Method and device for detecting defects of high-altitude cable protective sleeve

Abstract

The invention discloses a method and a device for detecting defects of a high-altitude cable protective sleeve. The method comprises the following steps: step 1, acquiring an image of the surface of the high-altitude cable protective sleeve; step 2, carrying out registration and alignment on the RGB image and the infrared thermal image; step 3, performing defect category and position marking on the image of the surface of the high-altitude cable protection sleeve, generating a surface defect data set of the high-altitude cable protection sleeve, and preprocessing the data set; step 4, based on the high-altitude cable protection sleeve surface defect data set, fusing the RGB image and the infrared thermal image to generate a corresponding fusion weight; step 5, constructing an HL-YOLO model based on the fused RGB image and the infrared thermal image; step 6, training the HL-YOLO model based on the surface defect data set of the high-altitude cable protection sleeve; and step 7, performing real-time detection on the high-altitude cable protection sleeve by using the trained HL-YOLO model. Compared with traditional manual detection, the method has the advantages of being good in detection effect, high in stability, high in detection speed and low in labor cost.

Description

Technical Field

[0001] The invention relates to the field of defect detection, and in particular to a method and a device for detecting defects in a high-altitude cable protective cover. Background Art

[0002] In the context of the continuous advancement of urban modernization construction today, high-voltage cables, as an indispensable and important part of the power supply system, play a vital role in the field of urban power supply. As an important protective part of high-voltage cables, the detection of surface defects of cable sheaths is particularly important. These defects are affected by many factors, including external impact, climate and environmental factors, cable aging, design defects and external corrosive substances. Any degree of damage may gradually evolve into a fault point, threatening the safe operation of the power system. The traditional manual inspection method has many shortcomings in detecting surface defects of high-voltage cable sheaths: strong subjectivity leads to inconsistent judgment standards; reliance on manpower is time-consuming and labor-intensive and susceptible to personnel fatigue; limited coverage may result in blind spots; low sensitivity to subtle defects and easy to miss detection; imperfect data recording and analysis have the risk of recording errors, which poses a serious safety hazard to the long-term operation of the cable.

[0003] At present, the methods for live detection of cables in operation mainly include ultrasonic and high-frequency partial discharge detection. However, these methods are easily affected by on-site interference. Another method is to use X-ray image analysis, which can intuitively display the internal structure of the equipment, but sometimes it is affected by ray scattering, affecting the clarity of the image. In general, these existing methods require engineers to make judgments based on waveforms or images combined with experience, which is not conducive to quickly and accurately determining the type of fault.

[0004] Chinese patent publication number CN117309885A, entitled A multi-spectral fusion high-voltage cable defect detection method and system, discloses a multi-spectral fusion high-voltage cable defect detection method and system, the detection system includes a spectral excitation module, a spectral receiving module, a system control module, and a data and graphics processing module; and provides a multi-spectral fusion high-voltage cable defect detection method and system, which integrates three types of spectra: visible light, infrared, and X-rays. The collected information is richer, and the processed fusion graphics are easier to find the location of the fault point than the existing methods; it is conducive to quickly and accurately judging the fault type. However, the invention is costly and still relies on manual analysis and judgment of the collected data and graphics, and cannot achieve automated detection. Summary of the invention

[0005] The purpose of the present invention is to overcome the shortcomings of the prior art that defect detection of high-altitude cable protective covers is easily affected by on-site interference, requires manual analysis and judgment, cannot achieve automated detection, and has low defect detection efficiency, and to provide a method and device for defect detection of high-altitude cable protective covers.

[0006] The purpose of the present invention is achieved through the following technical solutions: A method for detecting defects in a high-altitude cable protective cover comprises the following steps: Step 1, collecting images of the surface of the high-altitude cable protective cover, the images include RGB images and infrared thermal images; Step 2: align the RGB image and the infrared thermal map; Step 3, mark the defect category and position of the image of the surface of the high-altitude cable protective cover, generate a high-altitude cable protective cover surface defect dataset, and preprocess the dataset; Step 4: Based on the surface defect dataset of high-altitude cable protective covers, the RGB image and the infrared thermal map are fused to generate the corresponding fusion weights; Step 5, construct the HL-YOLO model based on the fused RGB image and infrared thermal map; Step 6, training the HL-YOLO model based on the surface defect dataset of high-altitude cable protective cover; Step 7: Use the trained HL-YOLO model to perform real-time detection of high-altitude cable protective covers.

