A high voltage breakdown point positioning method and system for insulation detection

Abstract

The present application relates to the technical field of image recognition, and particularly relates to a high-voltage breakdown point positioning method and system for insulation detection, which comprises the following steps: obtaining a high-voltage breakdown original image and a reference image; filtering out static reflection and dynamic noise based on brightness contrast and spatial neighborhood information, and identifying an arc reliable saturated area; calculating an adaptive noise baseline of a pixel; iteratively searching outward from the saturated area, and determining a ring-shaped glow area based on the adaptive noise baseline; constructing a point diffusion glow model, performing nonlinear fitting using glow area data, and obtaining model center coordinates as sub-pixel positioning results. The present application solves the positioning problem caused by arc blinding by positioning the saturated area through double confidence evaluation, constructing the glow area by using the adaptive noise baseline and iterative search, and inversely solving the pixel coordinates based on the physical model.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and in particular to a method and system for locating high-voltage breakdown points for insulation detection. Background Technology

[0002] In applications such as power electronics, new energy vehicles, and rail transportation, the insulation performance of insulated flexible copper busbars is crucial to the safe operation and long-term reliability of the integrated equipment. To ensure the quality of insulation components, high-voltage insulation withstand voltage testing is required during the production process. This test involves applying a high voltage between the internal conductor and the external conductive medium of the insulation component and monitoring changes in leakage current to determine the presence of defects in the insulation layer. If the leakage current increases significantly during the test, it indicates that the insulation layer has broken down, posing a potential failure risk.

[0003] Although high-voltage insulation withstand voltage testing can effectively identify whether there is a breakdown problem in the insulated soft copper busbar, this test cannot pinpoint the exact location of the breakdown point, thus failing to meet the actual needs of tracing breakdown defects. To pinpoint the exact location of the breakdown point, existing technologies typically use a high-speed camera to capture the arc flash generated at the moment of breakdown, and then use image processing algorithms to identify the flash area, thereby estimating the coordinates of the breakdown point.

[0004] However, this method has limitations: the arc brightness generated by high-voltage breakdown is extremely high, which easily causes pixel saturation in the core area of ​​visible light camera sensors, forming a flat-topped oversaturated region without gradient information, resulting in the loss of the true light intensity distribution at the center of the arc. Existing processing strategies usually use the geometric centroid of the oversaturated region as the breakdown point, or rely solely on its edge contour for positioning. This processing strategy does not consider the actual physical characteristics of the arc light intensity, limiting the positioning accuracy to the pixel level, and is easily affected by the randomness of the arc shape, making it difficult to achieve stable and accurate positioning. Summary of the Invention

[0005] To address the technical problems of insufficient positioning accuracy and stability caused by pixel saturation due to strong arc light, processing strategy defects, and arc morphology in the aforementioned breakdown point localization method based on visible light imaging, this invention provides solutions in the following aspects.

[0006] In a first aspect, the present invention provides a method for locating high-voltage breakdown points for insulation testing, the method comprising the steps of:

[0007] Acquire the original image of the high-voltage breakdown event and a reference image before breakdown; for any pixel in the original image, determine the candidate confidence level of the pixel based on the pixel brightness comparison between the original image and the reference image, combined with a preset brightness; determine the structural confidence level of the pixel based on the candidate confidence level of the pixel and the candidate confidence levels of each pixel in its neighborhood; calculate the structural confidence level of each pixel in the original image, and perform threshold segmentation based on it to obtain the saturation region of the arc in the breakdown event; construct an annular region for evaluating the glow around the saturation region; the distance between the inner boundary of the annular region and the saturation region is a preset value, and the outer boundary of the annular region is determined by iterative search, which is used to detect the distribution range of pixels in the original image whose brightness is higher than a preset noise baseline; use a preset point diffusion glow model to perform nonlinear fitting on the pixel coordinates and brightness within the annular region to obtain the center coordinates obtained by the nonlinear fitting; the center coordinates are the location result of the high-voltage breakdown point.

