickly and accurately found, safe and stable operation of a power transmission line is effectively guaranteed, and the method has important engineering application value.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent defect monitoring, and in particular relates to a method for detecting defects in a transmission line tension clamp based on an unmanned aerial vehicle and X-ray imaging. Background Art

[0002] The transmission line tension clamp is a key component to ensure the reliable connection between the conductor and the tower, and its health status directly affects the safe operation of the transmission line. However, since the tension clamp is in harsh outdoor environments for a long time, it is prone to defects such as corrosion, wear, and deformation, threatening the reliability of the transmission line. Traditional manual inspection methods are inefficient and difficult to detect hidden defects in a timely manner. The emergence of drone inspection technology has provided a new idea for solving this problem, but how to obtain high-quality images of tension clamp defects and quickly and accurately retrieve similar defects from massive inspection data is still a challenge. Therefore, based on the above technical problems, the present invention provides a transmission line tension clamp defect detection method using drones and X-ray imaging. Summary of the invention

[0003] In order to solve the above technical problems, the present invention proposes a method for detecting defects in transmission line tension clamps based on drones and X-ray imaging to solve the problems existing in the above-mentioned prior art.

[0004] To achieve the above object, the present invention provides a method for detecting defects of a transmission line tension clamp based on a drone and X-ray imaging, comprising the following steps:

[0005] Based on the material and thickness information of the tension clamp, the imaging parameter combination of the X-ray imaging device is optimized to obtain the optimized X-ray imaging device;

[0006] The optimized X-ray imaging device is equipped with a drone to obtain an image of the internal structure of the tension clamp;

[0007] Obtaining a defect image and a defect recognition result based on the internal structure image of the tension clamp;

[0008] Building a distributed database based on the defect image and the defect recognition result;

[0009] The newly acquired image of the tension clamp is compared and analyzed with the information in the distributed database to generate a health assessment report of the tension clamp.

[0010] Optionally, the process of optimizing the imaging parameter combination of the X-ray imaging device based on the material and thickness information of the tension clamp includes:

[0011] Obtain the ray energy range based on the material information of the tension clamp;

[0012] The imaging time range is obtained based on the thickness information of the tension clamp;

[0013] Setting constraints for the machine learning algorithm based on the ray energy range and imaging time range;

[0014] Iterating the machine learning algorithm based on the constraint conditions to obtain optimal ray energy and imaging time;

[0015] The optimal operating parameters of the X-ray imaging device are set based on the optimal ray energy and imaging time.

[0016] Optionally, the optimized X-ray imaging device is carried on a UAV to construct an image correction model by acquiring flight parameters of the UAV, and the collected X-ray image is processed based on the image correction model to obtain the internal structure image of the tension clamp, and the internal structure image of the tension clamp is preliminarily marked.

[0017] Optionally, the process of obtaining an image of the internal structure of the tension clamp and performing preliminary marking includes:

[0018] Performing distortion correction on the X-ray image to obtain a corrected image;

[0019] Performing contrast enhancement processing on the corrected image to obtain an enhanced image;

[0020] Obtaining an internal structure image of the tension clamp based on the enhanced image;

[0021] The internal structural features of the internal structural image of the tension clamp are extracted, and when the internal structural features are dark areas, they are marked as defects.

[0022] Optionally, the process of performing defect recognition on the internal structure image of the tension clamp includes:

[0023] Perform image segmentation on the marked internal structure image of the tension clamp to obtain a candidate defect area image;

[0024] Extracting feature vectors of the candidate defect region image to obtain morphological features and texture features;

[0025] Inputting the feature vector into a support vector machine classifier to perform defect judgment to obtain a first discrimination result, wherein the first discrimination result includes a defect type and a defect location coordinate;

[0026] Inputting the texture feature into a convolutional neural network to perform texture feature enhancement to obtain a second discrimination result;

[0027] The first discrimination result and the second discrimination result are combined to obtain a defect recognition result.

[0028] Optionally, the marked internal structure image of the tension clamp is segmented based on a region growing algorithm to obtain a candidate defect region image.

[0029] Optionally, the process of fusing the first discrimination result and the second discrimination result includes:

[0030] Set result thresholds;

[0031] Performing a weighted summation on the first discrimination result and the second discrimination result to obtain an actual value;

[0032] When the actual value is greater than the result threshold, the current candidate area is considered to be a real defect;

[0033] When the actual value is less than or equal to the result threshold, the current candidate area is considered to be non-defective.

[0034] Optionally, the process of building a distributed database includes:

[0035] Converting the defect image and the defect recognition result into a standardized format to obtain a standardized defect image and a standardized defect recognition result;

[0036] According to the feature information of the normalized defect image, a consistent hashing algorithm is used to calculate the hash index of the defect image, and a mapping relationship is established between the hash index of the defect image and the defect image;

[0037] According to the characteristic information of the normalized defect identification result, a consistent hashing algorithm is used to calculate the hash index of the defect identification result, and a mapping relationship is established between the hash index of the defect identification result and the defect identification result;

[0038] Building a distributed database based on data sharding technology to store defect images and defect recognition results;

[0039] The hash index based on the defect image and the hash index of the defect recognition result are retrieved and accessed in the distributed database.

