Image analysis-based transmission line strain clamp defect analysis method, system, device and medium

Abstract

The application relates to the technical field of defect detection, and discloses a power transmission line strain clamp defect analysis method, system, equipment and medium based on image analysis, which comprises the following steps: acquiring multi-dimensional image data by using a multi-view and multi-spectrum imaging device; constructing a feature extraction architecture to perform feature extraction; constructing a reinforcement learning model to realize analysis and decision-making; constructing a target detection screening model comprising multiple prediction heads, adopting an anchor frame generation strategy and a non-maximum suppression algorithm to perform defect target detection and positioning on the image features after analysis; constructing a classification evaluation architecture based on a strain clamp defect type feature database and a classification decision tree to perform category determination and severity evaluation on the detected defect targets; and outputting and storing the defect analysis results in a preset data format and storage strategy. The application overcomes the problems of insufficient complex defect feature extraction and rigid decision mechanism of traditional methods, and comprehensively improves the detection capability and operation and maintenance efficiency of the strain clamp defects.

Description

Technical Field

[0001] This invention relates to the field of defect detection and analysis technology, and in particular to a method, system, equipment and medium for defect analysis of transmission line tension clamps based on image analysis. Background Technology

[0002] With the continuous growth of power transmission demand, the safe and stable operation of transmission lines is of paramount importance. Tension clamps, as key components ensuring power transmission, directly affect the reliability of the lines. In traditional power operation and maintenance, defect detection of tension clamps mainly relies on manual inspection, which is not only inefficient but also limited by the subjective judgment and experience of inspection personnel, making it difficult to conduct comprehensive and accurate inspections of tension clamps in complex environments.

[0003] With the development of computer vision and artificial intelligence technologies, image analysis-based detection methods are gradually being applied to the field of transmission line inspection. However, existing image analysis-based tension clamp defect detection technologies still have significant shortcomings. On the one hand, in the feature extraction and analysis stage, traditional methods often employ a single perspective and fixed algorithms, making it difficult to effectively extract features of the complex structure and diverse defects of tension clamps. Their detection capabilities for minute and hidden defects are weak, failing to meet the demands for high-precision inspection. On the other hand, the decision-making and evaluation mechanisms of existing detection systems are not intelligent enough, lacking comprehensive consideration of multiple parameters of tension clamps. They cannot adaptively and accurately judge based on different operating conditions and defect types, resulting in poor reliability and stability of the detection results, making it difficult to meet the growing requirements of intelligent operation and maintenance in power systems. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a method, system, equipment, and medium for analyzing defects in transmission line tension clamps based on image analysis. This solves the problems of existing technologies that mainly rely on manual inspection and image analysis from a single perspective, making it difficult to comprehensively and accurately detect the complex structure and diverse defects of transmission line tension clamps, especially the insufficient ability to identify minute and hidden defects, resulting in low detection efficiency and poor reliability.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a method for analyzing defects in transmission line tension clamps based on image analysis, including:

[0008] Multi-view, multi-spectral imaging equipment is used with a preset spatial array layout and spectral response band combination to collect image information of the tension clamps of the transmission line and obtain multi-dimensional image data.

[0009] A feature extraction architecture is constructed, and features are extracted from the multidimensional image data using a preset combination of convolution kernel parameters and pooling strategy;

[0010] Based on the improved deep network structure, combined with the material properties and structural size parameters of the tension clamp, a reinforcement learning model is constructed, and the extracted image features are analyzed and decided through a preset neural network weight update mechanism and experience playback strategy.

[0011] Based on the size specifications and installation angle parameters of the tension clamp, a target detection and screening model containing multiple prediction heads is constructed. An anchor frame generation strategy and non-maximum suppression algorithm are used to detect and locate defect targets in the analyzed image features.

[0012] By constructing a classification and evaluation architecture based on a database of defects in tension clamps and a classification decision tree, the detected defect targets are classified and their severity is assessed to obtain defect analysis results.

[0013] The defect analysis results are output and stored in a preset data format and storage strategy.

[0014] As a preferred embodiment of the image analysis-based method for analyzing defects in transmission line tension clamps according to the present invention, the feature extraction of the multidimensional image data includes:

[0015] Based on the resolution and data volume of the multidimensional image data, determine the number of layers in the neural network and the number of nodes in each layer;

[0016] The convolutional layer of the feature extraction architecture first initializes the convolutional kernel parameters and trains them using historical tension clamp image data. It continuously adjusts the weights and biases of the convolutional kernel parameters to extract different levels of features from the multidimensional image data.

[0017] The pooling layer of the feature extraction architecture selects the pooling type and pooling window size according to the scale of the different levels of features and the limitations of computing resources, and performs further feature extraction on the multidimensional image data to obtain the edge features, shape features, texture features and spectral features of the tension clamp image at different scales and dimensions.

[0018] As a preferred embodiment of the image analysis-based transmission line tension clamp defect analysis method of the present invention, the construction of the reinforcement learning model includes:

[0019] Based on the improved deep network structure, and combined with the material properties and structural size parameters of the tension clamp, the state space, action space and reward function are defined.

[0020] The constructed state space S is: ,in, Indicates the first The first feature dimension Each quantified level status, The number of feature dimensions for the tension clamp image is determined based on the structural complexity of the tension clamp and the image acquisition resolution. The number of quantization levels for characteristic values ​​is determined by the accuracy requirements of the tension clamp parameters and the data processing capability.

[0021] The constructed action space A is based on the defect detection requirements of tension clamps and the system processing capabilities, and includes a series of decision actions for image feature analysis;

[0022] Constructed reward function for: ,in, The number of reward factors is set according to the type of defect in the tension clamp and the detection target; The weight of the reward factor is determined based on the importance of the tension clamp parameters to defect detection; In the state Next action The reward value corresponding to the reward factor at that time;

[0023] The improved deep network structure is trained using historical tension clamp image data and defect annotation data. During the training process, the network is continuously allowed to interact with the environment, that is, it is inputting extracted image features, performing actions, obtaining reward feedback, and updating the weight parameters of the neural network using gradient descent based on the reward function.