[0007] YOLO (You Only Look Once) is a target detection algorithm that realizes fast and accurate detection of objects in images by converting the target detection task into a single regression problem. The HL-YOLO model is an improvement on the YOLO model, and is particularly suitable for image recognition of the surface of high-altitude cable protective covers in this solution. The design of this solution can obtain more information about the surface of the cable protective cover by collecting RGB images and infrared thermal images. RGB images provide visual features such as color and texture, while infrared thermal images reveal temperature distribution, which helps to identify temperature changes caused by defects. The combination of the two can more accurately detect and identify defects in the protective cover. By constructing and preprocessing a dataset of surface defects of high-altitude cable protective covers, the model can learn more diverse defect samples during training, thereby improving its generalization ability. Using the trained HL-YOLO model for real-time detection, the defects of the protective cover can be quickly and accurately identified, providing strong support for timely maintenance and replacement. This helps to reduce cable failures caused by defects and ensure the stable operation of the power system.

[0008] Preferably, in the step 2, after the RGB image and the infrared thermal map are registered and aligned, the data is enhanced, and the specific operations are: brightness adjustment, random image flipping and grayscale conversion.

[0009] Preferably, in step 5, the HL-YOLO model is constructed by adding the following algorithm on the basis of the YOLOv8 algorithm: a) Using transfer learning, HL-YOLO has the ability to fuse RGB images and infrared thermal maps; b) Add deformable convolution; c) Introducing the MobileNet-V3 lightweight neural network; d) Optimize the CIoU loss function.

[0010] Preferably, the use of transfer learning enables HL-YOLO to have the ability to fuse RGB images and infrared thermal images, specifically: Step 301, using the canny operator to perform preprocessing on the edge blur problem of the infrared thermal image; Step 302, designing a feature fusion network FusionNet (a deep learning network that combines multiple data sources for depth map completion or semantic segmentation) to fuse the two types of images into one image; Step 303, training FusionNet based on the high-altitude cable protective cover defect dataset; Step 304, using transfer learning to extract the trained FusionNet encoding part separately; Step 305, placing the FusionNet encoding part in the YOLOv8 header to implement the network's fusion function for the two types of images.

[0011] Preferably, the adding of deformable convolution is specifically as follows: Step 401, the second layer of the three-layer shunt in the C2f module is fused with a deformable convolution to make the model receptive field closer to the defect characteristics of the cable protective cover, wherein the formula of the deformable convolution is: Among them, w is the weight matrix, x is the feature matrix covered by the convolution kernel, p0 is the tensor position passed by the center of the convolution kernel, and p n For each offset of p0 within the convolution kernel range, Δp n Then the additional learnable offset added to each offset of the convolution kernel is Step 402: Replace the original activation function in C2f with the H-Swish activation function, whose formula is: Among them, relu represents the ReLU function; Step S403, replacing the second and fourth layers in the backbone network of the HL-YOLO model with the improved C2f module, so that the model attention distribution is more consistent with the characteristics of the slender defect of the cable protective sheath.

[0012] Preferably, the optimized CIoU loss function is: On the basis of the original CIoU penalty term, the EIoU penalty for the length and width values ​​is introduced, and an adjustable parameter α is added to adjust the proportion of each penalty term. The formula is: Among them, ρ represents the Euclidean distance calculation formula, b, w and h represent the information of the center point, width and height of the prediction box respectively, b gt 、w gt and h gt They represent the center point, width and height of the target box respectively, c represents the diagonal line containing the minimum area of ​​the predicted box and the real box, and c w and c h They represent the width and height of the minimum bounding box covering the predicted box and the true box, respectively. α is an adjustable parameter used to manually adjust the penalty term.

[0013] Preferably, in step 6, the training process of the HL-YOLO model based on the surface defect dataset of the high-altitude cable protective cover is as follows: the training parameters used include the experimental training rounds epochs, the batch-size of the file fed into the batch, the maximum number of working cores workers, the initial learning rate Ir0 and the learning momentum momentum, wherein the learning momentum is used to adjust the learning rate of the cosine annealing algorithm.

