Power transmission line tower assembly part detection method, terminal and medium
Through image processing technology and the improved Faster R-CNN neural network, the automated detection and identification of transmission line tower components is realized, solving the problems of low efficiency, high safety risks and difficult to guarantee in traditional manual operation, and significantly improving construction efficiency and safety.
Patent Information
- Application Number
- CN202411770948.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-04
- Publication Date
- 2025-03-28
AI Technical Summary
Traditional manual operation has problems such as low efficiency, high safety risks and difficult to guarantee accuracy during the assembly of transmission towers, especially in harsh environments.
Image acquisition, image processing, feature extraction and object detection technology are used to construct image data sets of transmission line tower components and automatically detect and identify them using the improved Faster R-CNN neural network.
It realizes high-precision detection and identification of transmission line tower components, improves construction efficiency and safety, adapts to complex environments, and significantly improves technological progress and benefits in the construction field of the power industry.
Smart Images
Figure CN119723038A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of power systems, and in particular to a method, terminal and medium for detecting components of a transmission line tower assembly. Background Art
[0002] For a long time, transmission tower assembly projects usually require a lot of manpower, and heavy manual operations are required from carrying materials to assembly. This traditional manual operation method has some problems in terms of efficiency and safety. First, the high consumption and fatigue of human resources may have an impact on the construction progress and quality. In addition, the assembly of transmission towers is a highly dangerous operation, and manual operators may need to work at heights and face safety risks such as falling and falling objects. In addition, the assembly of transmission towers also requires a high degree of accuracy and stability to ensure the structure and stability of the tower. However, manual operation may be affected by factors such as skill level and fatigue, which may affect the accuracy and stability of assembly. In addition, the construction of transmission towers usually needs to be carried out under different environmental conditions, such as bad weather, high altitude areas, etc., which also increases the difficulty and risk of manual operation. Summary of the invention
[0003] The present application provides a method, terminal and medium for detecting components of a transmission line tower assembly, which realizes automatic detection and identification of components of a transmission line tower assembly through steps such as image acquisition, image processing, feature extraction and target detection, bringing significant technological progress and benefit improvement to the construction field of the power industry.
[0004] The technical solution of this application is as follows:
[0005] On the one hand, the present application provides a method for detecting components of a transmission line tower assembly, comprising the following steps:
[0006] Step 1: Construct an image dataset of transmission line tower components;
[0007] Step 2: Label the image data and divide the dataset into training set, validation set and test set;
[0008] Step 3: Build a fast-changing Faster R-CNN neural network and train the model; in the Faster R-CNN neural network structure, add the FPN feature pyramid structure;
[0009] Step 4: Post-process the feedback data results using the NMS model application;
[0010] Step 5: Use the Faster R-CNN neural network to identify, locate and detect tower assembly components.
[0011] Furthermore, step 1 includes:
[0012] Collect image data of various components of the transmission line tower, including towers, lines, lifting equipment and surrounding environment;
[0013] Perform image enhancement and size normalization operations on image data.
[0014] Furthermore, step 2 includes:
[0015] Annotate the collected image data, that is, draw bounding boxes for each component of the tower in each image or perform pixel-level segmentation annotations so that the bounding box or segmentation annotation of each component covers the area of the target component;
[0016] The data set is divided into training set, validation set and test set. The data of the training set is more than that of the validation set and the test set. When dividing the data set, random segmentation is adopted, and the data in the training set, validation set and test set all contain data from different targets.
[0017] Furthermore, the Faster R-CNN neural network includes:
[0018] Backbone network, which includes several convolutional layers and pooling layers, is used to extract features from the entire input image. The feature map is shared in the subsequent region proposal network and interest region pooling layer;
[0019] Region Proposal Network: The region proposal network performs traversal convolution on the feature map output by the backbone network in a sliding window manner. The result of the traversal convolution of the region proposal network is sent to its two fully connected layers, namely the classification fully connected layer and the positioning fully connected layer, which complete the classification and positioning tasks respectively. The classification fully connected layer performs binary classification to determine whether the region is a positive sample or a negative sample. The positioning fully connected layer combines the positive sample region proposal to generate an offset for the frame regression correction, obtain the region proposal, and remove the proposal boxes that are too small or exceed the boundary;
[0020] The region of interest pooling layer combines the feature map output by the original image with the region proposal information generated by the region proposal network, performs region of interest pooling on the feature maps of these region proposals, and obtains a fixed-size feature vector that is sent to the fully connected layer;
[0021] The classification and regression layer uses the Softmax classifier to calculate the category probability based on the region proposal features to complete the classification, and performs bounding box regression again to obtain the final position coordinates of the detection box.
