Disc-shaped suspension insulator steel foot eccentricity detection device, system and method

Abstract

The invention discloses a device, a system and a method for measuring the eccentricity of a steel foot of a disc-shaped suspension insulator. Belongs to the technical field of disc-shaped insulator detection, and comprises the following steps of: shooting by combining a shooting bracket of a three-axis sliding table of a load camera lens and a laser positioner and a sample bracket for fixing and adjusting the horizontal state of an insulator to obtain a front photo of a vertical disc-shaped insulator disc surface; and the contours of the disc surface and the steel foot of the disc-shaped insulator are identified by using a convolutional neural network learning algorithm, the coordinates of the circle center are obtained by fitting, and the eccentricity of the steel foot of the disc-shaped insulator is calculated according to the deviation distance between the two circle center points. The device and the system are easy and convenient to operate, the test result does not depend on subjective judgment of operators, and the device and the system are suitable for quality detection of insulator manufacturing enterprises and network access detection of power grid using units and have important application value for improving the quality grade of the disc-shaped insulators.

Description

Technical Field

[0001] This invention relates to the field of disc insulator testing technology, specifically to a device, system, and method for detecting the eccentricity of the steel foot of a disc suspension insulator. Background Technology

[0002] Disc suspension insulators are important insulating devices used in power transmission lines to suspend conductors. A disc suspension insulator consists of two parts: a disc-shaped insulator and connecting hardware. The disc-shaped insulator is typically an umbrella-shaped or inverted bowl-shaped structure made of ceramic or glass. A metal structure (steel cap) with a groove at the top and a hollow hemispherical bottom is fixed to the upper end of the ceramic or glass insulator using cement adhesive. A rod-shaped metal structure (steel foot) with enlarged ends is fixed to the lower end of the ceramic or glass insulator using cement adhesive. The enlarged ends of the steel foot can be combined and fixed with the groove on the upper part of the steel cap. Therefore, disc insulators can form insulator strings through the combination of the steel cap and steel foot. In power transmission lines, disc suspension insulators are interconnected using steel caps and steel feet to form insulator strings of a certain length, providing sufficient insulation length for the conductors and fixed ends. The production process of disc insulators involves fixing the steel cap and steel foot to the upper and lower ends of the insulator respectively with cement using a positioning device, followed by cement curing to obtain a single insulator.

[0003] Compared to steel caps, steel feet have a higher degree of structural asymmetry and are directly inserted into the cement-bonded structure. During insulator production, factors such as the amount of cement used, temperature and humidity changes during curing, and auxiliary positioning operations can all affect the position of the steel feet in the disc insulator, causing a certain degree of skewness. If the skewness of the steel feet is too large, it will cause bending deformation and uneven stress distribution in the insulator string, which may accelerate the uneven distribution of the surface electric field or stress concentration in the insulator string, leading to the risk of string breakage during long-term operation. Since the structural regularity of disc insulators is difficult to measure accurately, it is usually detected by visual inspection with the aid of gauges. This detection method lacks quantitative basis and is greatly affected by the subjective factors of the testers. Chinese invention patent CN114935317A discloses a method for detecting the skewness of disc suspension insulators. This technology designs a rotating table and a scanning frame, and uses a laser scanner to map the disc insulator after one rotation to obtain a three-dimensional point cloud on the surface of the insulator. The skewness of the insulator is then calculated by segmenting and fitting the point cloud. This technical solution requires a motion device to rotate the disc insulator one revolution and a laser scanner to scan the sample. The mechanical device and scanning equipment of this solution are complex and expensive. The fitted image obtained by laser scanning is affected by the accuracy of the laser scanner and the accuracy of the fitting algorithm, resulting in low testing efficiency. It is not a direct measurement of the insulator sample. Summary of the Invention

[0004] The purpose of this invention is to address the problems existing in the current technology and solutions for detecting the eccentricity of the steel feet in disc suspension insulators. This invention designs a device for detecting the eccentricity of the steel feet in disc suspension insulators and develops an algorithm system for eccentricity detection based on machine vision, thus solving the problems of reliance on the experience of testers, low accuracy, and low efficiency in the measurement of disc suspension insulators.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A device for detecting the eccentricity of the steel foot of a disc-shaped suspension insulator includes the following steps:

[0007] The shooting bracket is a four-legged or three-legged frame structure. A three-axis slide is fixed on the upper platform of the frame. The moving end of the three-axis slide is fixed to a camera lens and an optional laser positioner through hardware. The relative distance between the camera lens and the laser positioner is preset and fixed. The shooting bracket is also equipped with a height adjustment mechanism for adjusting the horizontal state of the upper platform.

