A spectrum detection method for insulator contamination

Abstract

The present invention discloses a spectrum detection method for insulator contamination. First, a first ultraviolet video is generated based on ultraviolet light, a light radiation intensity sequence is generated based on the first ultraviolet image of the first ultraviolet video, the discharge type is judged according to the characteristic parameters of the light radiation intensity sequence, and the ultraviolet gain is adjusted according to the discharge type to optimize the light intensity fluctuation ratio. Further, a second ultraviolet video is obtained again through ultraviolet gain, and the discharge spots of multiple second ultraviolet images are fused to generate a spot profile and a spot pixel variance. The ideal sampling time is obtained according to the similarity between the discharge spot of the second ultraviolet image and the spot profile, and a visible light image is extracted according to the ideal sampling time. The relative discharge position and the actual discharge area are calculated based on the visible light image and the spot profile, and finally the degree of contamination of the insulator is predicted according to the spot pixel variance, relative discharge position and actual discharge area of ​​multiple sampling periods. The present invention improves the reliability and accuracy of insulator surface contamination detection.

Description

Technical Field

[0001] The present invention relates to the technical field of spectrum detection, and in particular to a spectrum detection method for insulator contamination. Background Art

[0002] In high-voltage transmission lines, non-contact detection can not only significantly reduce manpower and time costs, but also achieve real-time monitoring of insulator status without affecting the normal operation of the power system. The Chinese patent application with publication number CN114565836A discloses a method for evaluating the status of contaminated insulators based on ultraviolet discharge images and hyperspectral images. The method obtains the discharge area based on the abnormal discharge area in the ultraviolet discharge image, constructs an electro-optical joint analysis feature matrix based on the discharge area, and analyzes the contamination level based on the electro-optical joint analysis feature matrix. The Chinese patent application with publication number CN115047010A discloses a method for detecting the degree of insulator contamination based on multi-source spectral sensing technology. The method constructs a neural network model based on ultraviolet feature quantities, infrared feature quantities and hyperspectral feature quantities to evaluate the degree of contamination of insulators. Among them, the ultraviolet feature quantity is generated only based on the number of photons and the spot area in the ultraviolet image.

[0003] However, in the case of weak discharge, the discharge signal is difficult to identify in the UV image, and in the case of strong discharge, overexposure may occur, resulting in the discharge area not being clearly displayed. Therefore, it is necessary to optimize the image quality by adjusting the gain of the UV camera to ensure that the discharge area is clearly visible. It is worth noting that if the gain is too small, it is difficult to detect weak signals, and if the gain is too large, the noise will increase. To ensure the validity of the data, an appropriate gain value must be selected. In addition, the range of the abnormal discharge area of ​​each frame of the UV discharge image varies. It is difficult to fully judge the discharge position and degree of contamination of the insulator based on a single frame of the discharge UV image. Therefore, the existing technology still needs to be further improved. Summary of the invention

[0004] The purpose of the present invention is to overcome the defect in the prior art that the gain of the ultraviolet camera cannot be adjusted, resulting in the ultraviolet image being difficult to recognize or unable to be clearly displayed, resulting in poor detection effect of insulator contamination, and to provide an insulator contamination spectral detection method.

[0005] The purpose of the present invention is achieved through the following technical solutions: A method for detecting insulator contamination spectra, comprising the following steps: Step 1: a spectrometer collects discharge light radiation generated on the surface of the insulator, and divides the discharge light radiation into ultraviolet light and visible light. A first camera unit receives the ultraviolet light based on an initial ultraviolet gain and generates a first ultraviolet video. Step 2, extracting the first ultraviolet image of the first ultraviolet video, extracting the discharge light spot of each ultraviolet image, calculating the light radiation intensity according to the discharge light spot and the initial ultraviolet gain, multiple groups of light radiation intensities constitute a light radiation intensity sequence, and calculating the light intensity fluctuation ratio of the light radiation intensity sequence; Step 3, extracting characteristic parameters of the light radiation intensity sequence, and judging the discharge type according to the characteristic parameters, if the discharge light radiation is corona discharge, increasing the ultraviolet gain until the light intensity fluctuation ratio is within the first threshold interval, if the discharge light radiation is arc discharge, reducing the ultraviolet gain until the light intensity fluctuation ratio is within the second threshold interval, if the discharge radiation is neither corona discharge nor arc discharge, not adjusting the ultraviolet gain; Step 4, the first camera unit receives ultraviolet light based on ultraviolet gain and generates a second ultraviolet video, extracts multiple second ultraviolet images of the second ultraviolet video, and generates a light spot profile and a light spot pixel variance according to the multiple second ultraviolet images; Step 5, calculating the actual spot area of ​​the spot profile based on the ultraviolet gain, and determining the ideal sampling time of the sampling period according to the similarity between the discharge spots of the plurality of second ultraviolet images and the spot profile; Step 6: The second camera unit receives visible light to generate a visible light video, extracts a visible light image at an ideal sampling time in the visible light video, matches the light spot contour with the visible light image, and calculates a relative discharge position of the light spot contour; Step 7, calculating the pixel height of the insulator in the visible light image, and calculating the actual discharge area based on the calibrated height of the insulator, the pixel height and the actual spot area; Step 8, repeating steps 1 to 7, generating a light spot timing matrix according to the light spot pixel variance, actual discharge area and relative discharge position of multiple sampling cycles, and predicting the insulator contamination data according to the light spot timing matrix.