[0024] As a preferred embodiment of the image analysis-based method for analyzing defects in transmission line tension clamps according to the present invention, the defect target detection and localization includes:

[0025] Based on the size specifications and installation angle parameters of the tension clamp, a target detection and screening model containing multiple prediction heads is constructed, with each prediction head responsible for detecting defective targets within a specific scale range;

[0026] Using a preset anchor frame generation strategy, anchor frames of different sizes and aspect ratios are generated as initial target candidate regions based on the common position distribution, standard size, and detection accuracy requirements of tension clamps in the image; the center position, width, and height parameters of the anchor frames are set according to the actual size characteristics of the tension clamps.

[0027] The image features after feature extraction and decision analysis are input into the target detection and screening model, and the predicted bounding box coordinates, category confidence and category probability of the defective target are output.

[0028] The prediction results are processed using a non-maximum suppression algorithm. The cross-union ratio between each predicted bounding box is calculated. Based on the preset overlap threshold and confidence threshold, the predicted boxes with high confidence and low overlap are retained to suppress redundant and erroneous detection results.

[0029] Based on the final preserved prediction bounding box information, the precise location coordinates and size of the defective target in the image are determined, thus completing the detection and localization of the defective target.

[0030] As a preferred embodiment of the image analysis-based transmission line tension clamp defect analysis method of the present invention, the data transmission optimization during defect target detection and localization is expressed as follows:

[0031] ;

[0032] in, For data transmission time, The transmission efficiency adjustment coefficient is determined based on the system hardware performance and network environment; The amount of feature data is determined by the resolution of the tension clamp image and the dimension of feature extraction; For data transmission bandwidth; The number of transmission tasks is determined based on the real-time requirements of tension clamp defect detection and the system's processing capacity. The number of parallel transmission channels is set based on the system hardware configuration and the characteristics of the tension clamp image data. This indicates the transmission time of a single task.

[0033] As a preferred embodiment of the image analysis-based method for analyzing defects in transmission line tension clamps according to the present invention, the defect analysis results obtained include:

[0034] Establish a database containing characteristic information of various common defect types;

[0035] Extract the feature information of the detected defect targets, compare and analyze it with the standard features in the defect type feature database, and calculate the feature matching degree;

[0036] Based on the feature matching degree, the category of the defect is determined using a classification decision tree; starting from the root node of the classification decision tree, the features of the defect target are compared with the judgment conditions of the decision tree nodes, and the tree is divided downwards step by step to finally determine the specific category of the defect.

[0037] The node splitting criterion of the classification decision tree adopts an improved information gain ratio calculation method, which comprehensively considers the information gain of the attribute on the sample set and the dispersion of the attribute values.

[0038] As a preferred embodiment of the image analysis-based method for analyzing defects in transmission line tension clamps according to the present invention, the severity assessment includes:

[0039] Based on the structural safety requirements and defect influencing factors of tension clamps, evaluation indicators are determined and weights are assigned to each evaluation indicator. The weight values ​​are allocated according to the importance of the tension clamp parameters to structural safety.

[0040] Calculate the score of each evaluation indicator based on the specific characteristics of the defective target;

[0041] The severity assessment value of the defect is calculated by weighted summation based on the scores of each assessment indicator.

[0042] Secondly, the present invention provides a transmission line tension clamp defect analysis system based on image analysis, comprising:

[0043] The multi-dimensional image acquisition unit is used to acquire image information of the tension clamps of the transmission line using a preset spatial array layout and spectral response band combination, and obtain multi-dimensional image data.

[0044] The image feature extraction unit is used to construct a feature extraction architecture and extract features from the multidimensional image data using a preset combination of convolutional kernel parameters and pooling strategy.

[0045] The deep network decision unit is used to build a reinforcement learning model based on the improved deep network structure, combined with the material properties and structural size parameters of the tension clamp, and to analyze and make decisions on the extracted image features through a preset neural network weight update mechanism and experience playback strategy.

[0046] The defect target screening unit is used to construct a target detection screening model containing multiple prediction heads based on the size specifications and installation angle parameters of the tension clamp. It uses an anchor frame generation strategy and a non-maximum suppression algorithm to detect and locate defect targets based on the analyzed image features.

[0047] The defect classification and evaluation unit is used to determine the category and severity of detected defects by constructing a classification and evaluation architecture based on a database of defects in tension clamps and a classification decision tree, and to obtain defect analysis results.

[0048] The results are output and stored in the storage unit, where the defect analysis results are output and stored in a preset data format and storage strategy.

[0049] Thirdly, the present invention provides an electronic device, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor executes the computer-executable instructions to implement the steps of a method for analyzing defects in transmission line tension clamps based on image analysis.

[0050] Fourthly, the present invention provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of an image analysis-based method for analyzing defects in transmission line tension clamps.

[0051] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention uses multi-view, multi-spectral imaging equipment to acquire multi-dimensional image data and constructs a feature extraction architecture, which can accurately capture the edge, shape, texture, and spectral features of tension clamps at different scales and dimensions. Compared with single-view and fixed algorithms, it greatly improves the detection capability of small and hidden defects and enhances the distinguishability of feature expressions. Existing systems suffer from insufficient intelligence in the decision-making and evaluation stages. This invention, through an improved deep Q-network structure, combines the material properties and structural parameters of the tension clamp to construct a state space and reward function, achieving intelligent analysis and adaptive decision-making of image features. Simultaneously, based on a multi-predictor target detection model and a dedicated anchor box generation strategy, it accurately detects and locates defect targets, improving the accuracy and robustness of defect location. Through the fusion of feature database and classification decision tree, it achieves accurate determination of defect types and refined assessment of severity. In this invention, the various units work collaboratively, overcoming the problems of insufficient feature extraction for complex defects and rigid decision-making mechanisms in traditional methods, comprehensively improving the detection capability and maintenance efficiency of tension clamp defects. Attached Figure Description

[0052] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0053] Figure 1 This is a schematic diagram of the overall process logic of a method for analyzing the defects of transmission line tension clamps based on image analysis, according to an embodiment of the present invention. Detailed Implementation

[0054] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0055] Example 1, referring to Figure 1As one embodiment of the present invention, a method for analyzing defects in transmission line tension clamps based on image analysis is provided, such as... Figure 1 The specific steps shown are as follows:

[0056] S100: Employs multi-view, multi-spectral imaging equipment with a preset spatial array layout and spectral response band combination to acquire image information of tension clamps of transmission lines and obtain multi-dimensional image data.