[0014] Preferably, in the step 7, the trained HL-YOLO model is used to perform real-time detection of the high-altitude cable protective cover under a human-computer interaction interface, and the human-computer interaction interface runs under a Windows system or a Linux system.

[0015] A device for detecting defects in a high-altitude cable protective cover, comprising a protective shell, a roller, a driving motor, a camera module, a control unit and an LED light strip; the camera module is installed in the middle part of the inner side of the protective shell; the driving motor and the roller are driven by a gear belt, and are installed at both ends of the protective shell to provide power for the entire device; the camera module collects data and transmits it to the control unit, and the control unit transmits the data to a terminal device such as a handle or a computer that can be connected to the network by wireless communication; the terminal device is equipped with a script file of a human-computer interaction interface, and the script file contains a pre-trained HL-YOLO model, which is used to detect the captured image in real time, and store the final detection result in a text box of the human-computer interaction interface to realize the recording function.

[0016] The beneficial effects of the present invention are as follows: compared with basic target detection algorithm model systems such as YOLOv8, the present invention has been improved to generate an HL-YOLO model, which has the ability to fuse RGB images and infrared thermal imaging, and has the advantages of low complexity, fast detection speed, high detection accuracy, and strong real-time performance, meeting the requirements of real-time performance and accuracy in actual detection; Based on YOLOv8, the present invention adopts transfer learning to combine the pre-trained fusion network with the YOLOv8 backbone network; the second channel of the original C2f module is enhanced by deformable convolution, and at the same time, in order to increase the computational efficiency, the original activation function is replaced by the H-Swish activation function, so that the network model has a further improvement in the feature extraction capability of slender defects.

[0017] The present invention replaces part of the backbone network with a lightweight network MobileNet-V3, which effectively improves the training speed and reduces the amount of calculation, and can realize real-time high-altitude cable protective cover defect detection. MobileNet-V3 adopts an inverted residual structure and a linear bottleneck design, which effectively reduces the computational overhead and the number of parameters in the network, while maintaining the representation ability of the model, bringing conditions for its operation on embedded devices.

[0018] The present invention uses α-IoU and EIoU to jointly optimize the original loss function as a new loss function. The penalty term of EIoU on the true length and width of the target box is added to the original loss function, and an adjustable parameter α is assigned to each of these penalty terms. The optimal parameters can be selected through manual screening, so that the model converges faster and better.

[0019] Compared with traditional manual detection, the present invention has the advantages of good detection effect, strong stability, fast detection speed and low labor cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 is a flow chart of the present invention; Figure 2 It is a structural diagram of a camera module of an image acquisition unit in the present invention; Figure 3 This is a schematic diagram of the fusion principle of RGB image and infrared thermal image in the present invention; Figure 4 It is a schematic diagram of the MobileNet-V3 lightweight neural network in the present invention; Figure 5 It is the backbone network diagram of the method for detecting defects of the protective cover of an aerial cable in the present invention; Figure 6 It is a schematic diagram of the human-machine interaction interface of the high-altitude cable protective cover defect detection system of the present invention; Figure 7 It is a schematic diagram of the mechanism of the high-altitude cable protective cover defect detection device of the present invention.

[0021] Among them: 1. Protective shell, 2. Roller, 3. Drive motor, 4. Camera module, 5. Control unit, 6. LED light strip. DETAILED DESCRIPTION

[0022] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this application will be more comprehensive and complete and fully convey the concept of the example embodiments to those skilled in the art.

[0023] In addition, described feature, structure or characteristic can be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present application. However, those skilled in the art will appreciate that the technical scheme of the present application can be put into practice without one or more of the specific details, or other methods, components, devices, steps, etc. can be adopted. In other cases, known methods, devices, realizations or operations are not shown or described in detail to avoid blurring the various aspects of the application.