[0040] Optionally, the process of performing health assessment on the newly acquired tension clamp image based on the distributed database includes:

[0041] Set similarity score threshold;

[0042] A K-nearest neighbor algorithm is used to calculate a similarity score between a newly acquired tension clamp image and a historical defect image in the distributed database;

[0043] When the similarity score is greater than the similarity score threshold, the newly acquired tension clamp image has the corresponding defect type;

[0044] When the similarity score is less than or equal to the similarity score threshold, the newly acquired tension clamp image does not have a corresponding defect.

[0045] Compared with the prior art, the present invention has the following advantages and technical effects:

[0046] The present invention provides an efficient method for detecting defects in transmission line tension clamps. By using an unmanned aerial vehicle equipped with an X-ray imaging device and combining it with a machine learning algorithm, accurate detection of the internal structure of the tension clamp is achieved. First, based on the material and thickness information of the tension clamp, the X-ray imaging parameters are optimized to ensure the imaging quality while reducing the radiation dose. Then, the unmanned aerial vehicle is equipped with the optimized X-ray device to obtain a high-definition internal structure image of the tension clamp. Next, image processing technology is used to extract defect features, and defect recognition is performed in combination with support vector machines and convolutional neural networks to construct a distributed database containing defect images and recognition results. Finally, a health assessment report for the tension clamp is generated by comparing the newly collected image with the database information. This method improves the automation and accuracy of detection, reduces the risk and cost of manual detection, and provides strong technical support for the maintenance of transmission lines. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0048] Figure 1 This is a flow chart of a method for detecting defects in a transmission line tension clamp based on drones and X-ray imaging according to an embodiment of the present invention;

[0049] Figure 2 It is a flow chart of defect recognition of the internal structure image of the tension clamp in an embodiment of the present invention;

[0050] Figure 3 The present invention is a flowchart of constructing a distributed database in an embodiment of the present invention. DETAILED DESCRIPTION

[0051] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0052] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0053] Embodiment 1

[0054] like Figure 1As shown, in this embodiment, a method for detecting defects in transmission line tension clamps based on drones and X-ray imaging is provided, comprising the following steps: optimizing the imaging parameter combination of an X-ray imaging device based on the material and thickness information of the tension clamp to obtain an optimized X-ray imaging device; carrying an drone on the optimized X-ray imaging device to obtain an internal structure image of the tension clamp; obtaining a defect image and a defect recognition result based on the internal structure image of the tension clamp; constructing a distributed database based on the defect image and the defect recognition result; and comparing and analyzing the newly collected tension clamp image with the information in the distributed database to generate a health assessment report for the tension clamp.

[0055] As a specific implementation of this embodiment, the following steps are included:

[0056] Step 1: According to the material and thickness information of the tension clamp, the radiation energy and imaging time parameters of the X-ray imaging equipment are optimized through machine learning algorithms. Under the premise of ensuring image quality, the radiation dose and imaging time are minimized to obtain the optimal imaging parameter combination.

[0057] Furthermore, the material information and thickness information of the tension clamp are obtained and used as input features of the machine learning algorithm. According to the material information of the tension clamp, the radiation energy range corresponding to the material is obtained from the preset material and radiation energy correspondence table. According to the thickness information of the tension clamp, the imaging time range corresponding to the thickness is obtained from the preset thickness and imaging time correspondence table. The radiation energy range and imaging time range are used as constraints of the machine learning algorithm, and the optimal combination of radiation energy and imaging time parameters is searched through the optimization algorithm. In the optimization process, the comprehensive evaluation index is calculated according to the radiation dose and imaging time corresponding to the current parameter combination, and it is used as the objective function of the optimization algorithm. Through iterative optimization, the radiation energy and imaging time parameters are continuously updated until the optimal parameter combination with the minimum radiation dose and imaging time is found under the premise of meeting the image quality requirements. The radiation energy and imaging time in the optimal parameter combination are set as the working parameters of the X-ray imaging device, and the tension clamp is imaged to obtain high-quality, low-radiation, and fast imaging image results.