[0008] This invention introduces a dual evaluation mechanism of candidate confidence and structural confidence, combined with the comparison of the original image and the reference image, to effectively filter out static reflections and dynamic noise interference, thereby accurately identifying the arc saturation region. At the same time, by using a ring region construction strategy, it avoids overflow interference from the saturation region through a preset inner boundary spacing, and includes effective glow information through iterative search based on the noise baseline. Finally, it uses a point-diffusion glow model to perform nonlinear fitting on the unsaturated glow data, solving the problem of losing center information due to excessive brightness in high-voltage breakdown arcs, and achieving sub-pixel level high-precision positioning.

[0009] Preferably, the step of determining the candidate confidence level of the pixel based on the comparison of pixel brightness in the original image and the reference image, combined with a preset brightness, includes: obtaining a first evaluation value based on the brightness value of the pixel in the original image by using a Gaussian function centered on the upper limit of the brightness saturation; constructing a preset nonlinear gain function based on the brightness gain of the pixel in the original image relative to the reference image to obtain a second evaluation value; the candidate confidence level of the pixel is the product of the first evaluation value and the second evaluation value.

[0010] This invention constructs a Gaussian function centered on the upper limit of brightness saturation to filter pixels in an oversaturated state. At the same time, it combines a nonlinear gain function relative to a reference image and uses a double weighting method to ensure that only pixels that meet both high brightness characteristics and obvious dynamic change characteristics are retained. This mechanism can accurately distinguish between high-brightness static background reflections and real dynamic arc signals.

[0011] Preferably, the candidate confidence scores of the pixels satisfy the following relationship:

[0012] ;

[0013] in, It is a pixel. Candidate confidence level; , They are pixels Brightness values ​​in the original image and reference image; It is the upper limit of image brightness saturation; It is a maximum value function; It is a preset brightness attenuation factor; It is a preset dynamic threshold factor; It is a natural exponential function.

[0014] This invention utilizes the mathematical properties of exponential functions and Gaussian distributions to establish a pixel evaluation standard through parameter control of brightness attenuation factors and dynamic threshold factors. This relationship can respond to minute changes in pixel brightness, while limiting negative fluctuations through a maximum value function, thereby achieving accurate digital characterization of the core region of the electric arc and providing a high-confidence data foundation for subsequent structural analysis.

[0015] Preferably, determining the structural confidence of a pixel based on its candidate confidence and the candidate confidence of each pixel in its neighborhood includes: performing multi-scale Gaussian difference analysis on the confidence map formed by the candidate confidence of all pixels to obtain the response value of each pixel at multiple scales; using the maximum response value at the multiple scales as the feature scale energy of the corresponding pixel; and the structural confidence of the pixel is the product of the sum of the feature scale energies of each pixel in its neighborhood and the candidate confidence of the pixel.

[0016] This invention introduces multi-scale Gaussian difference analysis, which can capture the spatial structural features of arc signals at different scales, distinguishing between arc clumps with solid morphology and isolated noise without structural features. By calculating the energy of the feature scale and combining it with neighborhood information, the spatial continuity of the arc signal is used to suppress isolated random noise points, improve the anti-interference ability of saturated region identification, and ensure the reliability of subsequent positioning references.

[0017] Preferably, obtaining the preset noise baseline includes: calculating the brightness standard deviation of all pixels in the reference image; performing mean filtering on the reference image to obtain the average brightness of each pixel; adding the average brightness to a preset multiple of the brightness standard deviation; and obtaining the noise baseline of the pixel.

[0018] Preferably, the iterative search includes: forming several search rings layer by layer outward from the inner boundary as the initial boundary; calculating the proportion of pixels with brightness higher than the noise baseline in each search ring; and when the proportion is lower than a preset ratio threshold, taking the outer boundary of the current search ring as the outer boundary of the annular region.

[0019] This invention employs an iterative search strategy from the inside out, combined with statistical determination of the proportion of pixel counts, to achieve dynamic optimization of the outer boundary of the glow region. This mechanism can adapt to the diffusion range of arcs of different intensities, ensuring sufficient fitting data while effectively truncating edge regions with low signal-to-noise ratios, preventing background noise data from entering the fitting model and thus affecting the accuracy of the final positioning.

[0020] Preferably, the preset point diffusion glow model is a rotatable elliptic Gaussian model.