[0022] Furthermore, the region proposal network is a convolutional neural network, and the loss function of the region proposal network is calculated and trained. The loss function of the region proposal network includes the loss value of the classification layer and the loss value of the corrected regression layer; the loss value of the region proposal network is:
[0023] R=R c +R m
[0024] R c is the loss value of the classification layer; R m is the loss value of the corrected regression layer; R is the loss value of RPN; R c With R m The calculations are:
[0025]
[0026] j is the target suggestion box number; o j is the probability that the jth target box is a positive sample; o j ′ is the actual probability of the jth target box, which is 1 for positive samples and 0 for negative samples; N c is the number of samples in the classification layer;
[0027]
[0028] D j is the deviation between the predicted coordinates and the actual coordinates; t j is the predicted coordinate set of the jth target box; t j ′ is the actual coordinate set of the jth target box; M R is the loss function.
[0029] Furthermore, during model training, the weight matrix w and the bias b are the objects that need to be updated for network training; error back propagation is used to train the network model and update the parameter matrix. When the error E between the output y of the network feedforward calculation and the expected value r does not meet the threshold condition, the model will back propagate the error along the original path, calculate layer by layer, and update the weight matrix w and the bias b according to the set learning rate until it reaches the input layer to achieve an error back propagation; feedforward calculation and back propagation are repeated alternately until the network output y is close enough to the expected value r, and the error E between the two meets the threshold condition, and the parameter update is completed.
[0030] Furthermore, in step 4, all bounding boxes of the current category are traversed and non-maximum suppression is performed. The specific steps are as follows:
[0031] Step 41: Take the bounding box with the highest score and regard it as the currently retained target;
[0032] Step 42: For the remaining bounding boxes, calculate their degree of overlap with the current retained target; wherein, the degree of overlap between the bounding box and the current retained target is calculated using the intersection over union (IoU), which is used as an indicator for measuring and marking the candidate boxes, and the calculation formula is:
[0033] IoU = (A∩B) / (A∪B)
[0034] The candidate boxes are filtered and marked using the following rules:
[0035] (1) Remove candidate areas that exceed the boundaries of the original image;
[0036] (2) If the IoU value between the candidate box and the real area is the largest, it is marked as a positive sample, label = 1;
[0037] (3) If the IoU between the candidate box and the real area is greater than 0.7, it is marked as a positive sample, label = 1;
[0038] (4) If the IoU between the candidate box and the real area is less than 0.3, it is marked as a negative sample, label = 0;
[0039] (5) The rest are neither positive samples nor negative samples and are not used in the final result, label = -1.
[0040] Step 43: If the degree of overlap between a certain bounding box and the currently retained target is higher than a set threshold, the bounding box is removed, otherwise it is retained;
[0041] Step 44: Repeat the above steps until all bounding boxes are processed.
[0042] Furthermore, in step 5, the trained network is configured to the terminal, the image of the power transmission line tower to be detected is detected in the terminal, the positions of different components are found and located in the image, and the detected components are identified through the classification branch of the model to determine their specific categories.
[0043] On the other hand, the present application provides a transmission line tower component detection terminal, including a memory and a processor, the memory storing a computer-readable program, and when the computer-readable program is called and executed by the processor, the transmission line tower component detection method as described above is implemented.
[0044] On the other hand, the present application provides a computer-readable medium, in which a computer-readable program is stored. When the computer-readable program is called and executed by a computer, the transmission line tower component detection method as described above is implemented.
[0045] In summary, the beneficial effects of this application are:
[0046] 1. The present invention applies machine vision technology to the intelligent hoisting system of power transmission tower construction, which is conducive to achieving high-precision positioning, intelligent hoisting control and adaptability to complex environments, etc., bringing significant technological progress and benefit improvement to the construction field of the power industry;
[0047] 2. This paper proposes an improved Faster R-CNN network architecture, which fully combines high-level semantic information and low-level pixel information, and adds an FPN feature pyramid structure to the original network structure. Using this neural network, different components can be detected and located in the image through image filtering and fast transformation techniques. BRIEF DESCRIPTION OF THE DRAWINGS
[0048] Figure 1 It is a flow chart of a method for detecting components of a transmission line tower assembly according to the present invention;
[0049] Figure 2 This is a diagram of the Faster R-CNN network architecture in an embodiment of the present invention;
[0050] Figure 3 It is a diagram of the FPN feature pyramid structure in an embodiment of the present invention. DETAILED DESCRIPTION
[0051] The specific implementation of the present application is described in detail below with reference to the accompanying drawings.
