A numerical control machining platform remote monitoring method for a high-voltage switch shell

Abstract

The present application relates to the technical field of image data processing, and particularly relates to a remote monitoring method for numerical control machining platform of high-voltage switch shell, comprising: acquiring a machining scene image of the numerical control machining platform containing the high-voltage switch shell; segmenting the machining scene image into a plurality of sub-regions, and determining the disorder degree of each sub-region; determining the attention degree of each sub-region; determining the thermal conductivity coefficient of each sub-region; based on the thermal conductivity coefficient, acquiring a filtered image of the machining scene image by using an anisotropic filtering algorithm; based on the filtered image, using target detection technology to detect defects of the high-voltage switch shell, so as to realize remote monitoring of the numerical control machining platform of the high-voltage switch shell. The present application overcomes the fixed parameter defect of the traditional anisotropic filtering algorithm, realizes dynamic balance between noise suppression and feature preservation, and significantly improves the accuracy and reliability of the high-voltage switch shell defect detection.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image data processing, and particularly relates to a remote monitoring method for numerical control machining platform of high-voltage switch shell. BACKGROUND

[0002] As the core project of power energy cross-regional transmission, the safe and stable operation of high-voltage transmission line is directly related to the overall reliability of the power system. Among them, the quality performance of high-voltage switch shell as the key equipment component of transmission line plays a decisive role in the operation safety of the whole line. The production and manufacturing of high-voltage switch shell is mainly realized by precision machining of thick-walled aluminum pipe through special machining platform, but if the platform has mechanical parameter abnormality, tool wear and other conditions during machining, it is easy to cause cracks, depressions and other machining defects of the shell, which may further cause serious safety hazards such as gas leakage in subsequent use. In addition, factors such as vibration, unstable clamping and uneven internal organization of materials during machining will also aggravate the machining error, affecting the sealing and structural strength of the shell, therefore, it is necessary to strictly control the process parameters of each link and improve the real-time and accuracy of the detection and feedback mechanism to ensure that the product meets the high standard quality requirements.

[0003] In the current industrial scene, the processing state of high-voltage switch shell is usually dynamically monitored by collecting real-time images of the machining platform. However, factors such as mechanical vibration, electromagnetic interference and light fluctuation in the industrial production environment will inevitably mix noise into the collected images. In order to eliminate the influence of noise on monitoring accuracy, an anisotropic filtering algorithm can be used to smooth the image, but the traditional anisotropic filtering algorithm will cause excessive blurring of the key defect features such as fine cracks on the surface of the shell while suppressing noise, affecting the accurate identification of processing abnormalities and reducing the timeliness and accuracy of remote monitoring of the high-voltage switch shell machining process. SUMMARY

[0004] In order to solve the problem that due to the interference of industrial production environment, there is certain noise in the collected images of the machining platform, and in the process of image smoothing by using an anisotropic filtering algorithm, the key defect features such as fine cracks on the surface of the high-voltage switch shell are also blurred while suppressing noise, which affects the accurate identification of processing abnormalities and reduces the timeliness and accuracy of remote monitoring of the high-voltage switch shell machining process, the present application proposes a remote monitoring method for numerical control machining platform of high-voltage switch shell, which comprises the following steps:

[0005] An image of a machining scene of a numerical control machining platform containing a high-voltage switch shell is acquired; a segmentation algorithm is used to segment the image of the machining scene into a plurality of sub-regions, any sub-region is recorded as a target sub-region, the degree of disorder of the target sub-region is determined according to the number of corner points in the target sub-region, the maximum and minimum gray scales of the pixel points in the target sub-region; the attention degree of the target sub-region is determined according to the degree of disorder, the mean value of the gray scales of the pixel points in the target sub-region, and the mean value of the degrees of disorder of all adjacent sub-regions of the target sub-region; the heat conduction coefficient of the target sub-region is determined according to the size of the attention degree and a preset attention degree threshold; based on the heat conduction coefficient, an anisotropic filtering algorithm is used to acquire a filtered image of the image of the machining scene; based on the filtered image, a target detection technology is used to detect defects of the high-voltage switch shell, so as to realize remote monitoring of the numerical control machining platform of the high-voltage switch shell.