[0008] An insulator support is a movable frame structure with a support platform on the upper part of the frame. The support platform has a circular opening adapted to the steel cap of the insulator and is equipped with a level and a height adjustment mechanism for fixing the insulator and adjusting the level of the disc. The bottom of the insulator support is equipped with sliding wheels.

[0009] Auxiliary facilities include a power module that powers the three-axis slide, camera, and laser positioner; a slide motion controller that controls the displacement of the three-axis slide; a shooting controller that controls the shooting of the camera; and a light-blocking curtain surrounding the shooting bracket and a shadowless light source mounted on the shooting bracket.

[0010] A system for detecting the eccentricity of the steel foot of a disc-type suspension insulator includes:

[0011] The image acquisition module, based on the aforementioned detection device, enables the acquisition of a frontal photograph perpendicular to the surface of the insulator disc;

[0012] The machine vision recognition module is based on the OpenCV open-source library and uses a U-Net convolutional neural network model. The U-Net convolutional neural network model is trained with front photos of insulators of different specifications. The training parameters include: DiceLoss loss function, Adam optimizer, 80 training epochs, and batch size of 8. The U-Net convolutional neural network model can automatically recognize the insulator disc surface and the outline of the steel foot.

[0013] The image processing and calculation module uses Canny edge detection and morphological operations to optimize the contours, fits the circles of the two contours and calculates the distance between the center d and the disk radius R. The eccentricity of the steel foot is obtained by the formula C=d / R×100%.

[0014] As a further technical solution of the present invention: the moving end of the three-axis slide table can achieve smooth sliding in six directions (up, down, left, right, forward, and backward) under the control of a computer or controller.

[0015] As a further technical solution of the present invention: the height adjustment mechanism of the bearing platform of the insulator support is a screw lift, which can monitor and adjust the level of the platform in real time through a level.

[0016] As a further technical solution of the present invention: the training dataset of the machine vision recognition module needs to contain at least 500 front photos of insulators of different specifications and manufacturers, and the dataset is divided into training set, validation set and test set, and multi-fold cross-validation is used to ensure the generalization ability of the model.

[0017] A detection method based on the above system includes the following steps:

[0018] S1: With the steel feet of the disc insulator facing upwards, the steel cap passes through the opening in the insulator support platform and is placed on the platform of the insulator support. Adjust the platform so that the surface of the insulator disc is horizontal.

[0019] S2: Place the bracket for horizontally placing the disc insulator directly below the shooting bracket;

[0020] S3: Adjust the shooting bracket and slide to make the fixed camera lens and laser positioner on the moving end of the slide horizontal, and adjust the position of the moving end to make the laser positioner accurately positioned at the center of the steel foot of the disc insulator below.

[0021] S4: Lower the light-blocking curtain of the shooting bracket, turn on the shadowless light source, turn on the camera lens, and adjust the slide table to move the specified displacement so that the camera lens is vertically aligned with the center of the steel foot of the disc insulator; since the relative position of the laser positioner and the camera lens is preset and fixed, after the laser positioner is aligned with the center of the steel foot of the disc insulator, the preset displacement can be moved through the slide table controller so that the camera lens returns to the original position of the laser positioner, that is, directly facing the center point of the steel foot of the disc insulator below;

[0022] S5: Adjust the height of the slide and the focal length of the camera lens to clearly image the structure of the insulator disc in the lens and take a picture of the insulator disc.