[0058] As a specific implementation of this embodiment, the material and thickness of the tension clamp are key input features of X-ray imaging. There are many ways to obtain this information, such as manual measurement, reading from a product database, or automatic extraction using image recognition technology. Assume that the material of a tension clamp is aluminum alloy and the thickness is 5 mm. The material information determines the penetration ability of X-rays. Different materials have different absorption rates of X-rays, and it is necessary to select the appropriate ray energy to obtain a clear image. The role of the preset material and ray energy correspondence table is to quickly determine the appropriate ray energy range according to the material. For example, the ray energy range corresponding to aluminum alloy may be 40-60kV. This range is an empirical value and can be adjusted according to actual conditions. The significance of setting this range is to avoid insufficient image contrast due to too low ray energy, or excessive penetration due to too high ray energy, and loss of image details. The thickness information determines the exposure time required for imaging. The greater the thickness, the longer the exposure time required. Similarly, the preset thickness and imaging time correspondence table can quickly determine the appropriate imaging time range according to the thickness. For example, the imaging time range corresponding to a 5 mm thick tension clamp may be 0.5-1 second. Reasonable imaging time can avoid underexposure or overexposure and ensure image quality. After obtaining the ray energy range (40-60kV) corresponding to the material and the imaging time range (0.5-1 second) corresponding to the thickness, it is necessary to select the optimal combination of ray energy and imaging time. Here, the optimization algorithm is needed. The goal of the optimization algorithm is to minimize the radiation dose and imaging time while meeting the image quality requirements. The comprehensive evaluation index can be defined as the weighted sum of the radiation dose and the imaging time, and the weight can be adjusted according to actual needs. For example, if you pay more attention to the radiation dose, you can increase the weight of the radiation dose. The optimization algorithm searches for the optimal solution within this parameter range. For example, it will first try a ray energy of 40kV and an imaging time of 0.5 seconds. Then calculate the comprehensive evaluation index based on the imaging results. Next, it will try a ray energy of 45kV and an imaging time of 0.7 seconds, and calculate the comprehensive evaluation index again. Through continuous iteration, an optimal combination of ray energy and imaging time is finally found, such as 50kV and 0.8 seconds. This combination can not only ensure image quality, but also control the radiation dose and imaging time within a reasonable range. The reason for adopting this method is that it can automatically adjust the imaging parameters according to the specific conditions of the inspected workpiece, avoid the limitations of manual experience, improve imaging efficiency and accuracy, and at the same time minimize the radiation dose to ensure the safety of the operator.

[0059] Step 2: Obtain the parameters such as the UAV flight attitude, ambient wind speed, temperature and humidity, use the convolutional neural network to build an image correction model, perform distortion correction and contrast enhancement on the collected X-ray images, improve the image quality, and output a high-definition image of the internal structure of the tension clamp.

[0060] Furthermore, the flight attitude, environmental wind speed, temperature and humidity of the UAV and other parameters are obtained, and these parameters are used as the input of the convolutional neural network. The image correction model is constructed by the convolutional neural network, and the model can perform distortion correction on the image according to the input UAV flight parameters. The X-ray image collected by the UAV is obtained, and it is input into the image correction model, and the image is subjected to distortion correction processing to obtain a corrected image. The corrected image is subjected to contrast enhancement processing to improve the contrast and clarity of the image to obtain an enhanced image. The internal structure features of the tension clamp are extracted from the enhanced image, and it is judged whether the internal structure of the tension clamp has defects according to the features. If it is judged that the internal structure of the tension clamp has defects, the defect location and type are output; if it is judged that the internal structure of the tension clamp is intact, the qualified result of the tension clamp detection is output. The internal structure image and the detection result of the tension clamp are visualized so that the staff can intuitively view and analyze the internal structure quality of the tension clamp.

[0061] As a specific implementation of this embodiment, if the drone is affected by the crosswind when collecting images, the flight attitude parameters will record this change, and the convolutional neural network model will predict the type of distortion that may appear in the image based on these parameters and make corresponding corrections. For example, the model will identify the stretching or compression of the image in a certain direction, and restore the original proportion of the image through algorithm adjustment. After the image is corrected, in order to further improve the usability of the image, contrast enhancement processing is an indispensable step. By adjusting the brightness and contrast of the image, the details in the image can be made clearer and more visible. For example, by increasing the grayscale contrast of the image, the internal structural features of the tension clamp can be made more obvious, thereby facilitating subsequent defect detection. After extracting the internal structural features of the tension clamp, image processing technology can be used to analyze these features to determine whether there are defects. For example, if unusual dark or bright areas are found in a certain area of ​​the tension clamp, this may indicate that there are cracks or corrosion there. At this time, the system can automatically mark these areas and output specific defect types and location information based on the shape and size of the defects. Finally, the internal structure image and test results of the tension clamp are visualized to enable the staff to intuitively understand and analyze the internal condition of the tension clamp. Through image visualization technology, complex data can be converted into intuitive graphics or images, so that non-professionals can quickly understand the test results, thereby improving work efficiency and accuracy.

[0062] Step 3: Use the region growing algorithm to segment the image, extract the morphology, texture and other features of the defect area, identify the defect type through the support vector machine classifier, and obtain the defect location and classification results.

