Intelligent positioning method and system for central axis of core rod in infrared image of composite insulator

Abstract

The invention discloses a composite insulator infrared image core rod central axis intelligent positioning method and system, and the method comprises the steps: obtaining a composite insulator infrared image data set, and carrying out the marking of key points on the central axis of an anchor frame and a core rod of a composite insulator, so as to obtain the marking data; establishing a Yolo deep learning model, and marking key points of the core rod position in the infrared image by combining a key point detection technology to obtain an initial detection model; training the initial detection model based on the annotation data to obtain an optimized detection model; and carrying out core rod central axis key point positioning on a to-be-detected composite insulator infrared image based on the optimized detection model. According to the method, deep learning and key point detection technologies are combined, meanwhile, a new infrared image of the composite insulator can be rapidly processed and recognized, real-time monitoring and analysis are achieved, manual intervention can be greatly reduced, the detection efficiency can be improved, maintenance or repair measures can be taken in time, and the method is suitable for industrial popularization.

Description

Technical Field

[0001] The present invention relates to the technical field of insulator state recognition, and more particularly to a composite insulator infrared image core rod central axis intelligent positioning method and system. Background Art

[0002] Composite insulators play a vital role in power systems. They support and isolate high-voltage power lines, ensuring the safety and reliability of power transmission. As the scale of power grids expands and transmission capacity increases, the stability of composite insulators directly affects the operational efficiency and safety of power systems. The temperature curve along the centerline of the composite insulator core rod is a key location that requires attention during power inspections. Abnormal heating is a major factor leading to power system failures. This heating can be caused by factors such as material aging, mechanical stress, environmental corrosion, or excessive electrical load. Therefore, timely and accurate monitoring and diagnosis of the thermal state of the centerline of the composite insulator core rod is of great significance for preventing power accidents and improving power grid reliability.

[0003] In recent years, aerial inspections using drones equipped with infrared thermal imagers have become an effective technology. They can quickly capture infrared images of composite insulators over a large area, monitoring the thermal state of the insulator core axis to identify potential overheating issues. However, this method still faces some challenges. While drone inspections can collect large amounts of infrared image data in a short period of time, the processing and analysis of this data is complex and time-consuming. Existing inspection methods also rely heavily on visual analysis and interpretation of infrared images by professionals, a process that is time-consuming and susceptible to the influence of personal experience and judgment, potentially resulting in analytical results that are inconsistent with actual conditions.

[0004] In summary, current monitoring methods lack highly automated and intelligent processing capabilities, making it difficult to effectively process and analyze massive amounts of data and achieve early warning and precise fault location. Therefore, it is necessary to conduct research on intelligent positioning of the central axis of composite insulators using infrared images. Summary of the Invention

[0005] The present invention proposes a composite insulator core rod central axis intelligent positioning method and system using infrared images to solve the problem of how to efficiently position the core rod central axis of the composite insulator.

[0006] In order to solve the above problems, according to one aspect of the present invention, a method for intelligently locating the central axis of a composite insulator core rod using infrared imaging is provided, the method comprising:

[0007] Obtain a composite insulator infrared image dataset and annotate key points on the composite insulator's anchor frame and core rod centerline to obtain annotated data.

[0008] A Yolo deep learning model was established and combined with key point detection technology to mark the key points of the core rod position in the infrared image to obtain an initial detection model;

[0009] Training the initial detection model based on the labeled data to obtain an optimized detection model;

[0010] Based on the optimized detection model, the key points of the core rod center axis are located in the infrared image of the composite insulator to be tested.

[0011] Preferably, the method further comprises:

[0012] Denoising and data enhancement processing are performed on the infrared image of the composite insulator; wherein the data enhancement processing includes: random flipping and rotation, cropping, scaling, adding noise, blurring and sharpening.

[0013] Preferably, the backbone extraction network of the model uses CSPDarknet, the feature fusion network of the model uses PAN-FPN, the detection head of the model uses Decoupled-Head, and the HRNet model is used for keypoint detection; the optimizer used during model training is the Adam algorithm; wherein, a PAN-FPN network is constructed on the output feature map of CSPDarknet to fuse high-dimensional and low-dimensional features from different levels; a Decoupled-Head is set on the output of the PAN-FPN network to be used for classification and regression tasks in object detection respectively.