[0006] The machining scene image is divided into a plurality of sub-regions by the segmentation algorithm, and the degree of disorder of each sub-region is quantified based on the number of corner points and the gray scale difference, so that fine analysis of local features of the image is realized; by introducing the attention degree mechanism, the context information of the sub-region itself and the adjacent region is comprehensively considered, so that important regions containing key defect features can be accurately identified; the heat conduction coefficient of each sub-region is dynamically determined by comparing the attention degree with the preset threshold, so that adaptive adjustment of the anisotropic filtering parameter is realized; a smaller heat conduction coefficient is used in a high-attention-degree region with dense defect features, so that key information such as fine cracks is protected from being excessively smoothed, and a larger heat conduction coefficient is used in a low-attention-degree region with a background or more noise, so that noise interference is effectively suppressed; the present application overcomes the disadvantages of the fixed parameters of the traditional anisotropic filtering algorithm, realizes dynamic balance between noise suppression and feature preservation, significantly improves the accuracy and reliability of the defect detection of the high-voltage switch shell, and the target detection technology based on the filtered image can more accurately identify machining abnormalities, thereby improving the timeliness and accuracy of remote monitoring.

[0007] Further, the image of the machining scene is an image processed by gray scale processing.

[0008] Further, the segmentation algorithm uses a superpixel segmentation algorithm.

[0009] Further, the degree of disorder satisfies:

[0010] In the formula, is the degree of disorder of the i th sub-region, is the number of corner points in the i th sub-region, and are the maximum and minimum gray scales of the pixel points in the i th sub-region, respectively, and are the maximum and minimum gray scales of the pixel points in the i th sub-region, respectively, The maximum and minimum gray levels of pixels within each sub-region It is a linear normalization function.

[0011] This invention constructs a quantitative index of clutter by linearly normalizing the product of the number of corner points and the gray level difference within a sub-region, thereby comprehensively reflecting the texture complexity and gray level variation characteristics of the sub-region. The number of corner points reflects the structural complexity within the region, while the gray level difference reflects the contrast variation of the image. The product of the two can effectively characterize the visual complexity of the sub-region. The linear normalization process eliminates the influence of different sub-region sizes and lighting conditions on the calculation results, making the clutter levels of each sub-region comparable.

[0012] Furthermore, the corner points are obtained using the FAST corner detection algorithm.

[0013] Furthermore, the level of attention satisfies:

[0014] In the formula, For the first The level of attention given to each sub-region For the first The degree of clutter in each sub-area For the first The average grayscale value of pixels within each sub-region To process the number of sub-regions in a scene image, For the first The mean of the clutter level of all adjacent subregions of a given subregion. It is a function for maximizing the value.

[0015] This invention constructs a comprehensive attention evaluation model by combining the exponential relationship between the average gray level of a sub-region and the average clutter level of adjacent regions. When the clutter level of the sub-region itself is low and the clutter level of the adjacent regions is high, the exponential term approaches 1, indicating that the region may be an important structural feature region and should be given high attention. Conversely, when the clutter level of the sub-region itself is high or the clutter level of the adjacent regions is low, the exponential term decreases, indicating that the region may be a noise or background region, and the attention should be reduced accordingly. This effectively distinguishes between defect feature regions and noise background regions, providing a basis for the subsequent adaptive adjustment of the thermal conductivity coefficient.

[0016] Furthermore, the thermal conductivity coefficient satisfies:

[0017] In the formula, For the first The thermal conductivity coefficient of each sub-region For the first The level of attention given to each sub-region For the first In all adjacent sub-regions of the first sub-region, the second sub-region is related to the third sub-region. The attention level of the adjacent sub-region with the smallest difference in attention among sub-regions. For the first The average gradient value of pixels within each sub-region For the first In all adjacent sub-regions of the first sub-region, the second sub-region is related to the third sub-region. The mean gradient value of pixels in adjacent sub-regions with the smallest difference in attention among sub-regions. This is the preset attention threshold.