[0063] Furthermore, if Figure 2As shown, according to the preset region growing rule, the region growing algorithm is used to segment the input image to be detected to obtain the candidate defect region image. For the candidate defect region image, its morphological features and texture features are extracted to obtain the feature vector of the candidate defect region. The feature vector of the candidate defect region is input into the pre-trained support vector machine classifier, and the classifier is used to determine whether the candidate defect region is a real defect. If so, the position coordinates of the defect region are obtained to determine the specific location of the defect. According to the classification result of the support vector machine classifier, the defect type to which the defect region belongs is obtained to obtain the category information of the defect. The position coordinates of the defect and the category information are associated to construct the defect detection result, and the position and corresponding defect type of each defect in the image are output, that is, the first discrimination result. The texture features of the candidate defect region are further extracted and learned by using the convolutional neural network. The discriminability of the texture features is enhanced through the feature map output of the network, and the accuracy of defect classification is improved, that is, the second discrimination result. The discrimination results of the support vector machine classifier and the convolutional neural network are fused, and the results of the two methods are weighted and summed by setting the result threshold to obtain the final defect detection result, thereby improving the accuracy of defect location and classification.

[0064] As a specific implementation of this embodiment, the region growing algorithm is an image segmentation method that merges pixels with similar features together to form a region. The preset region growing rules usually include the selection of seed points and growth criteria. For example, the pixel with the highest gray value in the image is selected as the seed point, and then the gray value difference between the pixel and its neighboring pixels is less than a certain threshold to determine whether to merge the pixel into the current region. For the X-ray image of the tension clamp to be detected, first, the pixel whose gray value is significantly lower than the surrounding area in the image is selected as the seed point, because the defective area is usually manifested as an area with a lower gray value in the X-ray image. Then, a gray value difference threshold is set, such as 10. If the gray value difference between a pixel and its neighboring pixels is less than 10, the pixel is merged into the current region. By continuously iterating this process, all pixels that meet the growth criteria can eventually be merged together to form a candidate defective area image. For the segmented candidate defective area image, its morphological features and texture features need to be extracted. Morphological features can describe information such as the shape and size of the defect, such as area, perimeter, circularity, etc. Texture features can describe the surface texture information of defects, such as gray-level co-occurrence matrix, local binary pattern, etc. Assuming that the area of ​​a candidate defect area is 100 pixels, the perimeter is 40 pixels, the circularity is 0.8, the energy of the gray-level co-occurrence matrix is ​​0.5, the contrast is 0.2, and the correlation is 0.8, then the feature vector of the candidate defect area can be expressed as [100, 40, 0.8, 0.5, 0.2, 0.8]. Inputting the feature vector of the candidate defect area into the pre-trained support vector machine classifier can determine whether the candidate area is a real defect. Assume that a support vector machine classifier has been trained to distinguish between defects and non-defects. Input the above feature vector into the classifier. If the output result of the classifier is 1, it means that the candidate area is a real defect; if the output result is 0, it means that the candidate area is not a defect. If the support vector machine classifier determines that the candidate defect area is a real defect, it is necessary to obtain the position coordinates of the defect area, including the center point coordinates or the bounding box coordinates, in order to determine the specific location of the defect. Assuming that the coordinates of the center point of a defect area are (100,200), it means that the defect is located at the 100th row and 200th column of the image. At the same time, according to the classification results of the support vector machine classifier, the defect type to which the defect area belongs can be obtained, such as cracks, pores, inclusions, etc. Assuming that the defect area is classified as a crack, the category information of the defect is "crack". By associating the position coordinates of the defect with the category information, the defect detection results can be constructed, such as "there is a crack defect at the position (100,200)". In this way, the position of each defect in the image and the corresponding defect type can be output, which is convenient for staff to view and analyze. In order to improve the accuracy of defect classification, a convolutional neural network can be used to further extract and learn the texture features of the candidate defect area.Convolutional neural networks can automatically learn the features of images and have strong feature expression capabilities. Through the feature map output of convolutional neural networks, the discriminability of texture features can be enhanced, thereby improving the accuracy of defect classification. For example, the image block of the candidate defect area is input into the convolutional neural network, and the network automatically learns and extracts the texture features of the image, and then outputs a feature vector. This feature vector can better describe the texture information of the defect, thereby improving the classification accuracy. Finally, the accuracy of defect location and classification can be further improved by fusing the discrimination results of the support vector machine classifier and the convolutional neural network. For example, a result threshold is set, and the results of the two methods are weighted and summed. If the result of the weighted sum is greater than the result threshold, the candidate area is considered to be a real defect; otherwise, the candidate area is considered to be a non-defect. This fusion method can combine the advantages of the two methods and improve the accuracy of defect detection.

[0065] Step 4: Store the defect images and recognition results in a distributed database in a standardized format, and use hash indexing and data sharding technology to improve data retrieval and access efficiency.