[0057] S200: Construct a feature extraction architecture and use preset convolution kernel parameter combinations and pooling strategies to extract features from multidimensional image data;

[0058] S300: Based on an improved deep network structure, combined with the material properties and structural size parameters of the tension clamp, a reinforcement learning model is constructed, and the extracted image features are analyzed and decided through a preset neural network weight update mechanism and experience playback strategy.

[0059] S400: Based on the size specifications and installation angle parameters of the tension clamp, a target detection and screening model containing multiple prediction heads is constructed. An anchor frame generation strategy and non-maximum suppression algorithm are used to detect and locate defect targets in the analyzed image features.

[0060] S500: By constructing a classification and evaluation architecture based on a database of defects in tension clamps and a classification decision tree, the detected defect targets are classified and their severity is assessed to obtain defect analysis results.

[0061] S600: Outputs and stores the defect analysis results in a preset data format and storage strategy.

[0062] It should be noted that, to address the shortcomings of existing technologies that rely primarily on manual inspections and single-view image analysis, which struggle to comprehensively and accurately detect the complex structure and diverse defects of transmission line tension clamps, particularly their insufficient ability to identify minute and hidden defects, resulting in low detection efficiency and poor reliability, steps S100-S600 employ multi-view, multi-spectral imaging equipment to acquire multi-dimensional image data and construct a feature extraction architecture. This architecture can accurately capture the edge, shape, texture, and spectral features of tension clamps at different scales and dimensions. Compared to single-view and fixed algorithms, this significantly improves the detection capability for minute and hidden defects and enhances the distinguishability of feature representations. Existing systems suffer from insufficient intelligence in the decision-making and evaluation stages. This invention, through an improved deep Q-network structure, combines the material properties and structural parameters of the tension clamps to construct a state space and reward function, achieving intelligent analysis and adaptive decision-making of image features. Simultaneously, based on a multi-predictor target detection model and a dedicated anchor box generation strategy, defect targets are accurately detected and located, improving the accuracy and robustness of defect localization. By integrating a feature database with a classification decision tree, accurate determination of defect types and refined assessment of severity are achieved. In this invention, the various units work collaboratively, overcoming the shortcomings of traditional methods in extracting complex defect features and rigid decision-making mechanisms, thus comprehensively improving the detection capability and maintenance efficiency of tension clamp defects.

[0063] In this embodiment of the invention, step S100 above uses a multi-view, multi-spectral imaging device with a preset spatial array layout and spectral response band combination to collect image information of the tension clamp of the transmission line and obtain multi-dimensional image data.

[0064] Specifically, multi-view, multi-spectral imaging equipment is used, operating through spatial array layout and spectral response band combinations. Multi-view imaging equipment can photograph the tension clamps of transmission lines from different angles, avoiding blind spots caused by a single viewpoint and comprehensively capturing information from all parts of the tension clamp surface; multi-spectral imaging can utilize the differences in sensitivity of different spectral bands to the tension clamp material and defects to obtain more detailed information that is difficult to detect with the naked eye.

[0065] Specifically, the spatial array layout is determined based on the structural characteristics of the tension clamp and the actual installation environment to ensure that the imaging equipment can cover the key areas of the tension clamp; the combination of spectral response bands is selected based on the material properties of the tension clamp and the spectral characteristics of common defects, so as to achieve effective acquisition of multi-dimensional image data such as surface texture, geometric contour and structural details of the tension clamp, providing a rich and accurate data foundation for subsequent analysis and processing.

[0066] Specifically, multi-view imaging equipment and multispectral imaging equipment are rationally combined and installed. For example, on transmission line towers, multiple cameras with different viewing angles are installed according to the installation position and angle of the tension clamp, forming a shooting array around the tension clamp to ensure that all sides of the tension clamp can be captured from all directions. For multispectral imaging equipment, appropriate spectral ranges, such as visible light and near-infrared light bands, are selected and precisely calibrated to ensure the accuracy and consistency of the acquired spectral image data.

[0067] In this embodiment of the invention, step S200, which constructs a feature extraction architecture and uses a preset combination of convolutional kernel parameters and pooling strategy to extract features from multidimensional image data, includes the following sub-steps B1 to B3:

[0068] In B1: Determine the number of layers and the number of nodes in each layer of the neural network based on the resolution and data volume of the multidimensional image data;

[0069] In B2: The convolutional layer of the feature extraction architecture first initializes the convolutional kernel parameters and trains them using historical tension clamp image data. It continuously adjusts the weights and biases of the convolutional kernel parameters to extract different levels of features from multidimensional image data.

[0070] Specifically, the calculation formula for the convolutional layer of the feature extraction architecture is as follows:

[0071] ;

[0072] in, This represents the i-th feature map of the l-th layer; For the first Number of feature maps in a layer; Connect the l-th layer to the l-th layer. A convolution kernel for the i-th input feature map and the i-th output feature map; * indicates a convolution operation; For the bias term of the i-th feature map in the l-th layer, Indicates the first The j-th feature map of the layer.