[0014] Preferably, the model improves the asymmetric weighted operation and uses VFL Loss, and the calculation formula is:

[0015]

[0016] Use DFL to adjust the loss, including:

[0017] DFL(S i , S i+1 )=-((y i+1 -y)log(S i )+(yy i )log(S i+1 )),

[0018] Among them, q is the sample; p is the predicted value; α and γ are hyperparameters that can be adjusted; y, y i and y i+1 are adjacent sample labels, where y i ≤y≤y i+1 ,at the same time Its meaning is to optimize the probability of the two positions closest to the label, one on the left and one on the right, in the form of cross entropy.

[0019] Preferably, the method further comprises:

[0020] Obtain a test set and evaluate the detection performance of the optimized detection model based on the test set.

[0021] According to another aspect of the present invention, a composite insulator core rod central axis intelligent positioning system using infrared imaging is provided, the system comprising:

[0022] A data acquisition unit is used to acquire an infrared image data set of a composite insulator and mark key points on the central axis of the anchor frame and core rod of the composite insulator to obtain marked data;

[0023] The model building unit is used to build a YOLO deep learning model and combine key point detection technology to mark key points of the mandrel position in infrared images in order to obtain an initial detection model.

[0024] A training unit, configured to train the initial detection model based on the labeled data to obtain an optimized detection model;

[0025] The positioning unit is used to locate the key points of the core rod center axis of the infrared image of the composite insulator to be tested based on the optimized detection model.

[0026] Preferably, the system further comprises:

[0027] The data processing unit is used to perform denoising and data enhancement processing on the infrared image of the composite insulator; wherein the data enhancement processing includes: random flipping and rotation, cropping, scaling, adding noise, blurring and sharpening.

[0028] Preferably, the backbone extraction network of the model uses CSPDarknet, the feature fusion network of the model uses PAN-FPN, the detection head of the model uses Decoupled-Head, and the HRNet model is used for keypoint detection; the optimizer used during model training is the Adam algorithm; wherein, a PAN-FPN network is constructed on the output feature map of CSPDarknet to fuse high-dimensional and low-dimensional features from different levels; a Decoupled-Head is set on the output of the PAN-FPN network to be used for classification and regression tasks in object detection respectively.

[0029] Preferably, the model improves the asymmetric weighted operation and uses VFL Loss, and the calculation formula is:

[0030]

[0031] Use DFL to adjust the loss, including:

[0032] DFL(S i , S i+1 )=-((y i+1 -y)log(S i )+(yy i )log(S i+1 )),

[0033] Among them, q is the sample; p is the predicted value; α and γ are hyperparameters that can be adjusted; y, y i and y i+1 are adjacent sample labels, where y i ≤y≤y i+1 ,at the same time Its meaning is to optimize the probability of the two positions closest to the label, one on the left and one on the right, in the form of cross entropy.

[0034] Preferably, the system further comprises:

[0035] The testing unit is used to obtain a test set and evaluate the detection performance of the optimized detection model based on the test set.

[0036] The present invention provides a method and system for intelligently locating the center axis of a composite insulator core rod using infrared images, comprising: acquiring a composite insulator infrared image dataset and annotating key points on the composite insulator's anchor frame and core rod center axis to obtain annotated data; establishing a Yolo deep learning model and combining key point detection technology to mark key points of the core rod position in the infrared image to obtain an initial detection model; training the initial detection model based on the annotated data to obtain an optimized detection model; and locating the key points of the core rod center axis in the infrared image of the composite insulator to be tested based on the optimized detection model. The present invention combines deep learning and key point detection technology to intelligently identify the core rod center axis in the composite insulator infrared image, avoiding the reliance on complex rules in traditional methods. It can also rapidly process and identify new composite insulator infrared images, enabling real-time monitoring and analysis. This can significantly reduce manual intervention and improve detection efficiency, allowing for timely implementation of maintenance or repair measures, making it suitable for industrial promotion. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] A more complete understanding of exemplary embodiments of the present invention may be obtained by referring to the following drawings:

[0038] Figure 1 Flowchart of a method 100 for intelligently locating the central axis of a composite insulator core rod using infrared images according to an embodiment of the present invention;

[0039] Figure 2 A schematic structural diagram of a Yolo model according to an embodiment of the present invention;

[0040] Figure 3 Schematic diagram of the structure of the composite insulator infrared image core rod central axis intelligent positioning system 300 according to an embodiment of the present invention. DETAILED DESCRIPTION

[0041] Exemplary embodiments of the present invention will now be described with reference to the accompanying drawings. However, the present invention may be embodied in many different forms and is not limited to the embodiments described herein. These embodiments are provided to provide a thorough and complete disclosure of the present invention and to fully convey the scope of the present invention to those skilled in the art. The terminology used in the exemplary embodiments shown in the accompanying drawings is not intended to limit the present invention. In the accompanying drawings, identical elements are denoted by the same reference numerals.

[0042] Unless otherwise specified, the terms used herein (including technical terms) have the meanings commonly understood by those skilled in the art. In addition, it is understood that terms defined in commonly used dictionaries should be understood to have the same meanings as those in the context of the relevant fields, and should not be understood as idealized or overly formal meanings.

[0043] Figure 1 FIG. 1 is a flow chart of a composite insulator infrared image core rod central axis intelligent positioning method 100 according to an embodiment of the present invention. Figure 1 As shown, the intelligent positioning method for the central axis of the core rod of a composite insulator using infrared images provided by an embodiment of the present invention combines deep learning and key point detection technology, and can intelligently identify the central axis of the core rod in the infrared image of the composite insulator, avoiding the reliance on complex rules in traditional methods. At the same time, it can quickly process and identify new infrared images of composite insulators, realize real-time monitoring and analysis, and can greatly reduce manual intervention and improve detection efficiency so that maintenance or repair measures can be taken in a timely manner. It is suitable for industrial promotion. The intelligent positioning method 100 for the central axis of the core rod of a composite insulator using infrared images provided by the embodiment of the present invention starts at step 101. In step 101, a composite insulator infrared image data set is obtained, and key points on the anchor frame and the central axis of the core rod of the composite insulator are annotated to obtain annotated data.

[0044] Preferably, the method further comprises:

[0045] Denoising and data enhancement processing are performed on the infrared image of the composite insulator; wherein the data enhancement processing includes: random flipping and rotation, cropping, scaling, adding noise, blurring and sharpening.

[0046] In the present invention, infrared images of composite insulators of power transmission lines taken by drones are collected and annotated, and the annotations include key points on the anchor frame and the central axis of the core rod of the composite insulator.

[0047] Furthermore, the acquired infrared images can be preprocessed to obtain new datasets, including: denoising the composite insulator infrared images using denoising algorithms to improve image quality, and simultaneously augmenting the dataset using data augmentation. Specifically, various image enhancement methods such as random flipping and rotation, cropping, scaling, adding noise, blurring, and sharpening are used to expand the composite insulator infrared image dataset, increasing data diversity and model robustness. Since the colors in infrared images contain temperature information, image enhancement methods related to color changes, such as color channel adjustment and contrast adjustment, are not used in this invention.

[0048] In step 102, a Yolo deep learning model is established, and key points of the core rod position in the infrared image are marked in combination with key point detection technology to obtain an initial detection model.

[0049] Preferably, the backbone extraction network of the model uses CSPDarknet, the feature fusion network of the model uses PAN-FPN, the detection head of the model uses Decoupled-Head, and the HRNet model is used for keypoint detection; the optimizer used during model training is the Adam algorithm; wherein, a PAN-FPN network is constructed on the output feature map of CSPDarknet to fuse high-dimensional and low-dimensional features from different levels; a Decoupled-Head is set on the output of the PAN-FPN network to be used for classification and regression tasks in object detection respectively.

[0050] Preferably, the model improves the asymmetric weighted operation and uses VFL Loss, and the calculation formula is:

[0051]

[0052] Use DFL to adjust the loss, including:

[0053] DFL(S i , S i+1 )=-((y i+1 -y)log(S i )+(yy i )log(S i+1 )),

[0054] Among them, q is the sample; p is the predicted value; α and γ are hyperparameters that can be adjusted; y, y i and y i+1 are adjacent sample labels, where yi ≤y≤y i+1 ,at the same time Its meaning is to optimize the probability of the two positions closest to the label, one on the left and one on the right, in the form of cross entropy.