[0066] Furthermore, if Figure 3 As shown, according to the preset defect image normalization format, the acquired defect image is format converted to obtain a normalized defect image. The normalized defect image is input into the pre-trained convolutional neural network model, and the defect recognition result is obtained through forward propagation calculation. According to the preset recognition result normalization format, the defect recognition result is format converted to obtain a normalized recognition result. According to the feature information of the defect image, the consistent hashing algorithm is used to calculate the hash index, and a mapping relationship is established between the hash index and the defect image. According to the feature information of the recognition result, the consistent hashing algorithm is used to calculate the hash index, and a mapping relationship is established between the hash index and the recognition result. According to the data distribution of the database, the data sharding technology is used to store the defect image and the recognition result in different data shards. In the distributed database, the defect image and the recognition result are quickly retrieved and accessed according to the hash index to improve the data processing efficiency.

[0067] As a specific implementation of this embodiment, the normalized defect image is to eliminate the differences in size, color, resolution, etc. between defect images from different sources, so as to facilitate subsequent processing and analysis. For example, the original defect images collected may be of different sizes, some of which are 1024x768 pixels, and some of which are 2048x1536 pixels. Through normalization, all images can be uniformly scaled to 256x256 pixels and converted into grayscale images. The advantage of doing so is that the amount of calculation can be reduced, the processing efficiency can be improved, and the feature extraction error caused by the difference in image size can be avoided. The conversion of color space is similar. The unified color space can avoid the interference caused by color differences. The pre-trained convolutional neural network model refers to a model that has been trained on a large-scale data set. For example, a ResNet50 model pre-trained on the ImageNet data set is used. Since these models have learned rich feature representations on a large amount of data, they can be directly used for defect recognition tasks without training from scratch. The normalized defect image is input into the pre-trained model, and the model automatically extracts the features of the image and outputs the probability of defect recognition. For example, the model may output that the probability of defect type A is 0.8 and the probability of defect type B is 0.2, then the defect is considered to belong to type A. The normalized format of the recognition results is to facilitate subsequent storage, retrieval and analysis. For example, the recognition results can be normalized into JSON format, which contains information such as defect type, location coordinates, and confidence. For example: {'type':'A', 'location':[100,50], 'confidence':0.8}. This normalized format facilitates data exchange and sharing between different systems. The consistent hashing algorithm can map defect images and recognition results to different data shards to achieve distributed storage of data. For example, suppose there are 10 data shards, each of which is responsible for storing a part of the data. Through the consistent hashing algorithm, the hash values ​​of the defect images and recognition results can be calculated, and then the data can be distributed to different shards according to the hash values. For example, if the hash value of image A is 1, it will be stored on shard 1; if the hash value of image B is 5, it will be stored on shard 5. The benefit of this is that it can achieve data load balancing and avoid excessive data volume in a certain shard, thereby improving the stability and performance of the system. Data sharding technology can store a large amount of data in different servers to improve the storage capacity and processing power of the system. Assuming that the database stores 1 million defect images and corresponding recognition results, if all the data is stored on one server, the server load will be too high, affecting the performance of the system. Through data sharding technology, this data can be stored in 10 servers, and each server is only responsible for storing 100,000 images and corresponding recognition results. The benefit of this is that it can improve the throughput and concurrent processing capabilities of the system.In a distributed database, defect images and recognition results can be quickly retrieved and accessed based on hash indexes. For example, to retrieve the recognition result of image A, you only need to calculate the hash value of image A, then find the corresponding shard based on the hash value, and then retrieve the data from the shard. Since the hash index has a high search efficiency, it can significantly improve data processing efficiency. For example, if you want to retrieve the recognition result of an image from 1 million images, if you use a sequential search method, you need to traverse all the data, which is very inefficient. However, using the hash index method, you only need to calculate the hash value and locate the data shard several times to find the target data, which is very efficient.

[0068] Step 5: When the newly collected tension clamp image is input into the system, its feature vector is extracted, and the most similar historical defect image is retrieved in the defect database through the K nearest neighbor algorithm to calculate the similarity score.

[0069] Furthermore, the input image of the tension clamp to be detected is obtained, and the image is preprocessed, including image enhancement, denoising and other operations, to improve the image quality. For the preprocessed tension clamp image, its texture, shape, color and other features are extracted to construct a multidimensional feature vector. According to the pre-constructed tension clamp defect image database, the K nearest neighbor algorithm is used to retrieve the K historical defect images that are most similar to the feature vector of the image to be detected. The similarity score between the tension clamp image to be detected and the K most similar historical defect images retrieved is calculated, and the higher the score, the greater the similarity. A similarity score threshold is set. If the highest similarity score exceeds the threshold, it is judged that the corresponding defect type exists in the tension clamp image to be detected. If the highest similarity score does not exceed the threshold, the tension clamp image to be detected is marked as normal and does not have defects. The defect judgment results, similarity scores and other information are stored in the database for subsequent statistical analysis and model optimization.

[0070] Feature vector representation: Assume that the feature vector of the tension clamp image to be detected is x = (x1, x2, ..., x n ), where X i Represents the value of the i-th feature, and n is the dimension of the feature.