[0023] S6: Import the insulator disc image into the machine vision processing software, use machine vision algorithms to identify the insulator disc and steel foot contour structure, and automatically draw a circular structure that fits the insulator disc contour and steel foot structure; if the circular structure identified by machine vision is significantly different from the insulator disc or steel foot contour, adjust the position or size of the circular structure to make the fitted circular structure perfectly match the insulator disc and steel foot contour.

[0024] S7: Calculate the distance between the points of the two fitted circular structures, denoted as d; the radius of the fitted circle of the insulator disc profile, denoted as R; the eccentricity of the steel foot of the disc insulator is denoted as C=d / R×100%.

[0025] As a further technical solution of the present invention: the machine vision algorithm is specifically an algorithm for recognizing and adjusting the contours of the insulator disc and the steel foot, including the following:

[0026] A1: Data Collection and Preparation: Collect disc insulator samples of various specifications and models from different manufacturers; use testing equipment to photograph the disc insulators; and collect front view photos of the disc insulators.

[0027] A2. Perform size unification and pixel value normalization processing on the collected photos, and set the front view of all disc insulators to a standard image of 256×256 pixels with pixel values ​​in the range of 0-1;

[0028] A3. Divide the processed disc insulator image dataset into training set, validation set and test set, and perform multi-fold cross-validation to establish a machine learning model for recognizing the disc surface contour and steel foot contour.

[0029] A4. Build a deep learning model based on U-Net convolutional neural network, and gradually extract image features through a series of convolutional and pooling layer operations; in the encoder part of the model, each convolutional operation uses the tf.keras.layers.Conv2D function, where the convolutional kernel size is set to 3, the activation function is ReLU to introduce non-linearity, and the padding method is same to keep the feature map size unchanged.

[0030] A5. Model Training: By defining a loss function with Dice Loss as the core, the model is compiled in conjunction with the Adam optimizer, and various key parameters are carefully set to effectively avoid overfitting and continuously optimize model performance.

[0031] A6. Once the model achieves the expected performance on the test set, it is saved.

[0032] A7. After establishing the model, edge detection and morphological operations are introduced as traditional image analysis algorithms to post-process the model's prediction results. In the edge detection stage, the Canny edge detection algorithm is used to extract edges from the segmented image obtained by the model, specifically through the cv2.Canny function in the OpenCV library. The key parameters of this function, the low threshold threshold1, are set to 50, and the high threshold threshold2, are set to 150. After completing the edge detection, morphological operations are used to optimize and refine the edge contours.

[0033] As a further technical solution of the present invention: the core steps of the Canny algorithm are as follows: First, Gaussian filtering is used for noise reduction. A Gaussian filter based on the principle of Gaussian function is used to perform convolution operation on the image to smooth the image and remove noise interference, laying the foundation for accurate edge detection. Next, the Sobel operator is used to calculate the gradient magnitude and direction. The cv2.Sobel function parameter ksize is set to 3 to calculate the gradient of the image in the x and y directions, thereby obtaining the gradient magnitude and direction, and thus determining the areas with drastic gray-level changes in the image, i.e., possible edges. Then, non-maximum suppression is automatically performed inside the cv2.Canny function. This operation compares the gradient magnitude of the current pixel with the gradient magnitude of the adjacent pixels along the gradient direction, and only retains the pixel with the largest gradient magnitude as the edge point, effectively avoiding edge coarsening and achieving edge refinement. Finally, dual threshold detection and edge connection are performed. By setting a low threshold and a high threshold, edge points that meet the threshold conditions are connected to form a complete edge contour, ensuring that the edge detection result does not miss important edges, nor does it contain too many false edges caused by noise.