[0052] A specific embodiment of the present application provides a method for detecting components of a transmission line tower assembly, such as Figure 1 As shown, the following steps are included:
[0053] Step 1: Construct an image dataset of transmission line tower components. Step 1 includes:
[0054] Collect image data of various components of the transmission line tower, including towers, lines, hoisting equipment, and surrounding environment; the data set covers images of different angles and distances under various lighting conditions to improve the robustness and performance of the algorithm. Preprocess the collected image data including image denoising, using Gaussian filters or median filters to remove noise interference.
[0055] Perform image enhancement, contrast enhancement, histogram equalization and size normalization on image data. Image enhancement, contrast enhancement and histogram equalization can improve image quality; size normalization adjusts all input images to the same size for subsequent processing.
[0056] Step 2: Label the image data and divide the dataset into training set, validation set and test set. Step 2 includes:
[0057] Annotate the collected image data, that is, draw bounding boxes for each component of the tower in each image or perform pixel-level segmentation annotations so that the bounding box or segmentation annotation of each component covers the area of the target component;
[0058] The dataset is divided into training set, validation set and test set. The data in the training set is more than that in the validation set and test set. Random segmentation is adopted when dividing the dataset. The data in the training set, validation set and test set all contain data from different targets. This ensures the randomness and representativeness of the dataset segmentation and avoids the images of the same group of tower components appearing in the training, validation and test sets. This constructs a dataset of images of various components of transmission line towers that are diverse and accurately labeled.
[0059] Step 3: Build a fast-changing Faster R-CNN neural network and train the model; in the Faster R-CNN neural network structure, add the FPN feature pyramid structure.
[0060] like Figure 2 As shown, the Faster R-CNN neural network includes:
[0061] Backbone network,The backbone network includes several convolutional layers and pooling layers, which are used to extract features from the entire input original image. The feature map is shared in the subsequent region proposal network and interest region pooling layer.
[0062] Region Proposal Network: The region proposal network performs traversal convolution on the feature map output by the backbone network in a sliding window manner, and generates an anchor point corresponding to the center point of each sliding window on the original image. Then, with the anchor point as the center, a predefined k anchor point box is generated on the original image, so that multi-scale and multi-ratio target detection can be achieved in a convolution feature map. The result of the traversal convolution of the region proposal network will be sent to the two fully connected layers inside it. The two fully connected layers are the classification fully connected layer and the positioning fully connected layer, which complete the classification and positioning tasks respectively. The classification fully connected layer performs binary classification to determine whether the region is a positive sample or a negative sample. The positioning fully connected layer combines the positive sample region proposal to generate an offset for the frame regression correction, obtain the region proposal, and remove the proposal box that is too small or exceeds the boundary.
[0063] The region of interest pooling layer combines the feature map output by the original image with the region proposal information generated by the region proposal network, performs region of interest pooling on the feature maps of these region proposals, and obtains a feature vector of a fixed size and sends it to the fully connected layer.
[0064] The classification and regression layer uses the Softmax classifier to calculate the category probability based on the region proposal features to complete the classification, and performs bounding box regression again to obtain the final position coordinates of the detection box.
[0065] The region proposal network is a convolutional neural network that generates a sliding vector block on the feature map. The vector block moves continuously with 3×3 pixels and generates a one-dimensional vector, and generates the target regression parameters through the fully connected layer. The 2k classification scores are mainly the foreground classification scores and background classification scores of the k anchor boxes. The 4k coordinate values are the coordinates of the target box regression values generated by the k anchor boxes. The feature map can also be viewed as multiple vectors, each of which is 512 dimensions. In order to make the results of RPN more reliable, the loss function of the region proposal network is calculated and trained. The loss function of the region proposal network includes the loss value of the classification layer and the loss value of the corrected regression layer; the loss value of the region proposal network is:
[0066] R=R c +R m
[0067] R c is the loss value of the classification layer; R m is the loss value of the corrected regression layer; R is the loss value of RPN; R c With R m The calculations are:
[0068]
[0069] j is the target suggestion box number; o j is the probability that the jth target box is a positive sample; o j ′ is the actual probability of the jth target box, which is 1 for positive samples and 0 for negative samples; N c is the number of samples in the classification layer;
[0070]
[0071] D j is the deviation between the predicted coordinates and the actual coordinates; t j is the predicted coordinate set of the jth target box; t j ′ is the actual coordinate set of the jth target box; M R is the loss function.