[0006] During the daily operation of the power grid, dirt and contaminants often accumulate on the surface of insulators. When the amount of these contaminants reaches a certain level or contains corrosive components, it will weaken the insulation performance of the insulators, leading to flashover phenomena in the line, thus triggering faults in the power system. The ultraviolet imaging method is a non-contact detection method that can judge the degree of contamination of insulators by observing the discharge conditions on their surfaces. This method can be carried out without power interruption, reducing operation risks and impacts on the power system. To optimize the detection effect, appropriate ultraviolet gain settings are crucial. A reasonable ultraviolet gain can help balance the signal intensity and background noise, thereby increasing the overall light intensity fluctuation ratio. However, an excessive ultraviolet gain will cause the image to overly amplify noise and background signals, making it difficult to detect discharge spots and potentially introducing false positive results. On the contrary, too small an ultraviolet gain will weaken the signal intensity of the discharge spots, resulting in some weak discharge phenomena not being effectively detected, thus reducing the detection sensitivity. Therefore, the present invention proposes a method for adjusting the corresponding appropriate ultraviolet gain for different types of discharge phenomena to ensure the best detection effect in various situations. The surface discharge state of insulators is random. Analyzing only a single-frame ultraviolet image is not comprehensive. The present invention comprehensively analyzes multiple frames of ultraviolet images collected continuously in multiple sampling periods, which can more accurately capture the changing trends and dynamic characteristics of discharge phenomena.

[0007] Preferably, in step 2, the first ultraviolet image of the first ultraviolet video is extracted, the discharge spots of each ultraviolet image are extracted, and the light radiation intensity is calculated based on the discharge spots and the initial ultraviolet gain. Specifically: Obtain the first binary image of the first ultraviolet image and extract the discharge spots of the first binary image. Calculate the area S1 of the discharge spots based on the pixel values of the discharge spots, and then calculate the light radiation intensity. Wherein, G0 is the initial ultraviolet gain, λ is the calibration coefficient of the first imaging unit, and B1 is the average pixel value of the discharge spots in the first ultraviolet image.

[0008] Preferably, the light intensity fluctuation ratio is R = I1 / σ, where I1 is the intensity average value of the light radiation intensity sequence and σ is the intensity standard deviation.

[0009] Preferably, the characteristic parameters of the light radiation intensity sequence include the intensity average value I1, the intensity variance σ2, and the proportion k of the light radiation intensity less than 3I1.

[0010] Preferably, in step 3, the first threshold interval is [R1, R2], and the second threshold interval is [R3, R4], where R2 < R3. If the discharge light radiation is corona discharge, the adjustment step size ΔG of the ultraviolet gain = G min (1 + R / R1), if the discharge light radiation is arc discharge, the adjustment step size ΔG of the ultraviolet gain = Gmin (1+R4 / R), where G min is the minimum adjustable gain.

[0011] Preferably, in step 4, the second ultraviolet image is a grayscale image, multiple second ultraviolet images are fused to generate a mean ultraviolet image, the mean ultraviolet image is converted into a second binary image based on a threshold segmentation method, the mean discharge spot in the second binary image is fitted to generate a spot contour.

[0012] Preferably, the step 5 is specifically as follows: The similarity between the discharge spot and the spot contour is D3 = w1(1-D1)+w2(1-D2), where D1 is the area similarity between the discharge spot and the spot contour, D2 is the contour similarity between the discharge spot and the spot contour, w1 and w2 are weight factors of the area similarity and contour similarity respectively, and the shooting time of the second ultraviolet image corresponding to the maximum similarity D3 is selected as the ideal sampling time.

[0013] Preferably, in step 6, the relative discharge position of the spot profile is calculated, specifically: the insulator profile of the visible light image is obtained, the coordinates (x', y') of the high-voltage side of the insulator are extracted, and the center coordinates of the spot profile are (x0, y0), then the relative discharge position

[0014] Preferably, in step 7, the actual discharge area S=S2H 2 / h 2 , where S2 is the actual spot area, h is the pixel height of the insulator, and H is the calibrated height of the insulator.