[0018] This invention constructs a heat conduction coefficient using a piecewise function, enabling intelligent adaptive adjustment of the filtering intensity. When the attention level of a sub-region exceeds a preset threshold, a weighted fusion of gradient information and attention level is comprehensively considered. Weights are allocated proportionally based on gradient values ​​to ensure effective noise suppression while protecting important defect features. When the attention level is below the threshold, a simplified averaging calculation method is used, balancing computational efficiency and filtering effect. It fully utilizes the crucial role of gradient information in edge preservation, dynamically adjusting the filtering intensity based on the ratio of its own gradient to the gradient of adjacent regions. This effectively maintains the clarity of defect edges while smoothing noisy areas. The selection mechanism of the adjacent region with the smallest attention difference ensures the spatial continuity of filtering parameters, avoiding image distortion caused by parameter abrupt changes. This enhances the anisotropic filtering algorithm's ability to protect against subtle defects in high-voltage switch housings, providing a high-quality image foundation for subsequent defect detection.

[0019] Furthermore, obtaining the filtered image of the processing scene image includes: based on the thermal conductivity coefficient of each sub-region in the processing scene image, using an anisotropic filtering algorithm to filter each sub-region in the processing scene image to obtain the filtered image of the processing scene image.

[0020] Furthermore, the target detection technology employs the YOLO algorithm.

[0021] Furthermore, the remote monitoring of the CNC machining platform for the high-voltage switch housing includes: in response to a defect on the high-voltage switch housing, pushing the specific information of the defect to the remote monitoring terminal of the CNC machining platform and issuing an early warning to handle the defect and complete the remote monitoring of the CNC machining platform for the high-voltage switch housing.

[0022] This invention achieves intelligent remote monitoring of the high-voltage switch housing processing by organically combining real-time defect detection with a remote alarm push mechanism. When a defect is detected, the system can immediately trigger an alarm and push specific defect information to the remote monitoring terminal, ensuring that operators can obtain abnormal situations in a timely manner and improving the timeliness of defect response and processing efficiency.

[0023] The present application has the following beneficial effects:

[0024] (1) Break through the limitations of traditional anisotropic filtering fixed parameters, quantify the region complexity through the sub-region disorder degree (corner point number, gray level difference), mark the high-disorder region containing key defects such as fine cracks as a high-attention region, dynamically reduce the heat conduction coefficient for the high-attention region, reduce the filtering smoothing strength, avoid the defect features from being excessively blurred, increase the heat conduction coefficient for the noise-intensive region with low attention, and strengthen the denoising effect. The differential filtering strategy can balance the contradiction between denoising and feature preservation, significantly improve the recognition sensitivity of defects such as fine cracks on the surface of the high-voltage switch shell, and solve the problem of abnormal missed detection caused by excessive smoothing in the traditional method.

[0025] (2) Fully consider the image noise characteristics caused by factors such as mechanical vibration, electromagnetic interference, and light fluctuation in industrial scenes, dynamically adjust the attention degree and heat conduction coefficient of each region through sub-region adjacency relationship and gray mean value analysis, which can specifically suppress noise interference while preserving key features such as edges and textures in the defect region, and improve the quality stability of the filtered image, providing a reliable data foundation for subsequent defect detection.

[0026] (3) Through sub-region fine analysis by segmentation algorithm, combined with target detection technology to form a closed loop from filtering to detection, the defect features in the high-attention region are clearly preserved, so that the defect detection model can more quickly and accurately identify processing abnormalities, shorten the response time from image acquisition to abnormal warning, improve the timeliness of remote monitoring, help operation and maintenance personnel to intervene and handle in the first time, reduce the scrap rate caused by defect expansion, and ensure the processing quality stability of the high-voltage switch shell.