[0021] Preferably, the acquisition of the original image includes: acquiring a continuous image segment before and after the breakdown event to obtain an image sequence; traversing each frame of the image sequence, calculating the total number of oversaturated pixels or the overall average brightness of each frame, and selecting the image frame with the highest total number of oversaturated pixels or the overall average brightness as the original image.

[0022] Preferably, the nonlinear fitting is implemented using the LM algorithm, which is used to minimize the sum of squared residuals between the brightness of the pixels within the annular region and the brightness of the pixels predicted by the preset point diffusion glow model.

[0023] In a second aspect, the present invention provides a high-voltage breakdown point location system for insulation detection. The high-voltage breakdown point location system for insulation detection includes a memory and a processor. The memory stores computer program instructions, which, when executed by the processor, implement a high-voltage breakdown point location method for insulation detection according to the first aspect of the present invention.

[0024] By adopting the above technical solution, a high-voltage breakdown point location method for insulation detection according to the first aspect of the present invention is generated into a computer program and stored in a memory so that it can be loaded and executed by a processor. A terminal device can then be made based on the memory and the processor for convenient use.

[0025] The beneficial effects of this invention are as follows: First, by utilizing dual-reset reliability assessment of brightness and structure, combined with reference image comparison, static reflections and random noise in the environment are effectively filtered out, thus pinpointing the arc saturation core. Then, by constructing a ring-shaped glow region that avoids the overflow area and adapts to local background noise, effective data containing the true light intensity gradient is extracted. Finally, a physical model is used to perform inverse fitting of the glow data, breaking through the accuracy limitations of traditional pixel-level positioning. This invention not only solves the problem of information loss caused by oversaturation but also overcomes the influence of background interference and irregular arc morphology, improving the accuracy and stability of high-voltage breakdown point positioning in insulation detection. Attached Figure Description

[0026] Figure 1 A flowchart of a high-voltage breakdown point location method for insulation testing provided in an embodiment of the present invention;

[0027] Figure 2 This is a structural block diagram of a high-voltage breakdown point location system for insulation detection provided in an embodiment of the present invention. Detailed Implementation

[0028] The first aspect of this invention provides a method for locating high-voltage breakdown points for insulation testing, such as... Figure 1 As shown, the method includes steps S100-S400:

[0029] Step S100: Obtain the original image of the high voltage breakdown event and the reference image before breakdown.

[0030] It should be noted that locating the breakdown point requires using the arc image at the moment of breakdown. The arc at the moment of breakdown has the physical characteristics of most concentrated energy, strongest brightness, and closest to the actual breakdown location. However, background noise before breakdown and the arc decay process after breakdown can both interfere with the judgment of the actual breakdown location. Therefore, it is necessary to collect a continuous image sequence at the moment of breakdown and for a period of time before and after, and eliminate interference information in the non-breakdown stage by temporal comparison to accurately locate the image frame that best reflects the breakdown characteristics.

[0031] Specifically, when the withstand voltage tester detects that the rate of change of the leakage current signal exceeds the rate of change threshold, and the absolute value of the leakage current signal reaches the breakdown judgment threshold, a breakdown event is determined to have occurred. At this time, a high-speed camera acquires continuous images for a period of time before and after the breakdown event, and sorts them according to the acquisition timestamp to form an image sequence. For example, the rate of change threshold is set to 100mA / ms, the breakdown judgment threshold is set to 50mA, and continuous images are acquired from 30ms before the breakdown event to 30ms after the event.

[0032] Subsequently, each frame in the image sequence is traversed, and the original image with the strongest arc flash is selected. To evaluate the brightness intensity and energy focusing degree of the arc, the total number of oversaturated pixels or the overall average brightness of each frame is calculated, and the image with the highest total number of oversaturated pixels or the highest overall average brightness is selected as the original image of the breakdown event. In addition, to eliminate static interference in subsequent calculations, a frame without arc flash, acquired before the breakdown event, needs to be selected from the image sequence as a reference image. Preferably, the first frame in the image sequence is selected as the reference image.

[0033] At this point, the original and reference images of the breakdown event have been obtained.

[0034] Step S200: For any pixel in the original image, based on the comparison of pixel brightness in the original image and the reference image, and combined with a preset brightness, determine the candidate confidence of the pixel; based on the candidate confidence of the pixel and the candidate confidence of each pixel in its neighborhood, determine the structural confidence of the pixel; calculate the structural confidence of each pixel in the original image, and perform threshold segmentation based on it to obtain the saturation region of the arc in the breakdown event.