[0024] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities may be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0025] The flowcharts shown in the accompanying drawings are only exemplary and do not necessarily include all contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps can be decomposed, and some operations / steps can be combined or partially combined, so the actual execution order may change according to actual conditions. A method for detecting defects in a high-altitude cable protective cover, such as Figure 1 As shown in the figure, the RGB image and infrared thermal image of the surface of the high-altitude cable protective cover are first collected by using the camera and the infrared thermal imaging module respectively. In order to improve the accuracy of image fusion, the collected RGB image and infrared thermal image are registered and aligned by Opencv, and then the LabelImg tool is used to mark the defect category and position of the registered image, and the high-altitude cable protective cover defect data set is generated and preprocessed. Then, the RGB image and the infrared thermal image are fused by the image fusion algorithm FusionNet, and the trained FusionNet is fused to YOLOv8 through transfer learning, and the deformable convolution is used to improve the algorithm model to HL-YOLO, and the data features are processed. Subsequently, the HL-YOLO model is trained based on the high-altitude cable protective cover defect data set to obtain the weight file. Finally, the trained weight file is embedded in the human-computer interaction interface of the high-altitude cable protective cover defect detection system built based on Tkinter to realize automatic real-time detection of high-altitude cable protective covers.

[0026] The specific steps are as follows: Step S1, using a camera and an infrared thermal imaging module to respectively collect an RGB image and an infrared thermal image of the surface of the high-altitude cable protective cover.

[0027] The defect image of the high-altitude cable protective cover can be obtained through Figure 7 After the device is fixed to the cable and the power is turned on, the device can crawl along the cable, and the camera module installed inside will collect real-time images and store them in two folders.

[0028] Step S2: Use Opencv to align the collected RGB image and infrared thermal map.

[0029] Since the camera and the infrared thermal imaging module are arranged in parallel, Figure 2 As shown, the perspective difference is caused. In order to correct the perspective difference and achieve better fusion, the two types of images are registered and corrected by using feature point matching.

[0030] Step S3: Use the LabelImg tool to mark the defect category and position of the collected high-altitude cable protective cover images, generate a high-altitude cable protective cover surface defect dataset, and preprocess the dataset.

[0031] Among them, the image is obtained through the high-altitude cable protective cover defect detection equipment, and the image is enhanced, mainly including random image flipping, grayscale conversion, brightness adjustment and other data enhancement operations. It should be noted that the corresponding two types of images need to use the same enhancement method when enhancing, especially the operations involving position transformation such as inversion cropping. After the processing is completed, 2000 high-altitude cable protective cover defect images are finally obtained, and the data set is divided into training set, verification set and test set in a ratio of 7:2:1. The experimental sample data are manually annotated using the labeling software LabelImg, and the corresponding xml format file is generated, and then converted into the txt format for YOLO training.

[0032] Step S4, based on the high-altitude cable protective cover defect dataset, use FusionNet to fuse the RGB image and the infrared thermal map to generate the corresponding fusion weight.

[0033] In order to achieve better fusion effect, a separate Figure 3 The FusionNet network shown is trained and fused. The specific steps are as follows: 1) To solve the edge blur problem of infrared thermal images, the canny operator is used for preprocessing; 2) Design a feature fusion network FusionNet to fuse the two types of images into one image; 3) Train FusionNet based on the high-altitude cable protective cover defect dataset; 4) Use transfer learning to extract the trained FusionNet encoding part separately; 5) Place the FusionNet encoding part in the YOLOv8 head to realize the network's fusion function for the two types of images.

[0034] In step S5, the trained FusionNet is integrated into YOLOv8 through transfer learning, and targeted improvements are made to the defects of the protective cover to generate the HL-YOLO model.

[0035] The HL-YOLO model is based on YOLOv8 and is improved in the following ways: FusionNet with good training In order to enable the model to have the ability to fuse two types of images, the trained FusionNet is used for transfer learning to make a Fusion module and placed at the head of the YOLOv8 backbone network to realize an end-to-end fusion detection network.

[0036] Fusing Deformable Convolutions The second layer of the three-layer shunt in the C2f module is fused with deformable convolution to make the model receptive field closer to the defect characteristics of the cable protective cover. The formula of deformable convolution is: Among them, w is the weight matrix, x is the feature matrix covered by the convolution kernel, p0 is the tensor position passed by the center of the convolution kernel, and p n For each offset of p0 within the convolution kernel range, Δp n Then an additional learnable offset is added to each offset of the convolution kernel.

[0037] Optimizing CIoU loss function On the basis of the original CIoU penalty term, the EIoU penalty for the length and width values ​​is introduced, and an adjustable parameter α is added to adjust the proportion of each penalty term. The formula is: Among them, ρ represents the Euclidean distance calculation formula, b, w and h represent the information of the center point, width and height of the prediction box respectively, b gt 、w gt and h gt They represent the center point, width and height of the target box respectively, c represents the diagonal line containing the minimum area of ​​the predicted box and the real box, and c w and c h They represent the width and height of the minimum bounding box covering the predicted box and the true box, respectively. α is an adjustable parameter used to manually adjust the effect of the penalty term.