[0073] In B3: The pooling layer of the feature extraction architecture selects the pooling type and pooling window size according to the scale of different levels of features and the limitations of computing resources, and performs further feature extraction on multidimensional image data to obtain the edge features, shape features, texture features and spectral features of tension clamp images at different scales and dimensions;

[0074] Specifically, the calculation formula for the pooling layer in the feature extraction architecture is as follows:

[0075] ;

[0076] in, The pooled feature map is represented by Pooling, which indicates the use of a preset pooling strategy. This strategy is determined based on the feature scale of the tension clamp image and the computational resource requirements.

[0077] Specifically, the convolutional layer computation involves multi-layer convolutional kernels operating on the feature maps of the previous layer and superimposing biases to extract features at different levels from the tension clamp image. The pooling layer, based on the image feature scale and computational resources, employs a pooling strategy to reduce data dimensionality. In implementation, a multi-layer neural network is constructed, and parameters such as the number, size, stride, and pooling window of convolutional kernels are adjusted according to the characteristics of the tension clamp image data. The network is trained and optimized with a large amount of image data to efficiently extract features such as edges, shapes, and textures from the tension clamp image.

[0078] In this embodiment of the invention, the number of network layers and the number of nodes in each layer are determined based on the resolution and data volume of the tension clamp images. The convolutional kernel parameters are initialized, and the network is trained using a large amount of tension clamp image data. The backpropagation algorithm is used to continuously adjust the weights and biases of the convolutional kernels, optimizing the network parameters so that the network can better extract features from the tension clamp images. The pooling operation selects an appropriate pooling type (such as max pooling, average pooling, etc.) and pooling window size based on the scale of the image features and the limitations of computational resources, improving the network's processing efficiency while ensuring no loss of feature information. The extracted image features are output in a specified data format.

[0079] It should be noted that the multi-level, multi-scale feature extraction architecture constructed in step S200 above processes the input multi-dimensional image data layer by layer through preset convolutional kernel parameter combinations and pooling strategies. The multi-level architecture can analyze images at different depths. Shallow networks can extract basic features such as edges and colors from tension clamp images. As the layers deepen, the network can gradually extract more complex and abstract features such as shapes and textures. The multi-scale strategy can capture features in image regions of different sizes, focusing on both the overall structural features of the tension clamp and detecting local features of minor defects. The convolutional kernel parameter combinations are optimized according to the characteristics of the tension clamp images and detection requirements. Different convolutional kernel sizes, numbers, and weights can extract different types of features. The pooling strategy is used to reduce data dimensionality and computational load while retaining key image features, thereby achieving effective extraction of edge features, shape features, texture features, and spectral features of tension clamp images at different scales and dimensions, providing representative image feature information for subsequent decision analysis.

[0080] In this embodiment of the invention, step S300, based on an improved deep network structure and combined with the material properties and structural size parameters of the tension clamp, constructs a reinforcement learning model, and analyzes and makes decisions on the extracted image features through a preset neural network weight update mechanism and experience replay strategy, including the following sub-steps C1 and C2:

[0081] In C1: Based on the improved deep network structure, combined with the material properties and structural size parameters of the tension clamp, the state space, action space and reward function are defined;

[0082] Specifically, the constructed state space S is as follows: ,in, Indicates the first The quantization level state of each feature dimension. The number of feature dimensions for the tension clamp image is determined based on the structural complexity of the tension clamp and the image acquisition resolution. The number of quantization levels for characteristic values ​​is determined by the accuracy requirements of the tension clamp parameters and the data processing capability.

[0083] Specifically, the constructed action space A is based on the defect detection requirements of tension clamps and the system processing capabilities, and includes a series of decision actions for image feature analysis;

[0084] Specifically, the constructed reward function for: ,in, The number of reward factors is set according to the type of defect in the tension clamp and the detection target; The weight of the reward factor is determined based on the importance of the tension clamp parameters to defect detection; In the state Next action The reward value corresponding to the time reward factor.

[0085] It should be noted that when constructing the state space, the number of feature dimensions is determined based on the structural complexity of the tension clamp and the image acquisition resolution. The quantization level of feature values ​​is set in conjunction with parameter accuracy and data processing capabilities, discretizing the image features into computable states. The action space designs decision-making actions around the defect detection requirements. The reward function comprehensively measures the merits of different state-action combinations by setting multiple reward factors and assigning weights. During implementation, a large amount of tension clamp images and defect data are first collected. During training, the parameters of each space and function are dynamically adjusted so that the network learns the optimal decision-making strategy for tension clamp defect detection.

[0086] In C2: The improved deep network structure is trained using historical tension clamp image data and defect annotation data. During the training process, the network is continuously allowed to interact with the environment, that is, the extracted image features are input, actions are executed, and reward feedback is obtained. The weight parameters of the neural network are updated using gradient descent based on the reward function.

[0087] Specifically, the weight parameters of a neural network The update formula is:

[0088] ;

[0089] in, The learning rate is determined based on the complexity of the tension clamp defect detection task and the system training efficiency. The target value is calculated based on the defect detection results of the tension clamp and the reward function; In the state Next action Action value function at time; The action value function with respect to the weight parameters The gradient.

[0090] Specifically, the learning rate is set based on the complexity of the tension clamp defect detection task and training efficiency. The weight parameters are adjusted by the difference between the target value and the action value function, and an experience replay strategy is used to avoid the influence of data correlation. During implementation, tension clamp image features are continuously input during training, and the weights are updated based on reward feedback to gradually optimize network performance, enabling the decision unit to accurately analyze tension clamp image features and make decisions.