[0035] It should be noted that in actual industrial testing environments, directly applying high-brightness thresholding to the original image to extract the arc core is unreliable due to two types of interference. These two types of interference include: static reflection interference, such as the stable high-brightness specular reflection of ambient light by the tested insulation component; these pixels may be saturated before breakdown occurs. Dynamic noise interference includes transient noise that may be generated by the image sensor in image acquisition equipment under extreme conditions. This noise is characterized by spatial isolation, specifically by a single pixel or a very small number of pixels suddenly becoming extremely bright, while their adjacent pixels remain unchanged, resembling isolated bright white salt grains in the image. Therefore, it is necessary to filter out these two types of interference layer by layer to extract a highly reliable, morphologically intact, and saturated region.

[0036] Specifically, obtaining saturated regions includes: obtaining candidate confidence; obtaining structural confidence; and identifying saturated regions.

[0037] First, candidate confidence scores are obtained. It's important to note that obtaining candidate confidence scores is to suppress static reflection interference. Considering that real arc pixels should be bright in the original image, these bright pixels should be the primary focus when calculating candidate confidence scores. However, static reflection points may also be saturated in the original image, and they may also be bright or saturated in the reference image. To filter out this interference, the candidate confidence score calculation must also ensure that the brightness value of the pixel is significantly increased compared to the brightness of the reference image.

[0038] Based on the above logic, the calculation of the candidate confidence of each pixel in the original image includes: constructing a Gaussian function centered on the upper limit of image brightness saturation to evaluate how close the brightness of the pixel in the original image is to the upper limit of saturation, denoted as the brightness prior term; constructing a nonlinear gain function to evaluate whether the brightness gain of the pixel relative to the reference image is significant enough, denoted as the dynamic gain term; and fusing the brightness prior term and the dynamic gain term to obtain the candidate confidence of the pixel.

[0039] Preferably, each pixel in the original image The candidate confidence scores satisfy the following relation:

[0040] ;

[0041] in, It is a pixel. Candidate confidence level; , They are pixels Brightness values ​​in the original image and reference image; It is the upper limit of image brightness saturation, usually 255; It is a maximum value function, used to guarantee ; It is a preset brightness attenuation factor; It is a preset dynamic threshold factor; It is a natural exponential function.

[0042] In this relation, It is a brightness prior, only when the pixel... When the value in the original image approaches the upper limit of brightness saturation, this term approaches 1; It is the dynamic gain term, and in this term... Used to calculate pixels The brightness gain from the reference image to the original image, if the pixel... Brightness of the original image Rather than the brightness of the reference image When they are approximately equal, it indicates that the brightness of that pixel has not changed significantly, corresponding to a static signal, at which point the dynamic gain term approaches 0; if Significantly greater than When the value is 1, it indicates that the pixel has a significant brightness gain, which corresponds to the dynamic signal. At this time, the dynamic gain term approaches 1.

[0043] It should be noted that the preset brightness attenuation factor The settings need to ensure that the Gaussian function is very sharp, only... A high score can only be obtained with 255 or 254 pixels, thus strictly enforcing the prior of saturation. It needs to be set relatively small. For example, The setting range is Preset dynamic threshold factor The settings should be configured based on a reference image to ensure that only statistically significant increases in brightness are considered significant gains. For example, Set to 3 times the global pixel brightness standard deviation of the reference image.

[0044] Secondly, when obtaining structural confidence, it's important to note that while candidate confidence suppresses static reflection interference, it may still retain dynamic noise interference. For example, spatially isolated transient noise points generated by image sensors in image acquisition devices. Real electric arc signals typically appear spatially as spots or clumps with a certain scale and structure, while noise lacks these structural features. Therefore, obtaining structural confidence can be achieved by analyzing the spatial structure dimension to distinguish between real signal clumps with structural features and unstructured random noise points, thus suppressing this isolated noise.