[0034] As a further technical solution of the present invention: morphological operations are used to optimize and refine the edge contours; the specific operations are as follows: The cv2.erode function in OpenCV is used for erosion. A structuring element is slid across the image; if all pixels within the covered area are foreground pixels, the center pixel is retained; otherwise, it is set as a background pixel, thus thinning the edge contours and removing isolated noise points and small burrs. The cv2.dilate function is used for dilation, the opposite of erosion. If any pixel within the structuring element's covered area is a foreground pixel, the center pixel is set as a foreground pixel, thus thickening the edge contours and filling in the gaps caused by erosion. The small gaps created by erosion restore edge continuity. The cv2.morphologyEx function is used to perform opening and closing operations. The opening operation first erodes and then dilates, using erosion to remove noise points and small interfering objects from the image, and then using dilation to restore some of the eroded edges, achieving the effect of removing noise while preserving the main edges. The closing operation first dilates and then erodes, using dilation to fill holes in the image, and then using erosion to remove the excess edges caused by dilation, thus connecting the broken edges. By combining morphological operations, the contours of the insulator disk and steel feet predicted by the model can be optimized and refined, improving the accuracy and clarity of contour recognition.

[0035] The present invention has the following advantages and beneficial effects:

[0036] 1. Simple operation and objective results: No manual visual inspection or complex mechanical rotation is required. The detection is completed through automated devices and algorithms, avoiding the influence of subjective experience of personnel and ensuring good repeatability of results;

[0037] 2. High accuracy and high efficiency: Based on U-Net network and post-processing algorithm, the contour recognition accuracy is ≥98%, and the single sample detection time is ≤5 minutes (far faster than laser scanning solution).

[0038] 3. Controllable equipment cost: No expensive laser scanner is required. The core components are a three-axis slide, camera and computer. The structure is simple and easy to mass-produce and promote.

[0039] 4. Wide range of applications: It can test disc insulators of different specifications made of ceramic and glass materials, meeting the quality inspection needs of manufacturers and grid connection inspection, which is of great significance for improving the quality level of insulators. Attached Figure Description

[0040] Figure 1 This is a schematic diagram of a device for detecting the eccentricity of the steel foot of a disc-type suspension insulator;

[0041] Figure 2 This is a schematic diagram of the insulator sample holder.

[0042] Figure 3 This is a photograph of a ceramic disc suspension insulator taken by the detection device in Example 1;

[0043] Figure 4 This is a photograph of a glass disc-shaped suspension insulator taken by the detection device in Example 2.

[0044] In the diagram: 1-Shooting bracket, 11-Three-axis slide, 12-Image acquisition device, 121-Camera lens, 122-Fittings, 123-Laser positioner, 13-Shadowless lamp, 14-Height adjustment mechanism, 2-Insulator sample holder, 21-Insulator placement platform, 211-Level, 212-Hole, 213-Adjusting bracket plane, 22-Insulator sample holder metal leg, 221-Sample holder sliding wheel, 3-Operating computer. Detailed Implementation

[0045] The present invention will be further described below with reference to the embodiments. It should be noted that these are merely examples and descriptions of the inventive concept. Those skilled in the art can make various modifications or additions to the specific embodiments described or use similar methods to replace them, as long as they do not deviate from the inventive concept or exceed the scope defined in the claims, they should all be considered to fall within the protection scope of the present invention.

[0046] like Figures 1-4 As shown, this invention discloses a device for detecting the eccentricity of the steel foot of a disc-shaped suspension insulator; comprising:

[0047] The shooting bracket 1 includes a three-axis slide 11 fixed to the top of the shooting bracket; an image acquisition device 12 is fixed on the three-axis slide 11, the image acquisition device 12 includes a camera lens 121 and a laser positioner 123; the camera lens 121 and the laser positioner 123 are fixed to the moving end of the slide 11 by a fitting 122; the displacement of the moving end of the three-axis slide 11 can be controlled by an external controller or an operating computer 3, so that the camera lens can be programmed to move to the position indicated by the laser positioner; it also includes a shadowless lamp 13 mounted on the shooting bracket and a height adjustment mechanism 14 for adjusting the horizontal state of the shooting bracket. A light-blocking curtain can be placed outside the shooting bracket to prevent external light sources from interfering with the clarity of the camera lens image. To avoid confusion, the power supplies and wires of the three-axis slide, lens, shadowless lamp, and laser positioner are not shown in the schematic diagram.