[0071] Distance calculation: K nearest neighbor algorithm usually uses Euclidean distance to measure the similarity between two feature vectors. For two feature vectors x and y, the Euclidean distance d(x,y) between them is defined as:

[0072]

[0073] y i represents the i-th eigenvalue of the eigenvector y.

[0074] Similarity score calculation:

[0075]

[0076] Among them, d(x,y j ) is the feature vector x of the image to be detected and the feature vector y of the jth historical defect image j The Euclidean distance between .

[0077] Defect judgment: Set the similarity score threshold θ. If the highest similarity score exceeds θ, it is judged that the image of the tension clamp to be detected has the corresponding defect type. If the highest similarity score does not exceed θ, the image of the tension clamp to be detected is marked as normal and has no defects.

[0078] Step 6: If the similarity score exceeds the preset threshold, it is determined to be a known type defect and the defect type and location information is output; if the similarity score is lower than the threshold, it is marked as a suspected defect and added to the manual review queue for further confirmation by professionals.

[0079] Furthermore, feature information of the detected target defect is obtained, and similarity matching is calculated between the feature information and the defect features in the pre-built known defect feature library to obtain a similarity matching score; it is determined whether the similarity matching score exceeds a preset similarity threshold, and if so, the target defect is determined to be a known defect of the corresponding type; if the similarity matching score does not exceed the preset similarity threshold, the target defect is marked as a suspected defect, and its related information is added to the manual review task queue; for the target defect determined to be a known defect, its corresponding defect type and location information is obtained from the known defect feature library, and the type, location, similarity score and other information of the defect are output; for the suspected defects added to the manual review task queue, professionals further determine whether they are real defects and the defect type based on their image information, similarity score, etc.; based on the results of the manual review, the suspected defects confirmed to be real defects are added to the known defect feature library, and the corresponding defect type and location information are updated; the known defect feature library is continuously updated iteratively to improve the accuracy of similarity matching, while optimizing the manual review process to improve the efficiency of defect detection and identification.

[0080] As a specific implementation of this embodiment, texture features (such as gray level co-occurrence matrix, local binary pattern, etc.), shape features (such as Hu moment, Fourier descriptor, etc.) and color features (such as color histogram, color moment, etc.) of the defect area are extracted. The extracted target defect features are matched with the pre-built known defect feature library for similarity matching calculation. The known defect feature library stores feature vectors of various known defect types, as well as the corresponding defect types and location information. Similarity matching can use a variety of algorithms, such as Euclidean distance, cosine similarity, Manhattan distance, etc. Assuming that the feature vector of the target defect is [0.2, 0.5, 0.8], the feature vector of the known defect A is [0.3, 0.6, 0.7], and the feature vector of the known defect B is [0.1, 0.2, 0.3], the similarity is calculated using the Euclidean distance, then the similarity between the target defect and defect A is 0.14, and the similarity with defect B is 0.58. Determine whether the similarity matching score exceeds the preset similarity threshold. The similarity threshold is a pre-set value used to determine whether the target defect is similar enough to the known defect. The threshold setting needs to be adjusted according to the actual situation. If the threshold is too high, some real defects may be missed; if the threshold is too low, some normal areas may be misdetected. Assuming that the similarity threshold is 0.5, the similarity between the target defect and defect A does not exceed the threshold, while the similarity with defect B exceeds the threshold. If the similarity matching score exceeds the preset similarity threshold, the target defect is determined to be a known defect of the corresponding type. For example, if the similarity between the target defect and defect B exceeds the threshold, the target defect is determined to be the same defect type as defect B. In this way, known defects can be quickly identified and corresponding treatment solutions can be provided. If the similarity matching score does not exceed the preset similarity threshold, the target defect is marked as a suspicious defect, and its related information is added to the manual review task queue. For example, if the similarity between the target defect and defect A does not exceed the threshold, it is marked as a suspicious defect, and the original image, feature vector, similarity score and other information are added to the manual review task queue, waiting for further judgment by professionals. Doing so can avoid missing some defects that are difficult to automatically identify and improve the accuracy of defect detection. For target defects that are determined to be known defects, the corresponding defect type and location information are obtained from the known defect feature library, and the type, location, similarity score and other information of the defect are output. For example, if the target defect is determined to be the same defect type as defect B, the type (e.g., loose bolts) and location information (e.g., upper left corner of the tension clamp) of defect B are obtained from the known defect feature library, and information such as "defect type: loose bolts, location: upper left corner of the tension clamp, similarity score: 0.58" is output to facilitate maintenance personnel to quickly locate and handle defects. For suspicious defects added to the manual review task queue, professionals will further determine whether it is a real defect and the defect type based on its image information, similarity score, etc.For example, professionals can view the original image of the target defect, and combine its feature vector and similarity score to determine whether it is a real defect, as well as the specific defect type (for example, rust). According to the results of manual review, the suspected defects confirmed as real defects are added to the known defect feature library, and the corresponding defect type and location information are updated. For example, if the professional confirms that the target defect is rust, the feature vector, defect type (rust) and location information of the target defect are added to the known defect feature library so that subsequent similarity matching can identify this type of defect. Continuously iterate and update the known defect feature library to improve the accuracy of similarity matching, while optimizing the manual review process to improve the efficiency of defect detection and identification. By continuously accumulating new defect samples and updating and optimizing the known defect feature library, the accuracy of similarity matching can be improved, the workload of manual review can be reduced, and the efficiency of defect detection and identification can be improved. For example, with the increase of rust samples in the known defect feature library, the system can more accurately identify rust defects of different degrees and forms.