[0072] In the classification layer branch, the candidate area is classified into foreground / background through the fully connected layer and the softmax function. The softmax function is also called the normalized exponential function, which can compress the k-dimensional real vector into a new k-dimensional real vector ranging from 0 to 1. Therefore, the softmax function can be used in multi-classification problems to normalize the output value into a probability value. The function form is:
[0073]
[0074] Among them, z i is the i-th component of the input; K is the total number of categories.
[0075] In neural networks, the softmax function can be used as the activation function of the output layer, where is the input of the jth neuron in the output layer. The output result of the softmax function represents the probability. The probability has two properties: the predicted probability is a non-negative number; the sum of the probabilities of various prediction results is equal to 1.
[0076] The purpose of neural network training is to obtain a series of model parameters so that the output of the network fits the expected value as much as possible. During model training, the weight matrix w and the bias b are the objects that need to be updated for network training; error back propagation is used to train the network model and update the parameter matrix. When the error E between the output y of the network feedforward calculation and the expected value r does not meet the threshold condition, the model will back propagate the error along the original path, calculate layer by layer, and update the weight matrix w and the bias b according to the set learning rate until it reaches the input layer to achieve an error back propagation; feedforward calculation and back propagation are repeated alternately until the network output y is close enough to the expected value r, and the error E between the two meets the threshold condition, and the parameter update ends.
[0077] Taking the update process of the weight parameter w of a single neuron as an example, the error function is defined as The activation function is the Sigmoid function After the neural network is trained, the model parameters are optimized through back-propagation so that it can accurately detect and identify components. The training process ends and the model parameters are fixed.
[0078] Faster R-CNN achieves accurate positioning and classification of target objects in images by converting the target detection problem into two stages: candidate region generation and classification. The original image is first extracted through a convolutional neural network to obtain a feature map; the feature map enters the region proposal network RPN to generate a large number of candidate boxes. One branch filters the candidate boxes and uses the softmax function to determine whether the candidate box belongs to the foreground or the background, that is, whether it is a target or background, which is a binary classification problem; at the same time, the other branch uses the loss function smooth to perform border regression to correct the candidate box, forming a more accurate proposed candidate area.
[0079] The pooling layer ROI combines the proposed candidate area generated by RPN and the feature map obtained by the last layer of the CNN feature extraction network to obtain a fixed-size proposed feature block; the proposed feature block generated by the last ROI layer enters the fully connected layer network, and after passing through the bounding box regression layer and the classification layer, the precise position of the detection box and the specific category of the object in the box are output.
[0080] In this example, ResNeSt is selected as the backbone network. Compared with VGG16, ResNeSt as the backbone network of Faster R-CNN is more inclined to extract effective features, which can improve the network's ability to detect targets. The core of ResNet is its single block ResNeSt Block, which mainly includes residual structure, grouping structure and Split-Attention. The overall process of ResNet Block: First, the input features are grouped according to the preset hyperparameter K; then the features are split in each group according to the hyperparameter R, and the split feature attention vector is generated and then the split features are weighted summed; finally, the features output by each group are spliced, and after passing through a 3×3 convolution layer and adding the residual information, a more targeted feature matrix is output.
[0081] Directly using the single-scale features generated by the backbone network makes it difficult to adapt the model to the detection of objects of different sizes. Figure 3 As shown in the figure, the introduction of FPN can improve the model's detection ability for different components by combining contextual features and generating multi-scale features. FPN not only brings global information to shallow local features, but also generates 5 sets of features of different scales, so that the subsequent RPN and RoIPooling can better generate interesting features. Therefore, the introduction of FPN can reduce the difficulty of model fitting and improve the robustness of the model in detecting targets of different scales.