[0091] It should be noted that step S300 above is based on an improved deep Q-network structure, combined with parameters such as the material properties and structural dimensions of the tension clamp, to construct a state space, action space, and reward function. The state space describes various combinations of image features of the tension clamp. Based on the structural complexity of the tension clamp and the image acquisition resolution, the image features are divided into multiple dimensions and quantization levels. Each dimension represents a feature attribute of the tension clamp, and each quantization level represents a different value range for that feature attribute. The action space contains a series of decision-making actions for image feature analysis. These actions are designed based on the defect detection requirements of the tension clamp and the system's processing capabilities, such as further analyzing features in specific areas and determining whether defects exist. The reward function is set according to the defect type and detection target of the tension clamp, and is used to evaluate the merits of actions performed under different states. By weighting the importance of the tension clamp parameters to defect detection, the network is guided to learn the optimal decision-making strategy. Simultaneously, through the neural network weight update mechanism and experience replay strategy, the extracted image features are analyzed and decisions are made, continuously optimizing the decision-making process and improving the accuracy and reliability of tension clamp defect judgment.

[0092] In this embodiment of the invention, step S400, based on the size specifications and installation angle parameters of the tension clamp, constructs a target detection and screening model containing multiple prediction heads, and uses an anchor frame generation strategy and a non-maximum suppression algorithm to perform defect target detection and localization on the analyzed image features, including the following sub-steps D1~D4:

[0093] In D1: Based on the size specifications and installation angle parameters of the tension clamp, a target detection screening model containing multiple prediction heads is constructed, with each prediction head responsible for detecting defective targets within a specific scale range;

[0094] Specifically, the output of the prediction head is calculated as follows:

[0095] ;

[0096] in, For the first The output of each prediction head; For activation functions; For the first The weight matrix of each prediction head is constructed based on the feature representation of tension clamp defects and the detection target; For input to the first The feature vector of each prediction head is generated by the features processed by the image feature extraction unit and the optimized deep Q network decision unit; For the first The bias vector of each prediction head;

[0097] Specifically, the prediction head outputs a weight matrix based on the feature representation of tension clamp defects and the detection target. The input is a feature vector processed by the preceding unit, and the prediction result is output through an activation function. During implementation, in the model training phase, the weight matrix and bias vector parameters of the prediction head are adjusted based on a large amount of tension clamp defect image data, enabling the prediction head to accurately identify tension clamp defect targets of different types and scales.

[0098] In D2: A preset anchor frame generation strategy is adopted. Based on the common position distribution, standard size and detection accuracy requirements of tension clamps in the image, anchor frames of different sizes and aspect ratios are generated as the initial target candidate areas. The center position, width and height parameters of the anchor frames are set according to the actual size characteristics of the tension clamps.

[0099] Specifically, the improved anchor frame generation formula is as follows:

[0100] ;

[0101] in, Indicates the first A generated anchor box, The coordinates of the anchor frame center are The width and height of the needle frame; The coordinates of the image center; The preset offset ratio parameter is determined based on the common positional distribution of tension clamps in the image; , These are the normalization factors for the image width and height, respectively; The basic anchor frame size is set by the standard size and testing accuracy requirements of the tension clamp; The scaling factor is determined based on the actual size fluctuation range of the tension clamp; the confidence update formula for the non-maximum suppression algorithm is: ,in, For the first Confidence level of each anchor box; For anchor frame and The crossover ratio is used to measure the degree of overlap between two anchor frames.

[0102] It should be noted that the anchor frame generation formula determines the center coordinates and base dimensions based on the common positions, standard sizes, and accuracy requirements of tension clamps in images, and adjusts the anchor frame shape through offset ratios and scaling factors. The confidence update formula of the non-maximum suppression algorithm retains target boxes with high confidence and no or low overlap by calculating the intersection-union ratio of anchor frames. In implementation, the model is trained based on a large amount of labeled tension clamp image data, and the anchor frame parameters and suppression algorithm thresholds are optimized to improve the accuracy of defect target detection and localization.

[0103] In D3: Image features after feature extraction and decision analysis are input into the target detection and screening model, which outputs the predicted bounding box coordinates, class confidence, and class probability of the defective target. The prediction results are processed using a non-maximum suppression algorithm to calculate the intersection-union ratio between each predicted bounding box. Based on preset overlap and confidence thresholds, predicted boxes with high confidence and low overlap are retained to suppress redundant and erroneous detection results. Based on the information of the finally retained predicted boxes, the precise location coordinates and size of the defective target in the image are determined, thus completing the detection and localization of the defective target.

[0104] It should be noted that the non-maximum suppression algorithm is used to process the initially predicted defect bounding boxes. First, the predicted boxes are sorted in descending order according to their confidence scores, and the highest-scoring boxes are selected as the retained results. The intersection-union ratio (IUU) of the highest-scoring boxes with the remaining boxes is calculated. A preset overlap threshold (usually set to 0.5 to 0.6) is used to determine whether they belong to the same target and to suppress redundant boxes. Then, a confidence threshold (usually set to 0.5) is used to further filter out low-confidence detection results, ultimately achieving accurate and non-redundant defect target localization.

[0105] In D4: The optimized data transmission for defect detection and localization is represented as follows:

[0106] ;

[0107] in, For data transmission time, The transmission efficiency adjustment coefficient is determined based on the system hardware performance and network environment; The amount of feature data is determined by the resolution of the tension clamp image and the dimension of feature extraction; For data transmission bandwidth; The number of transmission tasks is determined based on the real-time requirements of tension clamp defect detection and the system's processing capacity. The number of parallel transmission channels is set based on the system hardware configuration and the characteristics of the tension clamp image data. This indicates the transmission time of a single task.

[0108] It should be noted that, The bandwidth for data transmission is determined by the system's network environment and hardware performance. In a real system, the bandwidth cannot be zero (otherwise, there would be no basis for data transmission, and the system would not exist). The number of parallel transmission channels is set based on the system hardware configuration and the characteristics of the tension clamp image data. The number of parallel channels in the hardware configuration must be greater than 0 in the system design (otherwise, there would be no basis for parallel transmission). Therefore, and There is no case where the denominator is 0.