[0027] (4) Without manually presetting fixed filtering parameters, the heat conduction coefficient is adaptively optimized through automatic calculation of the sub-region disorder degree and attention degree, so that the algorithm can adapt to the image feature changes of different processing stages (such as rough machining and finishing) and different defect types (such as cracks and depressions), reducing the dependence on human experience and reducing the parameter debugging cost, while improving the practicality of the remote monitoring system under complex working conditions. BRIEF DESCRIPTION OF DRAWINGS

[0028] Figure 1 is a step flow chart of a remote monitoring method of a numerical control machining platform of a high-voltage switch shell according to an embodiment of the present application.

[0029] Figure 2 is a schematic diagram of a machining scene image of a remote monitoring method of a numerical control machining platform of a high-voltage switch shell according to an embodiment of the present application.

[0030] Figure 3It is a schematic view of a processing scene image after superpixel segmentation and corner point detection of a numerical control machining platform remote monitoring method of a high-voltage switch shell according to an embodiment of the present application. DETAILED DESCRIPTION

[0031] The technical solutions in the embodiments of the present application will be clearly and completely described below. The described embodiments are part of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of the present application.

[0032] The specific embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0033] Please refer to Figure 1 , which shows a step flowchart of a numerical control machining platform remote monitoring method of a high-voltage switch shell according to an embodiment of the present application. The method comprises the following steps:

[0034] S1: Obtain a processing scene image of a numerical control machining platform containing a high-voltage switch shell.

[0035] It should be noted that the processing scene image containing the high-voltage switch shell is first collected in real time by an image collection device (such as an industrial camera) deployed in front of the numerical control machining platform. For the convenience of subsequent processing, the collected color image is converted into a gray-scale image.

[0036] Specifically, the processing scene image is a gray-scale processed image.

[0037] S2: Segment the processing scene image into a plurality of sub-regions and determine the clutter degree of each sub-region.

[0038] It should be noted that, as Figure 2 shown, a large number of regions unrelated to the processing of the high-voltage switch shell in the initial processing scene image will interfere with the monitoring of the machining platform. Since the high-voltage switch shell is processed by thick-walled aluminum pipes, the aluminum pipes have a metallic luster and have high gray-scale values in the image, and the industrial production environment is complex and has complex structures in the image. Therefore, in this step, the image is segmented by a segmentation algorithm to obtain a plurality of sub-regions. For example, the image can be segmented into 100 sub-regions. As a preferred solution, the number of sub-regions can be set to 80-150. If the number of sub-regions is too small (for example, less than 80), it may cause a single sub-region to contain both the shell and the background, affecting the accuracy of subsequent analysis. If the number of sub-regions is too large (for example, more than 150), it will significantly increase the computational burden and affect the real-time performance of the monitoring.

[0039] The processing scene image is segmented into a plurality of sub-regions by using a segmentation algorithm.

[0040] Specifically, the segmentation algorithm adopts a superpixel segmentation algorithm.

[0041] It should be noted that, as shown in the figure, the red line area is a sub-region after superpixel segmentation, and the green dot is a corner point. In order to quantify the internal texture complexity and structural change of each sub-region, a degree of disorder is introduced in this step, and the construction basis is that the degree of disorder of a region can be jointly determined by the richness of its internal structure (reflected by the number of corner points) and the degree of change of the gray scale (reflected by the gray scale range). Figure 3

[0042] Any sub-region is recorded as a target sub-region, and the degree of disorder of the target sub-region is determined according to the number of corner points in the target sub-region, the maximum gray scale level and the minimum gray scale level of the pixel points in the target sub-region.

[0043] Specifically, the corner points are obtained by using a FAST corner point detection algorithm.

[0044] Specifically, the degree of disorder satisfies:

[0045] ;

[0046] In the formula, is the degree of disorder of the i-th sub-region, is the number of corner points in the i-th sub-region, and are the maximum gray scale level and the minimum gray scale level of the pixel points in the i-th sub-region, respectively, is a linear normalization function.