[0055] In the present invention, a Yolo model is constructed, and the backbone extraction network (Backbone) of the model uses CSPDarknet, which can reduce the amount of calculation and enhance the performance of gradients compared to ResNet; the feature fusion network (Neck) of the model uses PAN-FPN, and the model can fuse the high-dimensional and low-dimensional information extracted from the Backbone, enhancing the semantic expression and positioning capabilities of the model at multiple scales. The detection head (Head) of the model uses Decoupled-Head, so that the Heads of the classification and regression tasks no longer share parameters. A key point detection module is designed, and the HRNet model is used for key point detection. HRNet is a key point detection algorithm based on a high-resolution feature pyramid network, which has advantages in accuracy and speed. After determining the model structure, it is also necessary to select the training settings of the model, including the optimizer and loss function of the model.

[0056] When creating a model, the present invention simultaneously builds multiple models of different sizes. In theory, smaller models can achieve faster execution speeds and smaller memory requirements, thus reducing the hardware requirements for model operation. Larger models can achieve higher accuracy. Multiple models of different sizes can be applied in different application scenarios.

[0057] Furthermore, when setting the optimizer and loss function, the asymmetric weighting operation is improved to address the imbalance of positive and negative samples and the unequal weights in positive samples. The main method is to use VFL Loss, and its operation is shown in formula (1):

[0058]

[0059] Where q is the sample, p is the predicted value, and α and γ are adjustable hyperparameters.

[0060] In order to quickly focus on the distribution of locations close to the target location during model training, DFL is used to adjust the loss, as shown in formula (2):

[0061] DFL(S i , S i+1 )=-((y i+1 -y)log(S i )+(yy i )log(S i+1 )) (2)

[0062] Among them, y, yi and y i+1 are adjacent sample labels, where y i ≤y≤y i+1 ,at the same time Its meaning is to optimize the probabilities of the two positions closest to the label y in the form of cross entropy, so that the network can focus on the distribution of the neighboring area of ​​the target position more quickly.

[0063] In step 103, the initial detection model is trained based on the labeled data to obtain an optimized detection model.

[0064] Preferably, the method further comprises:

[0065] Obtain a test set and evaluate the detection performance of the optimized detection model based on the test set.

[0066] In this paper, the established composite insulator infrared image dataset is used to train the Yolo key point detection model. During the training process, the model minimizes the classification and regression loss values, uses the Adam optimizer to optimize the model gradient, and finally determines the model weight so that it can achieve the best composite insulator core rod centerline positioning effect. Among them, the model's prediction results are compared with the true label, and the model's performance and error are displayed in a visual way to help understand the model's performance.

[0067] The present invention can also prepare an independent test set containing a series of infrared images of composite insulators that do not overlap with the data in the training set. The trained model can be evaluated using the test set, and metrics such as the model's precision, recall, and F1 score on the test set can be calculated to assess model performance. The model's performance on the test set can also be analyzed to identify patterns and causes of model prediction errors. Based on the test results, model hyperparameters, such as the learning rate and batch size, can be adjusted to improve model performance.

[0068] In step 104, the infrared image of the composite insulator to be tested is used to locate the key points of the core rod center axis based on the optimized detection model.

[0069] In the present invention, after the training of the model is completed, the infrared image of the composite insulator to be tested is input into the model so as to locate the key points of the core rod center axis.

[0070] The method of the present invention can realize intelligent recognition of UAV inspection images, greatly reduce manual intervention and improve detection efficiency, so that maintenance or repair measures can be taken in a timely manner.

[0071] The following examples illustrate the embodiments of the present invention.