[0048] The insulator sample holder 2 includes an insulator placement platform 21, a level 211 on the placement platform 21, a hole 212 for supporting the insulator disc, a height adjustment mechanism 213 for adjusting the horizontal state of the holder plane 21, metal legs 22 of the insulator sample holder, and sample holder sliding wheels 221. The height of the metal legs 22 can prevent the steel cap of the disc insulator from contacting the ground or operating table where the insulator sample holder is placed.

[0049] Example:

[0050] Step 1: Place the ceramic disc insulator, manufactured by a certain manufacturer, with its steel feet facing upwards and the steel cap passing through the opening in the insulator support platform, on the insulator support. Use a level to help adjust the disc surface of the disc insulator to a horizontal state.

[0051] Step 2: Move the bracket holding the horizontally placed insulator directly below the shooting bracket.

[0052] Step 3: Adjust the shooting bracket and slide table to ensure the camera lens and laser positioner fixed on the moving end of the slide table are horizontal. Adjust the position of the moving end to accurately position the laser positioner at the center of the steel foot of the disc insulator below. Confirm that the horizontal distance between the center point of the camera lens and the positioning point of the laser positioner is Lx. That is, moving the slide table by Lx in the x-direction will move the camera lens to the initial position of the laser positioning point.

[0053] Step 4: Lower the light-blocking curtain of the shooting stand, turn on the shadowless light source, turn on the camera lens, and adjust the slide to move the specified displacement Lx so that the camera lens is vertically aligned with the center of the steel foot of the disc insulator.

[0054] Step 5: Adjust the height of the slide table and the focal length of the camera lens to clearly image the structure of the insulator disc in the lens and take a picture of the insulator disc.

[0055] Step Six: Import the insulator disc image into the machine vision model to identify the insulator disc and steel foot contours, and automatically draw a circular structure that fits the insulator disc and steel foot contours. If the circular structure identified by machine vision differs significantly from the insulator disc or steel foot contours, manually adjust the position or size of the circular structure to ensure a perfect fit. The fitted circle and the calculated center image are shown below. Figure 3 As shown.

[0056] Step 7: Calculation Figure 3 The distance between the two fitted circular points is d = 25 pixels, and the radius of the fitted circle of the insulator disc profile is R = 5000 pixels. Based on the eccentricity calculation formula C = d / R × 100%, the eccentricity of the steel foot of this disc insulator is calculated to be 0.5%.

[0057] Example 2:

[0058] A glass disc insulator manufactured by a certain manufacturer was placed on an insulator support according to the method described in Example 1, and a frontal photograph was taken. The photograph was input into a machine vision model to identify the outline of the insulator disc and the outline of the steel foot, and fitted into a circle, and the center point of the circle was calculated. Figure 4 As shown. Calculation Figure 4 The distance between the two fitted circular points is d = 40 pixels, and the radius of the fitted circle of the insulator disc profile is R = 5000 pixels. Based on the eccentricity calculation formula C = d / R × 100%, the eccentricity of the steel foot of this glass disc insulator is calculated to be 0.8%.

[0059] Figure 3 This is a photograph of a ceramic disc suspension insulator taken by the detection device in Example 1. In the figure, the blue ring is the fitted circle of the ceramic insulator disc surface and steel foot outline identified by the machine vision algorithm, and the red dot is the calculated center position.

[0060] Figure 3 Example 2: Fitted circles and dots for detecting the eccentricity of the steel feet of a glass disc suspension insulator.

[0061] Figure 3 This is a photograph of a glass disc-shaped suspension insulator taken by the detection device in Example 2. In the figure, the blue ring represents the fitted circle of the ceramic insulator disc surface and steel foot outline identified by the machine vision algorithm, and the red dot is the calculated center position.

[0062] The above is an exemplary description of the invention. Obviously, the specific implementation of the invention is not limited to the above-described manner. Any non-substantial improvement made using the inventive concept and technical solution of the invention, or the direct application of the inventive concept and technical solution to other situations without modification, is within the protection scope of the invention.