[0109] It should be noted that in the real-time transmission scenario of transmission line tension clamp defect analysis, each task has an inherent time cost. For example, the transmission task of a tension clamp image feature has a basic time consumption due to data interaction, hardware response, etc. At this time, This indicates the number of tasks each parallel channel needs to handle, and the time consumption for that part is the number of tasks multiplied by the base time cost of a single task. For example, if the transmission time of a single task is... , =10, =2, then each channel processes 5 tasks, and the time consumption is ,Right now .therefore, The association with the time dimension is indirectly bound through the task-time characteristics transmitted by the system, ultimately leading to... The formula creates an additive time dimension.

[0110] It should be noted that the data transmission time is quantitatively calculated by considering factors such as system hardware performance, network environment, image resolution, feature dimensions, number of transmission tasks, and number of parallel channels. During implementation, parameters such as the transmission efficiency adjustment coefficient are dynamically adjusted based on the actual hardware configuration and network conditions to optimize the data transmission process, ensuring fast and stable transmission of image feature data and guaranteeing system real-time performance.

[0111] It should be noted that step S400 above constructs a screening model containing multiple prediction heads based on parameters such as the size specifications and installation angle of the tension clamp. The design of multiple prediction heads allows for the detection of defect targets of different sizes and types. Each prediction head is responsible for predicting defect targets within a specific scale range, improving detection accuracy and efficiency. A special anchor frame generation strategy generates a series of anchor frames of different sizes and proportions based on the common location distribution of tension clamps in the image, standard dimensions, and detection accuracy requirements. These anchor frames serve as initial target candidate regions for subsequent target detection. The non-maximum suppression algorithm is used to filter the multiple overlapping target frames. By calculating the intersection-union ratio (IUU) between target frames, it retains target frames with high confidence and no overlap or low overlap, and removes redundant target frames, thereby achieving accurate detection and localization of defect targets in the tension clamp image. This enables the rapid and accurate determination of the position and size information of defect targets in the image, providing reliable data support for subsequent defect classification and evaluation.

[0112] In this embodiment of the invention, step S500 above constructs a classification and evaluation architecture based on a database of defects in tension clamps and a classification decision tree to determine the category and severity of detected defects, obtaining defect analysis results including the following sub-steps E1 and E2:

[0113] In E1: The specific steps for category determination include:

[0114] Establish a database containing characteristic information of various common defect types;

[0115] Extract the feature information of the detected defect targets, compare and analyze it with the standard features in the defect type feature database, and calculate the feature matching degree;

[0116] Based on feature matching degree, a classification decision tree is used to determine the category of the defect; starting from the root node of the classification decision tree, the features of the defect target are compared with the judgment conditions of the decision tree nodes, and the classification is carried out step by step downward to finally determine the specific category of the defect.

[0117] The node splitting criterion of the classification decision tree adopts an improved information gain ratio calculation method, which comprehensively considers the information gain of the attribute on the sample set and the dispersion of the attribute values.

[0118] In this embodiment of the invention, the formula for calculating the defect type matching degree is:

[0119] ;

[0120] in, Defect type matching degree; The number of features is determined based on the feature description of the tension clamp defect type; For the first The weights of each matching feature are set based on the importance of the tension clamp parameters to the identification of defect types. These are the detected defect feature values; These are standard defect characteristic values.

[0121] It should be noted that the number of matching features is determined based on the characteristic description of the tension clamp defect type, and each feature is assigned a weight. The degree of matching is measured by calculating the difference between the detected feature and the standard feature. During implementation, the defect type feature database is continuously updated, and the matching features and weights are optimized so that the system can accurately determine the degree of matching between the detected defect and the standard defect type, providing a reliable basis for defect classification.

[0122] In this embodiment of the invention, the classification decision tree constructed based on the feature database of tension clamp defect types uses an improved information gain ratio formula as the node splitting criterion:

[0123] ;

[0124] in, For sample set Information gain ratio; For sample set The information gain is calculated based on the characteristic distribution of the tension clamp defect samples. The inherent value, reflecting the action space The degree of dispersion of values ​​is determined by the diversity of parameters and the richness of defect types of tension clamps.

[0125] In E2: The specific steps for severity assessment include:

[0126] Based on the structural safety requirements and defect influencing factors of tension clamps, evaluation indicators are determined and weights are assigned to each evaluation indicator. The weight values ​​are allocated according to the importance of the tension clamp parameters to structural safety.

[0127] Calculate the score of each evaluation indicator based on the specific characteristics of the defective target;

[0128] The severity assessment value of the defect is calculated by weighted summation based on the scores of each assessment indicator.

[0129] In this embodiment of the invention, the severity assessment formula is as follows:

[0130] ;

[0131] in, This is the assessment value for the severity of the defect; The number of evaluation indicators is determined based on the structural safety requirements of tension clamps and the factors affecting defects; The weights of the evaluation indicators are set based on the importance of the tension clamp parameters to structural safety. The score for the evaluation index is determined by the specific characteristics of the tension clamp defects and relevant standards.

[0132] It should be noted that the node splitting criterion of the classification decision tree adopts an improved information gain ratio formula, which comprehensively considers the information gain of attributes on the sample set and the dispersion of attribute values, and is calculated based on the characteristic distribution of tension clamp defect samples. The severity assessment formula determines the assessment indicators based on structural safety requirements and defect influencing factors, assigns weights, and then calculates the severity of the defects. During implementation, the defect type feature database is improved, the decision tree is trained using historical data, and the assessment indicators and weights are continuously optimized to achieve accurate defect classification and severity assessment.

[0133] It should be noted that the S500 tension clamp defect type feature database described in the above steps stores feature information of various common defect types, including feature descriptions such as the shape, size, texture, and location of the defects. This information was obtained through the analysis and summarization of a large number of actual defect samples. The classification decision tree is based on this feature information and classifies the detected defect targets through a series of decision rules. Starting from the root node, it compares the characteristics of the defect target with the judgment conditions of the decision tree nodes, gradually dividing downwards to finally determine the category of the defect. In terms of severity assessment, an assessment index system based on the structural safety requirements of tension clamps and defect influencing factors is established. Each assessment index is assigned a corresponding weight, and the score of each index is calculated according to the specific characteristics of the defect target. Then, the severity assessment value of the defect is obtained by weighted summation, thereby providing maintenance personnel with accurate defect information so that appropriate maintenance measures can be taken.