[0072] In an embodiment of the present invention, positioning of key points of the mandrel centerline is achieved based on the following steps, including:

[0073] The drone is equipped with an infrared camera to ensure that it can capture infrared images of the composite insulators on the transmission line. The captured infrared images of the composite insulators are screened to eliminate incomplete, heavily obscured, unclear, and other poor quality images. The infrared images are annotated with anchor frames using the labelme annotation tool, marking their positions and bounding boxes. The coordinates of the upper left and lower right corners of the anchor frame are (x1, y1) and (x2, y2), respectively. Key points on the centerline of the composite insulator core rod are also annotated, with the top and bottom of the composite insulator selected as key points, k1 and k2, respectively. The anchor frame coordinates and key point coordinates are entered into a txt file in the Yolo dataset format as the dataset label file for subsequent use.

[0074] Preprocessing and data augmentation of existing infrared images are crucial steps for improving image quality and increasing data diversity. Denoising algorithms are used to reduce noise and improve image quality, primarily using Gaussian and bilateral filtering. Gaussian filtering blurs the image by taking a weighted average of the pixels surrounding each pixel's neighborhood. Bilateral filtering considers spatial distance and grayscale differences between pixels, preserving edge information while removing noise. Data augmentation techniques are used to increase dataset diversity and improve model robustness and generalization. These techniques include random flipping (randomly flipping images horizontally or vertically); random cropping (cropping a random region within an image); and noise addition (adding random Gaussian or salt-and-pepper noise). These image augmentation methods expand the dataset to eight times its original size.

[0075] Build a Yolo model, where the backbone is CSPDarknet, a lightweight deep learning model backbone network that reduces the amount of computation through cross-stage connections and enhances the performance of gradients, thereby improving the training and reasoning efficiency of the model. Figure 2 As shown in the figure, the CSPDarknet53 version is selected as the backbone network of the model. A PAN-FPN network is constructed on the output feature map of CSPDarknet, fusing high-dimensional and low-dimensional features from different layers to obtain a richer and more global feature representation. A Decoupled Head is set on the output of the PAN-FPN network to handle the classification and regression tasks in object detection, respectively, ensuring the effectiveness and independence of classification and localization capabilities.

[0076] A suitable HRNet model is selected and connected to the model as an additional key point detection module. At the same time, the features of various scales obtained by the PAN-FPN module in the model are input into HRNet for analysis. The features of different branches are fused through the feature fusion module, so that the model can directly analyze and obtain the key points of the center axis of the composite insulator core rod on the anchor frame output by the model.

[0077] Determine the optimizer and loss function to be used during model training: The Adam algorithm is chosen as the optimizer, and its formula is shown below:

[0078]

[0079] Where v dW (l) is the exponentially weighted average of the gradient, S dW (l) is the exponentially weighted average of the squared gradients, β1 and β2 are adjustable hyperparameters, α is the network learning rate, and ε is a very small real number to prevent the denominator from being zero. The Adam algorithm has the advantages of fast convergence and good learning effect, and is not prone to vanishing learning rates or violent fluctuations in the loss function.

[0080] The loss function formula for model training is shown below:

[0081] L train =L cls +L reg +L keypoint (4)

[0082] L train The total loss during training, where L cls For the classification loss of object detection, L reg For the regression loss of object detection, L keypoint Loss due to keypoint detection. L cls and L reg As shown in equations (1) and (2), L keypoint This is the mean squared error, and its formula is as follows:

[0083]

[0084] Where m is the number of keypoints in a single sample, which is set to 2 here, and k and are the true value and predicted value of the key point respectively. In the key point detection task, the mean square error between the predicted value and the true value of each key point is used as the loss function, and then the loss of all key points is averaged to measure the distance between the predicted value and the true value.

[0085] The model is trained by minimizing the total loss L during the training process. train, use the optimizer to adjust the weights of each layer in the model. During the training process, the weights with the best detection effect on the validation set are saved to obtain the final mandrel centerline detection model.

[0086] The model is evaluated using a test set. The main evaluation metrics include: precision (P), recall (R), F1 score, intersection-over-union (IOU), and average precision (AP). The formulas for each metric are shown below:

[0087]

[0088]

[0089]

[0090] Intersection over Union (IOU) is the overlap ratio between the generated candidate boxes and the original label boxes, that is, the ratio of their intersection to their union. The ideal situation is complete overlap, that is, a ratio of 1.

[0091] AP represents the average value of the detector under various R conditions. The formula is as follows:

[0092]

[0093] During testing, the above evaluation indicators are used to comprehensively evaluate the model performance.