Claims

1. A device for detecting the eccentricity of the steel foot of a disc-shaped suspension insulator, characterized in that, Include: The shooting bracket is a four-legged or three-legged frame structure. A three-axis slide is fixed on the upper platform of the frame. The moving end of the three-axis slide is fixed to a camera lens and an optional laser positioner through hardware. The relative distance between the camera lens and the laser positioner is preset and fixed. The shooting bracket is also equipped with a height adjustment mechanism for adjusting the horizontal state of the upper platform. An insulator support is a movable frame structure with a support platform on the upper part of the frame. The support platform has a circular opening adapted to the steel cap of the insulator and is equipped with a level and a height adjustment mechanism for fixing the insulator and adjusting the level of the disc. The bottom of the insulator support is equipped with sliding wheels. Auxiliary facilities include a power module that powers the three-axis slide, camera, and laser positioner; a slide motion controller that controls the displacement of the three-axis slide; a shooting controller that controls the shooting of the camera; and a light-blocking curtain surrounding the shooting bracket and a shadowless light source mounted on the shooting bracket.

2. A system for detecting the eccentricity of the steel foot of a disc-type suspension insulator, characterized in that, include: The image acquisition module, based on the detection device described in claim 1, enables the acquisition of a frontal photograph perpendicular to the surface of the insulator disc; The machine vision recognition module is based on the OpenCV open-source library to build a U-Net convolutional neural network model. The U-Net convolutional neural network model is trained with front photos of insulators of different specifications. The U-Net convolutional neural network model can automatically recognize the insulator disc surface and the outline of the steel foot. The image processing and calculation module uses Canny edge detection and morphological operations to optimize the contours, fits the circles of the two contours and calculates the distance between the center d and the disk radius R. The eccentricity of the steel foot is obtained by the formula C=d / R×100%.

3. The system according to claim 2, characterized in that, The moving end of the three-axis slide table can achieve smooth sliding in six directions (up, down, left, right, forward, and backward) under the control of a computer or controller.

4. The system according to claim 3, characterized in that, The height adjustment mechanism of the insulator support platform is a screw jack, which can monitor and adjust the platform's level in real time using a level.

5. The system according to claim 4, characterized in that, The training dataset for the machine vision recognition module must contain at least 500 front-facing photos of insulators of different specifications and manufacturers. The dataset is divided into training, validation, and test sets, and multi-fold cross-validation is used to ensure the model's generalization ability.

6. A detection method based on the system according to any one of claims 2-5, characterized in that, Includes the following steps: S1: With the steel feet of the disc insulator facing upwards, the steel cap passes through the opening in the insulator support platform and is placed on the platform of the insulator support. Adjust the platform so that the surface of the insulator disc is horizontal. S2: Place the bracket for horizontally placing the disc insulator directly below the shooting bracket; S3: Adjust the shooting bracket and slide to make the fixed camera lens and laser positioner on the moving end of the slide horizontal, and adjust the position of the moving end to make the laser positioner accurately positioned at the center of the steel foot of the disc insulator below. S4: Lower the light-blocking curtain of the shooting bracket, turn on the shadowless light source, turn on the camera lens, and adjust the slide table to move the specified displacement so that the camera lens is vertically aligned with the center of the steel foot of the disc insulator; since the relative position of the laser positioner and the camera lens is preset and fixed, after the laser positioner is aligned with the center of the steel foot of the disc insulator, move the preset displacement through the slide table controller so that the camera lens returns to the original position of the laser positioner, that is, directly facing the center point of the steel foot of the disc insulator below; S5: Adjust the height of the slide and the focal length of the camera lens to clearly image the structure of the insulator disc in the lens and take a picture of the insulator disc. S6: Import the insulator disc image into the machine vision processing software, use machine vision algorithms to identify the insulator disc and steel foot contour structure, and automatically draw a circular structure that fits the insulator disc contour and steel foot structure; if the circular structure identified by machine vision is significantly different from the insulator disc or steel foot contour, adjust the position or size of the circular structure to make the fitted circular structure perfectly match the insulator disc and steel foot contour. S7: Calculate the distance between the points of the two fitted circular structures, denoted as d; the radius of the fitted circle of the insulator disc profile, denoted as R; the eccentricity of the steel foot of the disc insulator is denoted as C=d / R×100%.