[0047] wherein, the greater the value is, the more rich the texture and the more complex the structure in the i-th sub-region are, and the greater the degree of disorder of the i-th sub-region is; the smaller the value is, the more single the texture and the structure in the i-th sub-region are, and the smaller the degree of disorder of the i-th sub-region is. represents the maximum gray scale difference between the pixel points in the i-th sub-region. The greater the value is, the more dramatic the change in the gray scale in the i-th sub-region is, and the greater the degree of disorder of the i-th sub-region is; the smaller the value is, the more gentle the change in the gray scale in the i-th sub-region is, and the smaller the degree of disorder of the i-th sub-region is.

[0048] ​​​​​​​​​​​​​S3: Determine the level of attention for each sub-region.

[0049] It should be noted that during the monitoring of the processing platform, the state changes of the high-voltage switch housing should be the main monitoring content. Pixels in different areas of the image have different characteristics (the image background corresponds to a sub-region with high clutter, the high-voltage switch housing area has low clutter, and the high-voltage switch housing area has high grayscale values). Therefore, this step obtains the attention of a sub-region based on the clutter level of the sub-region and the grayscale values ​​of the pixels within the sub-region. The basis for its construction is that the attention of a sub-region should be related to its brightness (the high-voltage switch housing is processed from aluminum tubes and usually has a high brightness feature), the purity of its own content (low clutter level), and the consistency of its neighborhood.

[0050] The attention level of the target sub-region is determined based on the clutter level, the average grayscale value of the pixels within the target sub-region, and the average clutter level of all adjacent sub-regions (sub-regions directly adjacent to the target sub-region).

[0051] Specifically, the level of attention satisfies:

[0052] ;

[0053] In the formula, For the first The level of attention given to each sub-region For the first The degree of clutter in each sub-area For the first The average grayscale value of pixels within each sub-region To process the number of sub-regions in a scene image, For the first The mean of the clutter level of all adjacent subregions of a given subregion. It is a function for maximizing the value.

[0054] in, The smaller the value, the better. The more complex the content of the processing scene image corresponding to each sub-region, the more complex the first... The more likely a sub-region is to correspond to the background region of the processing platform, the more likely the first sub-region is to be affected. The less attention each sub-region receives; The larger the value, the more likely it is to be the first. The purer the content of the processing scene image corresponding to each sub-region, the better. The more likely a sub-region is to correspond to the high-voltage switch housing area of ​​the processing platform, the more likely the first sub-region is to be affected. The greater the attention given to each sub-region. Indicates the first The relative brightness of each sub-region, the larger the value, the more significant the difference between the first and second sub-regions. The more likely a sub-region is to correspond to a region with a metallic sheen in the image, the more likely the sub-region is to be affected. The more sub-regions there are, the more likely they are to correspond to the high-voltage switch housing. The greater the attention given to each sub-region, the lower the value; the smaller the value, the higher the attention given to the first sub-region. The more likely a sub-region is to correspond to the background region, the more likely the first sub-region is to correspond to the background region. The less attention should be paid to each sub-region, the better. Because metal surfaces are relatively smooth, they are prone to reflection when exposed to light, causing some sub-regions of the high-voltage switch housing to be more cluttered than other sub-regions. To avoid this phenomenon leading to excessive differences in attention between sub-regions belonging to the same high-voltage switch housing, therefore... right Make an upward correction; The smaller, the more... The more likely the adjacent sub-regions of a given sub-region are to belong to the high-voltage switch housing region, the more likely the first sub-region will be. Each sub-region may also correspond to the high-voltage switch housing area, therefore The smaller, the better right The greater the upward correction, the more the first... Each sub-region receives a higher level of attention; The larger the number, the higher the number of... The more likely that adjacent sub-regions of a given sub-region are not all part of the high-voltage switch housing region, then the... The less likely a sub-region is to correspond to a high-voltage switch housing region, therefore The larger, the better right The smaller the upward correction, the less likely it is to cause problems. The individual sub-regions have received excessive attention.

[0055] S4: Determine the thermal conductivity coefficient of each sub-region.