[0045] Based on the above logic, the calculation of structural confidence includes: First, applying multi-scale Gaussian difference analysis to the candidate confidence map composed of the candidate confidence scores of all pixels. Specifically, a series of Gaussian kernels with different standard deviations are used to convolve the candidate confidence map to generate a set of images with different degrees of blur. After sorting the images in ascending order of Gaussian kernel standard deviation, adjacent images are subtracted pairwise according to the rule of subtracting the previous image from the next image, resulting in a set of Gaussian difference response maps. This Gaussian difference response map can highlight the speckle structure of a specific size, and the response value in the map represents the structural strength of the pixel at that scale. Second, each pixel in the response map is traversed, and the maximum response value of the pixel in all scale response maps is taken as the feature scale energy of the pixel. Then, based on the candidate confidence score of each pixel and the feature scale energy of its neighboring pixels, the structural confidence score of each pixel is obtained. Multi-scale Gaussian difference and convolution of images using Gaussian kernels are existing technologies and will not be elaborated on here.

[0046] Preferably, each pixel in the original image Structural confidence Satisfying the relation:

[0047] ;

[0048] in, It is a pixel. The final structural confidence level; It is a pixel. Candidate confidence level; It is a pixel. Neighboring pixels The characteristic scale energy; It refers to all pixels within the neighborhood window.

[0049] In this relation, pixel The sum of the feature scale energies of all pixels within the neighborhood window, and then multiplied by the pixel's energy. The candidate confidence scores are multiplied to ensure that only pixels themselves have high candidate confidence scores, and only when the neighborhood window to which that pixel belongs has high structure energy can a higher structure confidence score be obtained. For isolated salt grain noise, although its It might be very high, but its neighborhood window is almost zero, leading to its final... Approaching 0.

[0050] It should be further noted that the range of scales used in multi-scale Gaussian differences should be chosen to cover the expected arc spot size. Neighborhood window The size should be set relatively small to ensure that only the energy of the immediately adjacent feature scale is counted. Preferably, it is set to [size missing]. or .

[0051] Next, the saturation region is identified. It should be noted that the structural confidence score has integrated information from both temporal and spatial dimensions and suppressed static reflection interference and dynamic noise interference. Therefore, a reliable saturation region can be obtained based on the structural confidence score. An effective method for converting structural confidence scores into regions is thresholding. However, simple thresholding may retain isolated bright spots formed by tiny noise clumps. Opening operations, using small structural elements, can remove isolated clumps smaller than the structural element while maintaining the integrity of the arc body. Based on this, this invention combines thresholding with morphological opening operations to obtain the saturation region. Morphological opening operations are existing technology and will not be elaborated upon here.

[0052] Specifically, firstly, a high threshold is applied to the structure confidence score, filtering out all pixels with a structure confidence score greater than a preset threshold to form binary regions. Secondly, a morphological opening operation is performed on the binary regions to obtain saturated regions. It should be noted that the preset threshold should be sufficiently high to ensure high confidence; for example, it can be set to 0.8. The structuring element in the morphological opening operation should be small, such as... The disc or square shape can remove minor noise without over-eroding saturated areas.

[0053] At this point, the saturated region in the original image has been obtained.

[0054] Step S300: Construct an annular region for evaluating the glow around the saturated region; the distance between the inner boundary of the annular region and the saturated region is a preset value, and the outer boundary of the annular region is determined by iterative search, which is used to detect the distribution range of pixels in the original image whose brightness is higher than a preset noise baseline.

[0055] It should be noted that saturated regions themselves lose intensity information due to signal saturation and cannot be used directly. This invention considers utilizing the surrounding unsaturated glow regions. However, when selecting these glow regions, it is important to consider that pixels adjacent to saturated regions may be contaminated by charge overflow effects. Therefore, the selected glow region should avoid this contaminated area. Furthermore, in real-world testing scenarios, uneven background brightness, such as dark areas on the sample and bright areas on the fixture, causes spatial variations in noise levels. Therefore, the selected glow region should only include pixels with brightness values ​​significantly higher than their local background noise.

[0056] Specifically, obtaining the glow region includes: determining an adaptive noise baseline; iteratively searching for and generating the glow region.

[0057] First, an adaptive noise baseline is determined for each pixel. It should be noted that, to achieve local adaptability, this baseline is defined for each pixel in the original image. Calculate a noise threshold The noise threshold is determined by the local background brightness of the pixel in the reference image and the global noise of the image acquisition device.