[0134] In this embodiment of the invention, step S600, which outputs and stores the defect analysis results in a preset data format and storage strategy, includes:

[0135] Specifically, result output and storage are crucial interfaces for system-user interaction, responsible for displaying and storing the results obtained from the defect classification and evaluation unit. Regarding result output, information such as defect category and severity is organized and presented in a predefined data format. The data format design fully considers user habits and needs, facilitating maintenance personnel's intuitive understanding of the tension clamp's defect status. For example, basic defect information, including defect location, category, and severity, can be displayed in tabular form; alternatively, images can be used to mark the defect location on the original tension clamp image and display relevant detailed information. Regarding result storage, the detection results are stored in designated storage media according to a preset data storage strategy. The storage strategy considers data security, integrity, and retrieval, such as using redundant storage to ensure no data loss, establishing an indexing mechanism for easy user retrieval of historical detection results, and regularly backing up and cleaning the stored data to ensure efficient operation of the storage system.

[0136] It should be noted that step S600 above requires effective data integration with defect classification and assessment to ensure the accuracy of the received data. For result output, diverse output methods are provided based on different application scenarios and user needs, such as display on a local terminal or transmission to a remote monitoring center via network. Regarding data storage, appropriate storage devices and technologies are selected, such as hard disk arrays and cloud storage, and storage space is rationally planned and the storage structure optimized based on the size and growth trend of the data volume. Simultaneously, a strict data management mechanism is established to control access, modification, and deletion of data, ensuring data security and privacy.

[0137] Example 2: This example provides an image analysis-based system for analyzing defects in transmission line tension clamps, including:

[0138] The multi-dimensional image acquisition unit is used to acquire image information of the tension clamps of the transmission line using a preset spatial array layout and spectral response band combination, and obtain multi-dimensional image data.

[0139] The image feature extraction unit is used to construct a feature extraction architecture and extract features from multidimensional image data using a preset combination of convolutional kernel parameters and pooling strategies.

[0140] The deep network decision unit is used to build a reinforcement learning model based on the improved deep network structure, combined with the material properties and structural size parameters of the tension clamp, and to analyze and make decisions on the extracted image features through a preset neural network weight update mechanism and experience playback strategy.

[0141] The defect target screening unit is used to construct a target detection screening model containing multiple prediction heads based on the size specifications and installation angle parameters of the tension clamp. It uses an anchor frame generation strategy and a non-maximum suppression algorithm to detect and locate defect targets based on the analyzed image features.

[0142] The defect classification and evaluation unit is used to determine the category and severity of detected defects by constructing a classification and evaluation architecture based on a database of defects in tension clamps and a classification decision tree, and to obtain defect analysis results.

[0143] The results output and storage unit outputs and stores the defect analysis results in a preset data format and storage strategy.

[0144] It should be noted that the technical solution of the image analysis-based transmission line tension clamp defect analysis system is based on the same concept as the above-mentioned image analysis-based transmission line tension clamp defect analysis method. For details not described in detail in the technical solution of the image analysis-based transmission line tension clamp defect analysis system in this embodiment, please refer to the description of the above-mentioned image analysis-based transmission line tension clamp defect analysis method.

[0145] The aforementioned units can be embedded in the processor of the electronic device in hardware form or independent of it, or they can be stored in the memory of the electronic device in software form, so that the processor can call and execute the corresponding operations of the above units.

[0146] This embodiment also provides an electronic device, which includes a processor, a memory, a communication interface, a display screen, and an input device connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements an image analysis-based method for analyzing defects in transmission line tension clamps. The display screen can be a liquid crystal display (LCD) or an e-ink display. The input device can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the device's casing, or an external keyboard, touchpad, or mouse.

[0147] This embodiment also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method proposed in the above embodiments.

[0148] The storage medium proposed in this embodiment belongs to the same inventive concept as the method proposed in the above embodiments. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0149] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory, random access memory, flash memory, hard disk, or optical disk, and includes several instructions to cause an electronic device (which may be a personal computer, server, or network device, etc.) to execute the method of the embodiments of the present invention.

[0150] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for analyzing defects in transmission line tension clamps based on image analysis, characterized in that, include: Multi-view, multi-spectral imaging equipment is used with a preset spatial array layout and spectral response band combination to collect image information of the tension clamps of the transmission line and obtain multi-dimensional image data. A feature extraction architecture is constructed, and features are extracted from the multidimensional image data using convolution kernel parameter combinations and pooling strategies; Based on the improved deep network structure, combined with the material properties and structural size parameters of the tension clamp, a reinforcement learning model is constructed, and the extracted image features are analyzed and decided through the neural network weight update mechanism and experience replay strategy. Based on the size specifications and installation angle parameters of the tension clamp, a target detection and screening model containing multiple prediction heads is constructed. An anchor frame generation strategy and non-maximum suppression algorithm are used to detect and locate defect targets in the analyzed image features. By constructing a classification and evaluation architecture based on a database of defects in tension clamps and a classification decision tree, the detected defect targets are classified and their severity is assessed to obtain defect analysis results. The defect analysis results are output and stored in a preset data format and storage strategy; The construction of the reinforcement learning model includes: Based on the improved deep network structure, and combined with the material properties and structural size parameters of the tension clamp, the state space, action space and reward function are defined. The constructed state space S is: ,in, Indicates the first The first feature dimension Each quantified level status, The number of feature dimensions for the tension clamp image is determined based on the structural complexity of the tension clamp and the image acquisition resolution. The number of quantization levels for characteristic values ​​is determined by the accuracy requirements of the tension clamp parameters and the data processing capability. The constructed action space A is based on the defect detection requirements of tension clamps and the system processing capabilities, and includes a series of decision actions for image feature analysis; Constructed reward function for: ,in, The number of reward factors is set according to the type of defect in the tension clamp and the detection target; The weight of the reward factor is determined based on the importance of the tension clamp parameters to defect detection; In the state Next action The reward value corresponding to the reward factor at that time; The improved deep network structure is trained using historical tension clamp image data and defect annotation data. During the training process, the network is continuously allowed to interact with the environment, that is, the extracted image features are input, actions are executed, and reward feedback is obtained. The weight parameters of the neural network are updated using gradient descent based on the reward function. The data transmission optimization during defect target detection and localization is expressed as follows: ； in, For data transmission time, The transmission efficiency adjustment coefficient is determined based on the system hardware performance and network environment; The amount of feature data is determined by the resolution of the tension clamp image and the dimension of feature extraction; For data transmission bandwidth; The number of transmission tasks is determined based on the real-time requirements of tension clamp defect detection and the system's processing capacity. The number of parallel transmission channels is set based on the system hardware configuration and the characteristics of the tension clamp image data. This indicates the transmission time of a single task.