[0056] It should be noted that when using anisotropic filtering to smooth the image, the globally uniform thermal conductivity coefficient cannot simultaneously smooth noise and preserve shell cracks. Therefore, in order to better preserve potential defects on the high-voltage switch shell in the image while smoothing noise, this step obtains the thermal conductivity coefficient of the sub-region based on the attention of the sub-region and the gradient value of the pixels within the sub-region. The basis for this is that the high attention region (shell) should be weakly filtered to preserve details, and the low attention region (background) should be strongly filtered to remove noise. At the same time, the setting of the filtering intensity should also consider the gradient information within the region (a high gradient may mean a defect) and the smooth transition with the neighborhood.

[0057] The thermal conductivity coefficient of the target sub-region is determined based on the magnitude of the attention level and the preset attention threshold.

[0058] Specifically, the thermal conductivity coefficient satisfies:

[0059] ;

[0060] In the formula, For the first The thermal conductivity coefficient of each sub-region For the first The level of attention given to each sub-region For the first In all adjacent sub-regions of the first sub-region, the second sub-region is related to the third sub-region. The attention level of the adjacent sub-region with the smallest difference in attention among sub-regions. For the first The average gradient value of pixels within each sub-region For the first In all adjacent sub-regions of the first sub-region, the second sub-region is related to the third sub-region. The mean gradient value of pixels in adjacent sub-regions with the smallest difference in attention among sub-regions. This is the preset attention threshold.

[0061] Implementers can set a threshold for attention based on the specific implementation situation, for example, 0.6.

[0062] A higher thermal conductivity coefficient results in a smoother image. The larger the number, the higher the number of... The more likely a sub-region is to be located within the high-voltage switch housing area in the image, the smaller the thermal conductivity coefficient should be used to avoid smoothing out critical details such as cracks and minor defects contained within the sub-region. The smaller, the more... The less relevant a sub-region is to the processing monitoring of the high-voltage switch housing, the higher the thermal conductivity coefficient should be used to smooth out details in the image unrelated to the housing processing. To avoid excessive differences in thermal conductivity coefficients between adjacent sub-regions, which could cause artifacts at the junctions of adjacent sub-regions, [further measures should be taken]. and The weighted method smooths out the difference in thermal conductivity coefficients between adjacent sub-regions. High-voltage switch housings typically appear as a continuous solid object with specific textures and edge features in images; the housing does not exist as an isolated pixel or sub-region, but rather as a cluster. Therefore, spatially adjacent sub-regions with similar features are more likely to belong to the same type of region (housing or background). Thus, if the first... If a sub-region belongs to the shell or background, then its adjacent sub-regions with the smallest difference in attention are likely also part of the same type of region. When When this occurs, it means that both sub-regions are more likely to belong to the same shell region. The larger, the more The greater the possibility of shell defects in the first sub-region, the greater the weight of the first sub-region The greater the possibility of shell defects in the first sub-region, the greater the weight of the first sub-region The greater the possibility of shell defects in the first sub-region, the greater the weight of the first sub-region The greater the possibility of shell defects in the first sub-region, the greater the weight of the first sub-region The greater the possibility of shell defects in the first sub-region, the greater the weight of the first sub-region The greater the possibility of shell defects in the first sub-region, the greater the weight of the first sub-region The greater the possibility of shell defects in the first sub-region, the greater the weight of the first sub-region The greater the possibility of shell defects in the first sub-region, the greater the weight of the first sub-region The greater the possibility of shell defects in the first sub-region, the greater the weight of the first sub-region When the above principle is similar, it is not described here. When When the above principle is similar, it is not described here. When When the above principle is similar, it is not described here. When When the above principle is similar, it is not described here. When

[0063] S5: Based on the thermal conductivity coefficient, an anisotropic filtering algorithm is used to obtain a filtered image of the machining scene image.

[0064] Specifically, the filtered image of the machining scene image is obtained, comprising:

[0065] Based on the thermal conductivity coefficient of each sub-region in the machining scene image, an anisotropic filtering algorithm is used to filter each sub-region in the machining scene image to obtain a filtered image of the machining scene image.