[0058] Based on the above logic, the acquisition of the noise baseline for each pixel includes: calculating the brightness histogram of the reference image to determine the dark area, such as the set of pixels whose brightness is lower than the first quartile, calculating the standard deviation of the brightness of all pixels in the dark area, and recording it as global noise; performing mean filtering on the reference image to obtain the brightness mean of each pixel; adding the brightness mean to the global noise amplified by a preset factor to obtain the noise baseline of that pixel.

[0059] Preferably, each pixel in the original image noise baseline Satisfying the relation:

[0060] ;

[0061] in, It is a pixel. Noise baseline; It is a pixel. Average brightness on the reference image; It is the preset sensitivity factor; It is global noise.

[0062] In this relationship, pixels The noise threshold consists of two parts: the mean brightness and the global noise. The mean brightness is an adaptive term that varies with... The background, such as dark or bright areas, varies; global noise is a constant, representing the unavoidable and transient effects of the image acquisition device. Noise amplification factor. By combining the brightness mean with global noise, a noise baseline can be provided for each pixel, allowing it to automatically adapt to the background characteristics of different regions in the image.

[0063] It should be noted that the preset sensitivity factor The settings should strike a balance between signal detection sensitivity and noise suppression robustness, based on the noise characteristics of the image acquisition device and the expected intensity of the glow signal. In one feasible implementation, to ensure that only statistically significant signals are considered valid, The value of is typically between 3 and 5. In this invention, it is preferably set to 3.

[0064] Secondly, the glow region is iteratively searched and generated. It should be noted that after determining the noise baseline for each pixel, the outer boundary of the glow region needs to be determined. The actual intensity of the electric arc varies, and the position of this outer boundary cannot be fixed in advance. The iterative search algorithm can probe the signal layer by layer outward from the core region until the signal is submerged in its respective local noise baseline, making it suitable for dynamically finding this boundary. Based on this, this invention uses this algorithm to dynamically determine the outer boundary of the glow region and, combined with a fixed inner boundary, generates a ring-shaped glow region.

[0065] Specifically, the maximum effective region defined by the outer boundary is obtained through morphological dilation; the contaminated region defined by the inner boundary is obtained through morphological dilation; and the annular glow region is obtained by subtracting the contaminated region from the maximum effective region through set difference operation.

[0066] Preferably, inner boundary These are fixed preset values ​​used to set the isolation zone. Outer boundary The acquisition includes: First, initializing the search distance. Secondly, an iterative loop is entered, and in each iteration, the current search distance is extracted through morphological operations. The single-pixel ring; then, traversing each pixel on the single-pixel ring. The brightness of its original image Its noise baseline Comparison; statistically satisfying single-pixel ring band The proportion of reliable pixels Finally, Compared to the preset minimum ratio Compare, if If the signal is completely overwhelmed by noise, the loop terminates; otherwise, the current search distance is... Increment by 1, return to the beginning of the loop and continue the search. When the loop terminates, Assigned value The last ring distance with a qualified signal-to-noise ratio.

[0067] It should be noted that the inner boundary The optimal setting is 1 or 2 pixels to ensure that the glow area does not contain contaminated pixels adjacent to the saturated area. Preset minimum ratio. The preferred setting is 0.5, which ensures that the search stops when most pixels on the ring are already noisy.

[0068] At this point, the glow region in the original image has been obtained.

[0069] Step S400: Use a preset point diffusion glow model to perform nonlinear fitting on the pixel coordinates and brightness within the annular region to obtain the center coordinates obtained by the nonlinear fitting; the center coordinates are the location result of the high-voltage breakdown point.

[0070] It should be noted that this step uses the coordinates and brightness values ​​within the glow region to fit a physical model that can describe the diffusion of the point light source. Through nonlinear optimization, the center coordinates of the physical model are solved in reverse, and these coordinates are the accurate location of the arc center.

[0071] Specifically, firstly, all pixels within the glow region are traversed, and the integer coordinates of each pixel and its corresponding actual brightness value in the original image are extracted. These data are then combined into a set of coordinate-brightness triples to form a structured dataset for nonlinear fitting. ;in, , They are the first in the glow region The x and y coordinates of each pixel; It is the first The actual brightness value corresponding to each pixel in the original image.