2. The image analysis-based method for analyzing defects in transmission line tension clamps as described in claim 1, characterized in that, The feature extraction of the multidimensional image data includes: Based on the resolution and data volume of the multidimensional image data, determine the number of layers in the neural network and the number of nodes in each layer; The convolutional layer of the feature extraction architecture first initializes the convolutional kernel parameters and trains them using historical tension clamp image data. It continuously adjusts the weights and biases of the convolutional kernel parameters to extract different levels of features from the multidimensional image data. The pooling layer of the feature extraction architecture selects the pooling type and pooling window size according to the scale of the different levels of features and the limitations of computing resources, and performs further feature extraction on the multidimensional image data to obtain the edge features, shape features, texture features and spectral features of the tension clamp image at different scales and dimensions.

3. The image analysis-based method for analyzing defects in transmission line tension clamps as described in claim 1, characterized in that, The defect target detection and localization includes: Based on the size specifications and installation angle parameters of the tension clamp, a target detection and screening model containing multiple prediction heads is constructed, with each prediction head responsible for detecting defective targets within a specific scale range; Using a preset anchor frame generation strategy, anchor frames of different sizes and aspect ratios are generated as initial target candidate regions based on the common position distribution, standard size, and detection accuracy requirements of tension clamps in the image; the center position, width, and height parameters of the anchor frames are set according to the actual size characteristics of the tension clamps. The image features after feature extraction and decision analysis are input into the target detection and screening model, and the predicted bounding box coordinates, category confidence and category probability of the defective target are output. The prediction results are processed using a non-maximum suppression algorithm. The cross-union ratio between each predicted bounding box is calculated. Based on the preset overlap threshold and confidence threshold, the predicted boxes with high confidence and low overlap are retained to suppress redundant and erroneous detection results. Based on the final preserved prediction bounding box information, the precise location coordinates and size of the defective target in the image are determined, thus completing the detection and localization of the defective target.

4. The image analysis-based method for analyzing defects in transmission line tension clamps as described in claim 1, characterized in that, The defect analysis results obtained include: Establish a database containing characteristic information of various common defect types; Extract the feature information of the detected defect targets, compare and analyze it with the standard features in the defect type feature database, and calculate the feature matching degree; Based on the feature matching degree, the category of the defect is determined using a classification decision tree; starting from the root node of the classification decision tree, the features of the defect target are compared with the judgment conditions of the decision tree nodes, and the tree is divided downwards step by step to finally determine the specific category of the defect. The node splitting criterion of the classification decision tree adopts an improved information gain ratio calculation method, which comprehensively considers the information gain of the attribute on the sample set and the dispersion of the attribute values.

5. The image analysis-based method for analyzing defects in transmission line tension clamps as described in claim 4, characterized in that, The severity assessment includes: Based on the structural safety requirements and defect influencing factors of tension clamps, evaluation indicators are determined and weights are assigned to each evaluation indicator. The weight values ​​are allocated according to the importance of the tension clamp parameters to structural safety. Calculate the score of each evaluation indicator based on the specific characteristics of the defective target; The severity assessment value of the defect is calculated by weighted summation based on the scores of each assessment indicator.

6. A transmission line tension clamp defect analysis system based on image analysis, employing the transmission line tension clamp defect analysis method based on image analysis as described in any one of claims 1 to 5, characterized in that, include: The multi-dimensional image acquisition unit is used to acquire image information of the tension clamps of the transmission line using a preset spatial array layout and spectral response band combination, and obtain multi-dimensional image data. An image feature extraction unit is used to construct a feature extraction architecture and extract features from the multidimensional image data using convolution kernel parameter combinations and pooling strategies. The deep network decision unit is used to build a reinforcement learning model based on the improved deep network structure, combined with the material properties and structural size parameters of the tension clamp, and to analyze and make decisions on the extracted image features through the neural network weight update mechanism and experience replay strategy. The defect target screening unit is used to construct a target detection screening model containing multiple prediction heads based on the size specifications and installation angle parameters of the tension clamp. It uses an anchor frame generation strategy and a non-maximum suppression algorithm to detect and locate defect targets based on the analyzed image features. The defect classification and evaluation unit is used to determine the category and severity of detected defects by constructing a classification and evaluation architecture based on a database of defects in tension clamps and a classification decision tree, and to obtain defect analysis results. The results are output and stored in the storage unit, where the defect analysis results are output and stored in a preset data format and storage strategy.

7. An electronic device comprising a memory and a processor, characterized in that: The memory is used to store computer-executable instructions, and when the processor executes the computer-executable instructions, it implements the steps of the image analysis-based transmission line tension clamp defect analysis method according to any one of claims 1 to 5.

8. A computer-readable storage medium storing computer-executable instructions thereon, characterized in that: When the computer-executable instructions are executed by the processor, they implement the steps of the image analysis-based transmission line tension clamp defect analysis method according to any one of claims 1 to 5.

Images