[0081] Step 7. According to the type and severity of the defect, a health assessment report of the tension clamp is automatically generated, and the corresponding defect warning and maintenance work order are triggered, and the operation and maintenance personnel are notified to handle it to ensure the safe and stable operation of the transmission line.

[0082] Furthermore, the defect type and severity data of the tension clamp are obtained, and the health status of the tension clamp is judged by comparing with the preset defect type and severity threshold. According to the judged health status of the tension clamp, a corresponding health assessment report is automatically generated, which describes in detail the specific defect type, severity and comprehensive health status of the tension clamp. If the health status judgment result of the tension clamp is lower than the preset health threshold, a defect warning is triggered, and the warning information is automatically sent to the relevant operation and maintenance personnel. At the same time as the defect warning is triggered, a corresponding maintenance work order is automatically generated, which contains detailed information such as the location, defect type, and severity of the defective tension clamp. The generated maintenance work order is automatically issued to the corresponding operation and maintenance personnel, and the operation and maintenance personnel are notified by SMS, email, etc. to deal with the defective tension clamp in time. The operation and maintenance personnel repair or replace the defective tension clamp according to the instructions of the maintenance work order, and enter the maintenance results into the system after completion. The system automatically updates the health status of the tension clamp and stores the maintenance process data in the transmission line operation and maintenance management database, so as to analyze the safe and stable operation of the transmission line in the future.

[0083] As a specific implementation of this embodiment, this embodiment obtains an image of a tension clamp through drone inspection, and the image recognition algorithm identifies that the tension clamp has a bolt loosening defect, and the severity is slight. At the same time, the sensor monitoring data shows that the temperature of the tension clamp is slightly increased. After obtaining the defect data, it is necessary to compare it with the preset defect type and severity threshold to determine the health status of the tension clamp. The preset threshold needs to be set according to the actual situation. For example, the severity threshold of the bolt loosening defect can be set to three levels: slight, medium and severe. For example: Assume that the preset severity threshold of the bolt loosening defect is: slight-healthy, medium-sub-healthy, severe-unhealthy. Then, according to the defect data obtained in the above example, the health status of the tension clamp is judged to be healthy. According to the judged health status of the tension clamp, the corresponding health assessment report is automatically generated. The report describes in detail the specific defect type, severity and comprehensive health status of the tension clamp.

[0084] In view of the above actual situation, the generated health assessment report includes the following contents: Strain clamp number: XXX, defect type: loose bolts, severity: minor, overall health status: healthy, and continuous attention is recommended. If the health status judgment result of the strain clamp is lower than the preset health threshold, a defect warning is triggered. The warning information contains detailed information such as the location, defect type, and severity of the defective strain clamp, and is automatically sent to relevant operation and maintenance personnel through SMS, email, etc.

[0085] Assuming that the health status of another tension clamp is judged to be unhealthy, the system will automatically trigger a defect warning and send a warning message to the operation and maintenance personnel: tension clamp number: YYY, location: XX tower of XXX line, defect type: bolt fracture, severity: severe, please handle immediately. At the same time as the defect warning is triggered, the corresponding maintenance work order is automatically generated. The work order contains detailed information such as the location, defect type, and severity of the defective tension clamp, and is automatically sent to the corresponding operation and maintenance personnel.

[0086] For the above unhealthy tension clamps, the system will automatically generate a maintenance work order, which includes: tension clamp number: YYY, location: XX tower of XXX line, defect type: bolt breakage, severity: severe, maintenance requirement: bolt replacement. The operation and maintenance personnel will repair or replace the defective tension clamp according to the instructions of the maintenance work order, and enter the maintenance results into the system after completion. The system automatically updates the health status of the tension clamp and stores the maintenance process data in the transmission line operation and maintenance management database.

[0087] After receiving the work order, the operation and maintenance personnel went to the site to replace the broken bolts and entered the maintenance results into the system. The system updated the health status of the tension clamp to healthy and stored information such as the maintenance time and the model of the replaced bolts into the database. Storing the maintenance process data in the database can facilitate subsequent analysis of the safe and stable operation of the transmission line, such as analyzing the occurrence patterns of different types of defects and the health status of tension clamps in different regions, providing data support for the operation and maintenance of the transmission line.