[0038] Step S6, training the HL-YOLO model based on the high-altitude cable protective cover defect dataset.

[0039] The improved HL-YOLO model was trained based on the high-altitude cable protective cover defect dataset; the experimental training round epochs was set to 300, the batch-size of the feeding batch file was set to 4, the maximum number of working cores workers was set to 0, the initial learning rate Ir0 was set to 0.01, and the learning momentum momentum of the cosine annealing algorithm used to adjust the learning rate was set to 0.937. Figure 5 It is the backbone network diagram of the present invention. After the training is completed, the obtained recognition model weight file is saved, and the model performance is evaluated using the test set. The final output of the network is the location box of the high-altitude cable protective cover defect and the probability of belonging to a specific category.

[0040] Step S7, using Tkinter to create a human-computer interaction interface to perform real-time detection of the high-altitude cable protective cover.

[0041] like Figure 6As shown in the figure, the human-computer interaction interface consists of two parts. The left side is the real-time monitoring screen, which is used to display the current cable monitoring screen in real time; the right side is the function area. If there are multiple devices, they can be selected here by ID number. At the same time, the top area can also display real-time information and status, allowing personnel to monitor the equipment in real time. If there is any damage that causes the equipment to be offline, it can also be inferred from this.

[0042] This embodiment also relates to a high-altitude cable protective cover defect detection device, such as Figure 7 The device includes a protective housing 1, a roller 2, a drive motor 3, a camera module 4, a control unit 5, and an LED light strip 6; the camera module 4 is installed in the middle part of the inner side of 1; the drive motor 3 and the roller 2 are driven by a gear belt, which is installed at both ends of 1 to provide power for the entire device; the camera module 4 collects data and transmits it to the control unit 5, and the control unit 5 transmits the data to the terminal device through wireless communication, and the terminal device is equipped with a human-machine interaction interface such as Figure 6 As shown in the figure, the script file contains a pre-trained HL-YOLO model, which can detect the captured images in real time and save the final detection results in the text box of the human-computer interaction interface to realize the recording function.

[0043] The defect detection camera module is composed of an industrial camera and a thermal imaging module. Figure 2 As shown in the figure, the camera uses a CMOS chip with a resolution of 1280*720, supports MJPGE format transmission, has high detection accuracy, and meets industrial needs. The drive motor is a DC motor, which achieves torque amplification through an external reducer, and can be both light and practical. The terminal configuration is as follows: the operating system is Windows 10, the graphics card is NVIDIA GeForce RTX 3060 and above, and the processor is the 11th Gen Intel (R) Core (TM) i5-11400H @ 2.70GHz 2.69GHz hardware environment.

[0044] Those skilled in the art will readily appreciate other embodiments of the present application after considering the specification and practicing the embodiments disclosed herein. The present application is intended to cover any variations, uses or adaptations of the present application, which follow the general principles of the present application and include common knowledge or customary technical means in the art that are not disclosed in the present application.

[0045] It should be understood that the present application is not limited to the precise structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.

Claims

1. A method for detecting defects in a high-altitude cable protective cover, characterized in that: The following steps are involved: Step 1, collecting images of the surface of the high-altitude cable protective cover, the images include RGB images and infrared thermal images; Step 2: align the RGB image and the infrared thermal map; Step 3, mark the defect category and position of the image of the surface of the high-altitude cable protective cover, generate a high-altitude cable protective cover surface defect dataset, and preprocess the dataset; Step 4: Based on the surface defect dataset of high-altitude cable protective covers, the RGB image and the infrared thermal map are fused to generate the corresponding fusion weights; Step 5, construct the HL-YOLO model based on the fused RGB image and infrared thermal map; Step 6, training the HL-YOLO model based on the surface defect dataset of high-altitude cable protective cover; Step 7: Use the trained HL-YOLO model to perform real-time detection of high-altitude cable protective covers.

2. A method for detecting defects in a high-altitude cable protective cover according to claim 1, characterized in that: In the step 2, after the RGB image and the infrared thermal map are registered and aligned, the data is also enhanced, and the specific operations are: brightness adjustment, random image flipping and grayscale conversion.