7. The method according to claim 6, characterized in that, The machine vision algorithm is specifically an algorithm for recognizing and adjusting the contours of the insulator disc and steel feet, and includes the following: A1: Data Collection and Preparation: Collect disc insulator samples of various specifications and models from different manufacturers; use testing equipment to photograph the disc insulators; and collect front view photos of the disc insulators. A2. Perform size unification and pixel value normalization processing on the collected photos, and set the front view of all disc insulators to a standard image of 256×256 pixels with pixel values ​​in the range of 0-1; A3. Divide the processed disc insulator image dataset into training set, validation set and test set, and perform multi-fold cross-validation to establish a machine learning model for recognizing the disc surface contour and steel foot contour. A4. Build a deep learning model based on U-Net convolutional neural network, and gradually extract image features through a series of convolutional and pooling layer operations; in the encoder part of the model, each convolutional operation uses the tf.keras.layers.Conv2D function, where the convolutional kernel size is set to 3, the activation function is ReLU to introduce non-linearity, and the padding method is same to keep the feature map size unchanged. A5. Model Training: By defining a loss function with Dice Loss as the core, the model is compiled in conjunction with the Adam optimizer, and various key parameters are carefully set to effectively avoid overfitting and continuously optimize model performance. A6. Once the model achieves the expected performance on the test set, it is saved. A7. After establishing the model, edge detection and morphological operations are introduced as traditional image analysis algorithms to post-process the model's prediction results. In the edge detection stage, the Canny edge detection algorithm is used to extract edges from the segmented image obtained by the model, specifically through the cv2.Canny function in the OpenCV library. The key parameters of this function, the low threshold threshold1, are set to 50, and the high threshold threshold2, are set to 150. After completing the edge detection, morphological operations are used to optimize and refine the edge contours.

8. The method according to claim 7, characterized in that, The core steps of the Canny algorithm are as follows: First, Gaussian filtering is used for noise reduction. A Gaussian filter based on the Gaussian function principle is used to perform convolution operations on the image to smooth the image and remove noise interference, laying the foundation for accurate edge detection. Next, the Sobel operator is used to calculate the gradient magnitude and direction. The cv2.Sobel function parameter ksize is set to 3 to calculate the gradient of the image in the x and y directions, thereby obtaining the gradient magnitude and direction to determine the areas with drastic gray-level changes in the image, i.e., possible edges. Then, non-maximum suppression is automatically performed inside the cv2.Canny function. This operation compares the gradient magnitude of the current pixel with the gradient magnitude of the adjacent pixels along the gradient direction, and only retains the pixel with the largest gradient magnitude as the edge point, effectively avoiding edge coarsening and achieving edge refinement. Finally, dual threshold detection and edge connection are performed. By setting a low threshold and a high threshold, edge points that meet the threshold conditions are connected to form a complete edge contour, ensuring that the edge detection results neither miss important edges nor contain too many false edges caused by noise.

9. The method according to claim 7, characterized in that, Morphological operations are used to optimize and refine the edge contours. The specific operations are as follows: The `cv2.erode` function in OpenCV is used for erosion. A structuring element slides across the image; if all pixels within the covered area are foreground pixels, the center pixel is retained; otherwise, it is set as a background pixel. This thins the edge contours and removes isolated noise points and small burrs. The `cv2.dilate` function performs dilation, the opposite of erosion. If any pixel within the structuring element's covered area is a foreground pixel, the center pixel is set as a foreground pixel. This thickens the edge contours and fills in the small gaps created by erosion. The morphological model restores edge continuity by using the cv2.morphologyEx function to perform opening and closing operations. The opening operation first erodes and then dilates, using erosion to remove noise points and small interfering objects from the image, and then using dilation to restore some of the eroded edges, achieving the effect of removing noise while preserving the main body edges. The closing operation first dilates and then erodes, using dilation to fill holes in the image, and then using erosion to remove the excess edges caused by dilation, thereby connecting broken edges. By combining morphological operations, the model-predicted contours of the insulator disk and steel feet can be optimized and refined, improving the accuracy and clarity of contour recognition.

Images