[0072] Secondly, in constructing the point-diffusion glow model, it should be noted that physically, when an ideal point-shaped electric arc source propagates in a medium, its energy distribution can be approximated as a centrally symmetric two-dimensional Gaussian distribution. This model can accurately describe the energy attenuation law from the center outwards. However, real arc diffusion may exhibit asymmetry, for example, due to the influence of the medium or electromagnetic field. Considering that a rotatable elliptical Gaussian model can simulate not only different diffusion degrees along the principal and secondary axes but also ellipses with arbitrary rotation angles, this invention preferably employs a rotatable elliptical Gaussian model to fit the potentially asymmetric real brightness distribution in the glow region.

[0073] Preferably, the rotatable elliptic Gaussian model satisfies the following relation:

[0074] ;

[0075] in, Is the model at the point Predicted pixel brightness; It is the peak amplitude of the Gaussian distribution, representing the theoretical peak brightness at the center of the electric arc; The coordinates of the center of the Gaussian distribution are both the core parameters to be solved and the center of the arc pixel. ; It is the standard deviation of the Gaussian distribution along the principal and secondary axes, describing the degree of diffusion of the glow; It is the rotation angle of the Gaussian distribution; It is the background light intensity offset, used to fit the average background brightness of the glow region; It is any coordinate of the glow region in the original image; It is a natural exponential function.

[0076] This relationship describes a rotatable two-dimensional Gaussian function, and the model predicts the brightness. With point To the center of the electric arc The distance decreases exponentially, and the exponential part of the equation is used to calculate the distance to the point. Scaling distance along the Gaussian elliptical coordinate system, Control the diffusion range of the glow in the two principal axis directions; Control the overall brightness of the glow; This is used to upscale the model to the actual background brightness of the image. Through this set of parameters... The combined effect of these factors allows the model to define an elliptical Gaussian brightness distribution.

[0077] Then, the fitting and center coordinate acquisition are performed. It should be noted that this is a structured dataset. The data is real data, while the point spread glow model is a physical model. This step uses an optimization algorithm to find the optimal model parameters so that the difference between the brightness predicted by the model and the actual brightness in the structured dataset is minimized.

[0078] Specifically, the structured dataset will be concentrated in All data points, i.e., the actual pixel coordinates and brightness values ​​within the glow region, are substituted into the point-diffusion glow model. Iterative optimization is performed using the nonlinear least squares method, preferably the LM algorithm. This algorithm is a standard algorithm for solving nonlinear least squares problems, possessing both stability and fast convergence, and is suitable for Gaussian model fitting in this invention. When using it... The initial value is set to the geometric center of the saturation region. The initial value is set to the maximum brightness value of the glow region. The algorithm iteratively adjusts the model parameters to minimize the sum of the squared differences between the model-predicted pixel brightness of all pixels within the glow region and the actual brightness of that pixel in the original image. After optimization, the LM algorithm outputs a set of optimal model parameters. These are the coordinates of the arc pixel center obtained by this invention.

[0079] The second aspect of this embodiment provides a high-voltage breakdown point location system for insulation detection, such as... Figure 2 As shown, the high-voltage breakdown point location system for insulation testing includes a memory and a processor. The memory stores computer program instructions, which, when executed by the processor, implement a high-voltage breakdown point location method for insulation testing according to the first aspect of the present invention.

[0080] The high-voltage breakdown point location system for insulation testing also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the setup and functions of which are known in the art and will not be described in detail here.

[0081] In this invention, the aforementioned memory can be any tangible medium containing or storing a program that can be used or combined with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as resistive random access memory (DRAM), dynamic random access memory (DRAM), static random access memory (SRAM), enhanced dynamic random access memory (DRAM), high-bandwidth memory, hybrid memory cube, etc., or any other medium that can be used to store desired information and can be accessed by an application, module, or both. Any such computer storage medium can be part of a device or accessible to or connected to a device.