3. A method for detecting defects in a high-altitude cable protective cover according to claim 1, characterized in that: In step 5, the HL-YOLO model is constructed by adding the following algorithms to the YOLOv8 algorithm: a) Using transfer learning, HL-YOLO has the ability to fuse RGB images and infrared thermal maps; b) Add deformable convolution; c) Introducing the MobileNet-V3 lightweight neural network; d) Optimize the CIoU loss function.

4. A method for detecting defects in a high-altitude cable protective cover according to claim 3, characterized in that: The use of transfer learning enables HL-YOLO to have the ability to fuse RGB images and infrared heat maps, specifically: Step 301, using the canny operator to perform preprocessing on the edge blur problem of the infrared thermal image; Step 302, designing a feature fusion network FusionNet to fuse the two types of images into one image; Step 303, training FusionNet based on the high-altitude cable protective cover defect dataset; Step 304, using transfer learning to extract the trained FusionNet encoding part separately; Step 305, placing the FusionNet encoding part in the YOLOv8 header to implement the network's fusion function for the two types of images.

5. A method for detecting defects in a high-altitude cable protective cover according to claim 3, characterized in that: The adding of deformable convolution is specifically as follows: Step 401, the second layer of the three-layer shunt in the C2f module is fused with a deformable convolution to make the model receptive field closer to the defect characteristics of the cable protective cover, wherein the formula of the deformable convolution is: Among them, w is the weight matrix, x is the feature matrix covered by the convolution kernel, p0 is the tensor position passed by the center of the convolution kernel, and p n For each offset of p0 within the convolution kernel range, Δp n Then the additional learnable offset added to each offset of the convolution kernel is Step 402: Replace the original activation function in C2f with the H-Swish activation function, whose formula is: Among them, relu represents the ReLU function; Step S403, replacing the second and fourth layers in the backbone network of the HL-YOLO model with the improved C2f module, so that the model attention distribution is more consistent with the characteristics of the slender defect of the cable protective sheath.

6. A method for detecting defects in a high-altitude cable protective cover according to claim 3, characterized in that: The optimized CIoU loss function is: On the basis of the original CIoU penalty term, the EIoU penalty for the length and width values ​​is introduced, and an adjustable parameter α is added to adjust the proportion of each penalty term. The formula is: Among them, ρ represents the Euclidean distance calculation formula, b, w and h represent the information of the center point, width and height of the prediction box respectively, b gt 、w gt and h gt They represent the center point, width and height of the target box respectively, c represents the diagonal line containing the minimum area of ​​the predicted box and the real box, and c w and c h They represent the width and height of the minimum bounding box covering the predicted box and the true box, respectively. α is an adjustable parameter used to manually adjust the penalty term.

7. A method for detecting defects in a high-altitude cable protective cover according to claim 1, characterized in that: The training process of the HL-YOLO model based on the surface defect dataset of the high-altitude cable protective cover in the step 6 is as follows: the training parameters used include the experimental training rounds epochs, the batch-size of the file fed into the batch, the maximum number of working cores workers, the initial learning rate Ir0 and the learning momentum momentum, wherein the learning momentum is used to adjust the learning rate of the cosine annealing algorithm.

8. The method for detecting defects in a high-altitude cable protective cover according to claim 1 is characterized in that: In the step 7, the trained HL-YOLO model is used to perform real-time detection of the high-altitude cable protective cover under the human-computer interaction interface, and the human-computer interaction interface runs under the Windows system or the Linux system.

9. A high-altitude cable protective cover defect detection device, characterized in that: The device comprises a protective shell (1), a roller (2), a driving motor (3), a camera module (4), a control unit (5) and an LED light strip (6); the camera module (4) is mounted on the middle part of the inner side of the protective shell (1); the driving motor (3) and the roller (2) are driven by a gear belt and are mounted on both ends of the protective shell (1) to provide power for the entire device; After the camera module (4) collects data, it transmits it to the control unit (5). The control unit (5) transmits the data to a terminal device such as a handle or a computer that can be connected to the Internet through wireless communication. The terminal device is equipped with a script file of a human-computer interaction interface. The script file contains a pre-trained HL-YOLO model, which is used to detect the captured image in real time and store the final detection result in a text box of the human-computer interaction interface to realize a recording function.

Images