[0094] Analyze the model's performance on the test set. Analyze the samples in the test set to determine which samples the model performs poorly on. Also, identify common types of prediction errors, such as missed detections, false detections, and inaccurate target positioning. Check whether the model suffers from underfitting or overfitting. Based on the test results, adjust the model's hyperparameters, such as the learning rate and batch size, to improve performance. Retrain the model using the adjusted hyperparameters and evaluate it on the test set. Continue iterating these steps, further adjusting the model's hyperparameters based on test results and feedback, until satisfactory performance is achieved.

[0095] Finally, the trained model is used to locate the key points of the core rod centerline in the infrared image of the composite insulator to be tested.

[0096] Figure 3 FIG. 3 is a schematic structural diagram of an intelligent positioning system 300 for a composite insulator core rod central axis using infrared imaging according to an embodiment of the present invention. Figure 3 As shown, the composite insulator infrared image core rod central axis intelligent positioning system 300 provided by the embodiment of the present invention includes: a data acquisition unit 301, a model building unit 302, a training unit 303 and a positioning unit 304.

[0097] Preferably, the data acquisition unit 301 is used to acquire infrared image datasets of composite insulators and to annotate key points on the central axis of the anchor frame and core rod of the composite insulators to obtain annotation data.

[0098] Preferably, the system further comprises:

[0099] The data processing unit is used to perform denoising and data enhancement processing on the infrared image of the composite insulator; wherein the data enhancement processing includes: random flipping and rotation, cropping, scaling, adding noise, blurring and sharpening.

[0100] Preferably, the model building unit 302 is used to build a YOLO deep learning model and combine key point detection technology to mark key points of the mandrel position in the infrared image to obtain an initial detection model.

[0101] Preferably, the backbone extraction network of the model uses CSPDarknet, the feature fusion network of the model uses PAN-FPN, the detection head of the model uses Decoupled-Head, and the HRNet model is used for keypoint detection; the optimizer used during model training is the Adam algorithm; wherein, a PAN-FPN network is constructed on the output feature map of CSPDarknet to fuse high-dimensional and low-dimensional features from different levels; a Decoupled-Head is set on the output of the PAN-FPN network to be used for classification and regression tasks in object detection respectively.

[0102] Preferably, the model improves the asymmetric weighted operation and uses VFL Loss, and the calculation formula is:

[0103]

[0104] Use DFL to adjust the loss, including:

[0105] DFL(S i , S i+1 )=-((y i+1 -y)log(S i )+(yy i )log(S i+1 )),

[0106] Among them, q is the sample; p is the predicted value; α and γ are hyperparameters that can be adjusted; y, y i and y i+1 are adjacent sample labels, where y i ≤y≤y i+1 ,at the same time Its meaning is to optimize the probability of the two positions closest to the label, one on the left and one on the right, in the form of cross entropy.

[0107] Preferably, the training unit 303 is used to train the initial detection model based on the labeled data to obtain an optimized detection model.

[0108] Preferably, the system further comprises:

[0109] The testing unit is used to obtain a test set and evaluate the detection performance of the optimized detection model based on the test set.

[0110] Preferably, the positioning unit 304 is used to locate key points of the core rod axis based on the infrared image of the composite insulator under test using the optimized detection model.

[0111] The composite insulator infrared image core rod center axis intelligent positioning system 300 of the embodiment of the present invention corresponds to the composite insulator infrared image core rod center axis intelligent positioning method 100 of another embodiment of the present invention, and will not be repeated here.

[0112] The present invention has been described with reference to a few embodiments. However, it is apparent to those skilled in the art that other embodiments than the ones disclosed above are equally within the scope of the present invention.

[0113] Generally, all terms used in this disclosure are to be interpreted according to their ordinary meaning in the art, unless explicitly defined otherwise herein. All references to "a / the / the [device, component, etc.]" are to be interpreted openly as referring to at least one instance of the device, component, etc., unless explicitly stated otherwise. The steps of any method disclosed herein do not necessarily need to be performed in the exact order disclosed, unless explicitly stated otherwise.

[0114] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0115] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0116] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0117] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0118] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the present invention.