[0082] Step 4: Post-process the feedback data results using the NMS model application;
[0083] NMS stands for non-maximum suppression. It starts to traverse all bounding boxes of the current category and perform non-maximum suppression processing. The specific steps are as follows:
[0084] Step 41: Take the bounding box with the highest score and regard it as the currently retained target;
[0085] Step 42: For the remaining bounding boxes, calculate their degree of overlap with the current retained target; wherein, the degree of overlap between the bounding box and the current retained target is calculated using the intersection over union (IoU), which is used as an indicator for measuring and marking the candidate boxes, and the calculation formula is:
[0086] IoU = (A∩b) / (A∪B)
[0087] The candidate boxes are filtered and marked using the following rules:
[0088] (1) Remove candidate areas that exceed the boundaries of the original image;
[0089] (2) If the IoU value between the candidate box and the real area is the largest, it is marked as a positive sample, label = 1;
[0090] (3) If the IoU between the candidate box and the real area is greater than 0.7, it is marked as a positive sample, label = 1;
[0091] (4) If the IoU between the candidate box and the real area is less than 0.3, it is marked as a negative sample, label = 0;
[0092] (5) The rest are neither positive samples nor negative samples and are not used in the final result, label = -1.
[0093] Step 43: If the degree of overlap between a certain bounding box and the currently retained target is higher than the set threshold (0.5 in this example), the bounding box is removed, otherwise it is retained;
[0094] Step 44: Repeat the above steps until all bounding boxes are processed.
[0095] After completing the post-processing, for each category, after non-maximum suppression processing, the target result after filtering out redundant bounding boxes is obtained. It can be visualized, saved or further processed as needed.
[0096] Step 5: Use the Faster R-CNN neural network to identify, locate and detect tower assembly components.
[0097] The trained network is configured to the terminal, and the image of the transmission line tower to be detected is detected in the terminal. The positions of different components are found and located in the image. For the detected components, they are identified through the classification branch of the model to determine their specific categories, such as towers, lines, etc.
[0098] Another specific embodiment of the present application provides a transmission line tower component detection terminal, including a memory and a processor, wherein the memory stores a computer-readable program, and when the computer-readable program is called and executed by the processor, the transmission line tower component detection method as described above is implemented.
[0099] Another specific embodiment of the present application provides a computer-readable medium, in which a computer-readable program is stored. When the computer-readable program is called and executed by a computer, the above-mentioned method for detecting components of a transmission line tower assembly is implemented.
[0100] The parts not involved in the present invention are the same as the prior art or are implemented by using the prior art.
[0101] The above is only a preferred implementation of the present application. It should be pointed out that a person skilled in the art can make several modifications and improvements without departing from the inventive concept of the present application, and these all fall within the scope of protection of the present application.
Claims
1. A method for detecting components of a transmission line tower assembly, characterized in that: The following steps are involved: Step 1: Construct an image dataset of transmission line tower components; Step 2: Label the image data and divide the dataset into training set, validation set and test set; Step 3: Build a fast-changing Faster R-CNN neural network and train the model; in the Faster R-CNN neural network structure, add the FPN feature pyramid structure; Step 4: Post-process the feedback data results using the NMS model application; Step 5: Use the Faster R-CNN neural network to identify, locate and detect tower assembly components.
2. The method for detecting transmission line tower components according to claim 1, characterized in that: Step 1 includes: Collect image data of various components of the transmission line tower, including towers, lines, lifting equipment and surrounding environment; Perform image enhancement and size normalization operations on image data.
3. The method for detecting transmission line tower components according to claim 1, characterized in that: Step 2 includes: Annotate the collected image data, that is, draw bounding boxes for each component of the tower in each image or perform pixel-level segmentation annotations so that the bounding box or segmentation annotation of each component covers the area of the target component; The data set is divided into training set, validation set and test set. The data of the training set is more than that of the validation set and the test set. When dividing the data set, random segmentation is adopted, and the data in the training set, validation set and test set all contain data from different targets.