[0066] S6: Based on the filtered image, a target detection technology is used to detect defects of the high-voltage switch shell to realize remote monitoring of the numerical control machining platform of the high-voltage switch shell.

[0067] Specifically, the target detection technology uses YOLO algorithm.

[0068] Specifically, the remote monitoring of the numerical control machining platform of the high-voltage switch shell comprises:

[0069] When there is a defect on the high-voltage switch shell, the specific information of the defect (including the defect position, type, severity and local screenshot of the corresponding filtered image) is pushed to the remote monitoring terminal of the numerical control machining platform, and a warning prompt is sent to notify the maintenance personnel to handle the machining failure in time to realize the processing of the defect and complete the remote monitoring of the numerical control machining platform of the high-voltage switch shell.

[0070] The above only describes the preferred embodiment of the present application and does not limit the present application. Any modification, equivalent replacement, improvement, etc. within the principles of the present application shall be included in the protection scope of the present application.

Claims

1. A numerical control machining platform remote monitoring method of a high-voltage switch housing, characterized in that, The method comprises the following steps: acquire a machining scene image of a numerical control machining platform including a high-voltage switch shell, the machining scene image being a grayscale processed image; segment the machining scene image into a plurality of sub-regions by using a segmentation algorithm, any sub-region being recorded as a target sub-region, and determine the disorder degree of the target sub-region according to the number of corner points in the target sub-region, the maximum and minimum grayscale levels of the pixel points in the target sub-region, the disorder degree satisfying: ； In the formula, is the number of sub-regions, is the degree of disorder of the sub-region, is the number of corner points in the sub-region, are the maximum and minimum gray levels of the pixel points in the sub-region, respectively, is a linear normalization function. determine the attention degree of the target sub-region: ； In the formula, is the attention degree of the th sub-region, is the clutter degree of the th sub-region, is the average value of the gray values of the pixels in the th sub-region, is the number of sub-regions in the processing scene image, is the average value of the clutter degrees of all the adjacent sub-regions of the th sub-region, is a maximum function; determine the heat conduction coefficient of the target sub-region: ； In the formula, the heat conduction coefficient of the first sub-region, the attention degree of the first sub-region, the average value of the gradient values of the pixels in the first sub-region, the attention degree of the adjacent sub-region with the least difference in the attention degree from the first sub-region among all the adjacent sub-regions of the first sub-region, the average value of the gradient values of the pixels in the first sub-region, the average value of the gradient values of the pixels in the adjacent sub-region with the least difference in the attention degree from the first sub-region among all the adjacent sub-regions of the first sub-region, is a preset attention degree threshold. based on the heat conduction coefficient, acquire a filtered image of the machining scene image by using an anisotropic filtering algorithm, comprising: based on the heat conduction coefficient of each sub-region in the machining scene image, filter each sub-region in the machining scene image by using the anisotropic filtering algorithm to obtain the filtered image of the machining scene image; based on the filtered image, detect defects of the high-voltage switch shell by using a target detection technology to realize remote monitoring of the numerical control machining platform of the high-voltage switch shell.

2. The numerical control machining platform remote monitoring method of a high-voltage switch shell according to claim 1, characterized in that, The segmentation algorithm adopts a superpixel segmentation algorithm.

3. The numerical control machining platform remote monitoring method of a high-voltage switch shell according to claim 1, characterized in that, The corner points are acquired by using a FAST corner point detection algorithm.

4. The numerical control machining platform remote monitoring method of a high-voltage switch shell according to claim 1, characterized in that, The target detection technology adopts a YOLO algorithm.

5. The numerical control machining platform remote monitoring method of a high-voltage switch shell according to claim 1, characterized in that, The remote monitoring of the numerical control machining platform of the high-voltage switch shell comprises: in response to the existence of defects on the high-voltage switch shell, push the specific information of the defects to a remote monitoring terminal of the numerical control machining platform and issue a warning prompt to realize the processing of the defects and complete the remote monitoring of the numerical control machining platform of the high-voltage switch shell.

Images