Claims

1. A composite insulator infrared image core rod center axis intelligent positioning method, characterized in that: The method comprises: Acquire a composite insulator infrared image dataset, and annotate key points on the center axis of the composite insulator anchor frame and core rod to obtain annotated data; A Yolo deep learning model was established and key point detection technology was used to mark the key points of the core rod position in the infrared image to obtain an initial detection model. Training the initial detection model based on the labeled data to obtain an optimized detection model; Based on the optimized detection model, the key points of the core rod center axis are located in the infrared image of the composite insulator to be tested.

2. The method according to claim 1, characterized in that The method further comprises: The composite insulator infrared image is subjected to denoising and data enhancement processing; wherein the data enhancement processing includes: random flipping and rotation, cropping, scaling, adding noise, blurring and sharpening.

3. The method according to claim 1, characterized in that The backbone extraction network of the model uses CSPDarknet, the feature fusion network of the model uses PAN-FPN, the detection head of the model uses Decoupled-Head, and the HRNet model is used for key point detection; the optimizer used in model training selects the Adam algorithm; among them, the PAN-FPN network is constructed on the output feature map of CSPDarknet to fuse high-dimensional and low-dimensional features from different levels; the Decoupled-Head is set on the output of the PAN-FPN network to process classification and regression tasks in target detection respectively.

4. The method according to claim 1, characterized in that: The model improves the asymmetric weighted operation and uses VFLLoss. The calculation formula is: Use DFL to adjust the loss, including: DFL(S i ,S i+1 )=-((y i+1 -y)log(S i )+(y-y i )log(S i+1 )), Among them, q is the sample; p is the predicted value; α and γ are hyperparameters that can be adjusted; y, y i and i+1 are adjacent sample labels, where y i ≤y≤y i+1 ,at the same time Its meaning is to optimize the probabilities of the two positions closest to the label, one on the left and one on the right, in the form of cross entropy.

5. The method according to claim 1, characterized in that The method further comprises: A test set is obtained, and detection performance of the optimized detection model is evaluated based on the test set.

6. An intelligent positioning system for the central axis of a composite insulator core rod using infrared imaging, characterized in that: The system comprises: A data acquisition unit is used to acquire an infrared image data set of a composite insulator, and to mark key points on the central axis of an anchor frame and a core rod of the composite insulator to obtain marked data; A model building unit is used to build a Yolo deep learning model and mark the key points of the core rod position in the infrared image in combination with key point detection technology to obtain an initial detection model; A training unit, used for training the initial detection model based on the labeled data to obtain an optimized detection model; The positioning unit is used to locate the key points of the core rod center axis of the infrared image of the composite insulator to be tested based on the optimized detection model.

7. The system according to claim 6, characterized in that The system further comprises: The data processing unit is used for performing denoising and data enhancement processing on the composite insulator infrared image; wherein the data enhancement processing includes: random flipping and rotation, cropping, scaling, adding noise, blurring and sharpening.

8. The system according to claim 6, characterized in that The backbone extraction network of the model uses CSPDarknet, the feature fusion network of the model uses PAN-FPN, the detection head of the model uses Decoupled-Head, and the HRNet model is used for key point detection; the optimizer used in model training selects the Adam algorithm; among them, the PAN-FPN network is constructed on the output feature map of CSPDarknet to fuse high-dimensional and low-dimensional features from different levels; the Decoupled-Head is set on the output of the PAN-FPN network to process classification and regression tasks in target detection respectively.

9. The system according to claim 6, characterized in that The model improves the asymmetric weighted operation and uses VFLLoss. The calculation formula is: Use DFL to adjust the loss, including: DFL(S i ,S i+1 )=((y i+1 -y)log(S i )+(y-y i )log(S i+1 )), Among them, q is the sample; p is the predicted value; α and γ are hyperparameters that can be adjusted; y, y i and i+1 are adjacent sample labels, where y i ≤y≤y i+1 ,at the same time Its meaning is to optimize the probabilities of the two positions closest to the label, one on the left and one on the right, in the form of cross entropy.

10. The system according to claim 6, characterized in that The system further comprises: The testing unit is used to obtain a test set and evaluate the detection performance of the optimized detection model based on the test set.

Images