4. The method for detecting transmission line tower components according to claim 1, characterized in that: The Faster R-CNN neural network includes: Backbone network, which includes several convolutional layers and pooling layers, is used to extract features from the entire input image. The feature map is shared in the subsequent region proposal network and interest region pooling layer; Region Proposal Network: The region proposal network performs traversal convolution on the feature map output by the backbone network in a sliding window manner. The result of the traversal convolution of the region proposal network is sent to its two fully connected layers, namely the classification fully connected layer and the positioning fully connected layer, which complete the classification and positioning tasks respectively. The classification fully connected layer performs binary classification to determine whether the region is a positive sample or a negative sample. The positioning fully connected layer combines the positive sample region proposal to generate an offset for the frame regression correction, obtain the region proposal, and remove the proposal boxes that are too small or exceed the boundary; The region of interest pooling layer combines the feature map output by the original image with the region proposal information generated by the region proposal network, performs region of interest pooling on the feature maps of these region proposals, and obtains a fixed-size feature vector that is sent to the fully connected layer; The classification and regression layer uses the Softmax classifier to calculate the category probability based on the region proposal features to complete the classification, and performs bounding box regression again to obtain the final position coordinates of the detection box.
5. The method for detecting transmission line tower components according to claim 1, characterized in that: The region proposal network is a convolutional neural network. The loss function of the region proposal network is calculated and trained. The loss function of the region proposal network includes the loss value of the classification layer and the loss value of the corrected regression layer. The loss value of the region proposal network is: R=R c +R m R c is the loss value of the classification layer; R m is the loss value of the corrected regression layer; R is the loss value of RPN; R c With R m The calculations are: j is the target suggestion box number; o j is the probability that the jth target box is a positive sample; o j ′ is the actual probability of the jth target box, which is 1 for positive samples and 0 for negative samples; N c is the number of samples in the classification layer; D j is the deviation between the predicted coordinates and the actual coordinates; t j is the predicted coordinate set of the jth target box; t j ′ is the actual coordinate set of the jth target frame; M R is the loss function.
6. The method for detecting transmission line tower components according to claim 1, characterized in that: During model training, the weight matrix w and bias b are the objects that need to be updated for network training; error back propagation is used to train the network model and update the parameter matrix. When the error E between the output y of the network feedforward calculation and the expected value r does not meet the threshold condition, the model will back propagate the error along the original path, calculate layer by layer, and update the weight matrix w and bias b according to the set learning rate until it reaches the input layer to achieve an error back propagation; feedforward calculation and back propagation are repeated alternately until the network output y is close enough to the expected value r, and the error E between the two meets the threshold condition, and the parameter update is completed.
7. The method for detecting transmission line tower components according to claim 1, characterized in that: In step 4, all bounding boxes of the current category are traversed and non-maximum suppression is performed. The specific steps are as follows: Step 41: Take the bounding box with the highest score and regard it as the currently retained target; Step 42: For the remaining bounding boxes, calculate their degree of overlap with the current retained target; wherein, the degree of overlap between the bounding box and the current retained target is calculated using the intersection over union (IoU), which is used as an indicator for measuring and marking the candidate boxes, and the calculation formula is: IoU = (A∩B) / (A∪B) The candidate boxes are filtered and marked using the following rules: (1) Remove candidate areas that exceed the boundaries of the original image; (2) If the IoU value between the candidate box and the real area is the largest, it is marked as a positive sample, label = 1; (3) If the IoU between the candidate box and the real area is greater than 0.7, it is marked as a positive sample, label = 1; (4) If the IoU between the candidate box and the real area is less than 0.3, it is marked as a negative sample, label = 0; (5) The rest are neither positive samples nor negative samples and are not used in the final result, label = -1. Step 43: If the degree of overlap between a certain bounding box and the currently retained target is higher than a set threshold, the bounding box is removed, otherwise it is retained; Step 44: Repeat the above steps until all bounding boxes are processed.
8. The method for detecting components of a transmission line tower assembly according to claim 1, characterized in that: In step 5, the trained network is configured to the terminal, the image of the power transmission line tower to be detected is detected in the terminal, the positions of different components are found and located in the image, and the detected components are identified through the classification branch of the model to determine their specific categories.
9. A transmission line tower component detection terminal, characterized in that: It comprises a memory and a processor, wherein the memory stores a computer-readable program, and when the computer-readable program is called and executed by the processor, the method for detecting tower components of a transmission line assembly as claimed in any one of claims 1 to 8 is implemented.
10. A computer-readable medium, characterized in that The computer-readable medium stores a computer-readable program, and when the computer-readable program is called and executed by a computer, the method for detecting tower components of a transmission line assembly as described in any one of claims 1 to 8 is implemented.