[0082] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A high voltage breakdown point location method for insulation detection, characterized in that, The method comprises the following steps: obtaining an original image of a high-voltage breakdown event and a reference image before breakdown; for any pixel point in the original image, based on the pixel brightness comparison of the pixel point in the original image and the reference image, and in combination with a preset brightness, determining a candidate confidence of the pixel point; based on the candidate confidence of the pixel point and the candidate confidence of each pixel point in the neighborhood of the pixel point, determining a structure confidence of the pixel point; calculating the structure confidence of each pixel point in the original image, and performing threshold segmentation processing based thereon to obtain a saturated region of an arc in the breakdown event; constructing a ring-shaped region for evaluating the glow around the saturated region; the inner boundary of the ring-shaped region has a preset distance from the saturated region, and the outer boundary of the ring-shaped region is determined through iterative search; the iterative search is used to detect the distribution range of the pixel points with brightness higher than a preset noise baseline in the original image; using a preset point spread glow model to perform nonlinear fitting on the pixel coordinates and brightness in the ring-shaped region to obtain a center coordinate obtained through nonlinear fitting; the center coordinate is a positioning result of the high-voltage breakdown point; based on the pixel brightness comparison of the pixel point in the original image and the reference image, and in combination with a preset brightness, determining a candidate confidence of the pixel point, comprising: based on the brightness value of the pixel point in the original image, a first evaluation value is obtained through a Gaussian function with the saturation upper limit of brightness as the center; based on the brightness gain of the pixel point in the original image relative to the reference image, a preset nonlinear gain function is constructed to obtain a second evaluation value; the candidate confidence of the pixel point is the product of the first evaluation value and the second evaluation value; based on the candidate confidence of the pixel point and the candidate confidence of each pixel point in the neighborhood of the pixel point, determining a structure confidence of the pixel point, comprising: performing multi-scale Gaussian difference analysis on a confidence map composed of the candidate confidences of all pixel points to obtain a response value of each pixel point at multiple scales; the maximum response value at multiple scales is taken as a feature scale energy of the corresponding pixel point; the structure confidence of the pixel point is the product of the sum of the feature scale energies of each pixel point in the neighborhood of the pixel point and the candidate confidence of the pixel point.

2. The high voltage breakdown point location method for insulation detection according to claim 1, characterized in that, The candidate confidence of the pixel point satisfies the following relationship: ； wherein, is a candidate confidence of a pixel point ; , are luminance values of a pixel point in the original image, the reference image, respectively; is a saturation upper limit of image luminance; is a maximum function; is a preset luminance attenuation factor; is a preset dynamic threshold factor; is a natural exponential function.

3. The high voltage breakdown point location method for insulation detection according to claim 1, wherein, The acquisition of the preset noise baseline comprises: calculating the brightness standard deviation of all pixel points of the reference image; performing mean filtering on the reference image to obtain the brightness mean value of each pixel point; adding the brightness mean value and a preset multiple of the brightness standard deviation to obtain the noise baseline of the pixel point.

4. The high voltage breakdown point location method for insulation detection according to claim 1, wherein, The iterative search comprises: taking the inner boundary as an initial boundary to form a plurality of layers of search annular zones outward layer by layer; in each layer of the search annular zone, calculating the proportion of the number of pixel points with brightness higher than the noise baseline; when the proportion is lower than a preset proportion threshold, taking the outer boundary of the current search annular zone as the outer boundary of the ring-shaped region.

5. The high voltage breakdown point location method for insulation detection according to claim 1, wherein, The preset point spread glow model is a rotatable elliptical Gaussian model.

6. The high voltage breakdown point location method for insulation detection according to claim 1, wherein, The acquisition of the original image comprises: collecting a continuous image before and after the breakdown event to obtain an image sequence; traversing each frame of image in the image sequence, calculating the total number of oversaturated pixels or the overall average brightness of each frame, and selecting the image frame with the highest total number of oversaturated pixels or overall average brightness as the original image.

7. The high voltage breakdown point location method for insulation detection according to claim 1, wherein, The nonlinear fitting is realized by an LM algorithm, which is used to minimize the residual square sum between the luminance of the pixel points in the annular region and the luminance of the pixel points predicted by the preset point diffusion glow model.

8. A high voltage breakdown point location system for insulation detection, characterized by, The high-voltage breakdown point positioning system for insulation detection comprises a processor and a memory, and the memory stores computer program instructions which, when executed by the processor, implement the method for positioning a high-voltage breakdown point for insulation detection according to any one of claims 1-7.

Images