[0088] The present invention discloses a method for detecting defects in power transmission line tension clamps using drones and X-ray imaging. The method optimizes X-ray imaging parameters to achieve low-dose, high-quality imaging; uses neural networks to correct and enhance images; uses regional growing and support vector machine algorithms to achieve defect detection and classification; builds a distributed defect database to support efficient retrieval; combines the K nearest neighbor algorithm to achieve similarity matching of newly acquired images, automatically identifies known defects or marks suspicious defects; and finally generates a health assessment report based on the defect situation and triggers a maintenance process. The present invention integrates a variety of intelligent algorithms to achieve automated non-destructive testing of the internal structure of the tension clamp, can quickly and accurately discover potential defects, and effectively ensure the safe and stable operation of the transmission line, and has important engineering application value.

[0089] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for detecting defects in power transmission line tension clamps based on drones and X-ray imaging, characterized in that: The following steps are involved: Based on the material and thickness information of the tension clamp, the imaging parameter combination of the X-ray imaging device is optimized to obtain the optimized X-ray imaging device; The optimized X-ray imaging device is equipped with a drone to obtain an image of the internal structure of the tension clamp; Obtaining a defect image and a defect recognition result based on the internal structure image of the tension clamp; Building a distributed database based on the defect image and the defect recognition result; The newly acquired image of the tension clamp is compared and analyzed with the information in the distributed database to generate a health assessment report of the tension clamp.

2. The method according to claim 1, characterized in that The process of optimizing the imaging parameter combination of the X-ray imaging device based on the material and thickness information of the tension clamp includes: Obtain the ray energy range based on the material information of the tension clamp; The imaging time range is obtained based on the thickness information of the tension clamp; Setting constraints for the machine learning algorithm based on the ray energy range and imaging time range; Iterating the machine learning algorithm based on the constraint conditions to obtain optimal ray energy and imaging time; The optimal operating parameters of the X-ray imaging device are set based on the optimal ray energy and imaging time.

3. The method according to claim 1, characterized in that The optimized X-ray imaging device is equipped with a drone to construct an image correction model by acquiring the flight parameters of the drone, and the collected X-ray image is processed based on the image correction model to obtain the internal structure image of the tension clamp, and the internal structure image of the tension clamp is preliminarily marked.

4. The method according to claim 3, characterized in that The process of obtaining an image of the internal structure of the tension clamp and making preliminary markings includes: Performing distortion correction on the X-ray image to obtain a corrected image; Performing contrast enhancement processing on the corrected image to obtain an enhanced image; Obtaining an internal structure image of the tension clamp based on the enhanced image; The internal structural features of the internal structural image of the tension clamp are extracted, and when the internal structural features are dark areas, they are marked as defects.

5. The method according to claim 4, characterized in that The process of identifying defects in the internal structure image of the tension clamp includes: Perform image segmentation on the marked internal structure image of the tension clamp to obtain a candidate defect area image; Extracting feature vectors of the candidate defect region image to obtain morphological features and texture features; Inputting the feature vector into a support vector machine classifier to perform defect judgment to obtain a first discrimination result, wherein the first discrimination result includes a defect type and a defect location coordinate; Inputting the texture feature into a convolutional neural network to perform texture feature enhancement to obtain a second discrimination result; The first discrimination result and the second discrimination result are combined to obtain a defect recognition result.

6. The method according to claim 5, characterized in that Based on the region growing algorithm, the image of the marked internal structure of the tension clamp is segmented to obtain a candidate defect region image.

7. The method according to claim 6, characterized in that The process of fusing the first discrimination result and the second discrimination result includes: Set result thresholds; Performing a weighted summation on the first discrimination result and the second discrimination result to obtain an actual value; When the actual value is greater than the result threshold, the current candidate area is considered to be a real defect; When the actual value is less than or equal to the result threshold, the current candidate area is considered to be non-defective.

8. The method according to claim 7, characterized in that The process of building a distributed database includes: Converting the defect image and the defect recognition result into a standardized format to obtain a standardized defect image and a standardized defect recognition result; According to the feature information of the normalized defect image, a consistent hashing algorithm is used to calculate the hash index of the defect image, and a mapping relationship is established between the hash index of the defect image and the defect image; According to the characteristic information of the normalized defect identification result, a consistent hashing algorithm is used to calculate the hash index of the defect identification result, and a mapping relationship is established between the hash index of the defect identification result and the defect identification result; Building a distributed database based on data sharding technology to store defect images and defect recognition results; The hash index based on the defect image and the hash index of the defect recognition result are retrieved and accessed in the distributed database.

9. The method according to claim 8, characterized in that The process of health assessment of newly acquired tension clamp images based on a distributed database includes: Set similarity score threshold; A K-nearest neighbor algorithm is used to calculate a similarity score between a newly acquired tension clamp image and a historical defect image in the distributed database; When the similarity score is greater than the similarity score threshold, the newly acquired tension clamp image has the corresponding defect type; When the similarity score is less than or equal to the similarity score threshold, the newly acquired tension clamp image does not have a corresponding defect.

Images