[0015] Preferably, in step 8, the insulator contamination data includes contamination components and contamination levels, a contamination classification model is trained according to the light spot pixel variance, actual discharge area and relative discharge position of different contamination components and contamination levels, and the light spot timing matrix is ​​input into the contamination classification model to predict the insulator contamination data.

[0016] The beneficial effects of the present invention are as follows: the present invention first generates a first ultraviolet video based on ultraviolet light, extracts the discharge spot in the first ultraviolet image to generate a light radiation intensity sequence, and determines the discharge type according to the light radiation intensity sequence. According to the different discharge types, the ultraviolet gain is dynamically adjusted to optimize the light intensity fluctuation ratio, thereby improving the detection sensitivity and accuracy. Further, the second ultraviolet video is obtained according to the ultraviolet gain, the discharge spots of multiple second ultraviolet images are fused to generate a spot profile, and the ideal sampling time is obtained based on the spot profile and the discharge spots of multiple second ultraviolet images. By extracting the visible light image according to the ideal sampling time, the relative discharge position and actual discharge area of ​​the spot profile can be accurately calculated, and finally the contamination data of the insulator is predicted according to the relative discharge position and actual discharge area of ​​the spot profile of multiple sampling cycles. This method not only improves the reliability and accuracy of insulator surface contamination detection, but also can be monitored without power outages. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a flow chart of the spectrum detection method of insulator contamination of the present invention; Figure 2 Schematic diagram of the optical paths of ultraviolet light and visible light of the present invention; Figure 3 is a schematic diagram of the light radiation intensity sequence of the present invention; Figure 4 is a schematic diagram of the morphological operation of the present invention; Figure 5 A schematic diagram of generating a mean discharge spot according to the present invention; Figure 6 is a schematic diagram of the relative discharge position of the present invention; Figure 7 It is a schematic diagram of fitting a mean discharge spot and generating a spot profile according to the present invention. DETAILED DESCRIPTION

[0018] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this application will be more comprehensive and complete and fully convey the concept of the example embodiments to those skilled in the art.

[0019] In addition, described feature, structure or characteristic can be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present application. However, those skilled in the art will appreciate that the technical scheme of the present application can be put into practice without one or more of the specific details, or other methods, components, devices, steps, etc. can be adopted. In other cases, known methods, devices, realizations or operations are not shown or described in detail to avoid blurring the various aspects of the application.

[0020] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities may be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0021] The flowcharts shown in the accompanying drawings are only exemplary and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps can be decomposed, and some operations / steps can be combined or partially combined, so the actual execution order may change according to actual conditions.

[0022] Embodiment 1: Reference Figure 1 The insulator contamination spectrum detection method of the present invention described in detail in this embodiment includes the following steps: Step 1: The spectrometer collects the discharge light radiation generated on the surface of the insulator, and divides the discharge light radiation into ultraviolet light and visible light. The first camera unit receives the ultraviolet light based on the initial ultraviolet gain to generate a first ultraviolet video. Figure 2 The spectrometer divides the discharge light radiation into ultraviolet light and visible light, wherein the ultraviolet light passes through the solar-blind ultraviolet filter and the image intensifier to irradiate the first camera unit, the reflector is used to adjust the direction of the ultraviolet light, the solar-blind ultraviolet filter only allows ultraviolet light in the wavelength range of 200-280nm to pass, and the ultraviolet light in the wavelength range of 200-280nm is not affected by solar radiation, which can reduce the background interference of the sun and other light sources, and the image intensifier is used to achieve ultraviolet gain. The initial ultraviolet gain can be set according to the device. In this embodiment, the adjustment range of the ultraviolet gain is [0dB, 100dB], and the initial ultraviolet gain is 60dB.

[0023] Step 2: Extract the first ultraviolet image of the first ultraviolet video, extract the discharge spot of each first ultraviolet image, calculate the light radiation intensity according to the discharge spot and the ultraviolet gain, construct a light radiation intensity sequence from multiple groups of light radiation intensities, and calculate the light intensity fluctuation ratio of the light radiation intensity sequence. In this embodiment, for example, 10 frames are selected from the first ultraviolet video per second as the first ultraviolet image, and the continuous acquisition is for example 10 seconds, and a total of 100 frames of the first ultraviolet image are extracted. The resolution of the first ultraviolet image is X×Y, and the background noise is removed by background subtraction of the first ultraviolet image, and then the first binary image of the first ultraviolet image is generated based on the threshold segmentation method. When the pixel point (x, y) of the first ultraviolet image belongs to the discharge spot, the pixel value of the first binary image is set to 1, otherwise it is set to 0.

[0024] In this embodiment, the discharge spot of the first binary image is extracted, and the discharge spot area S1 is calculated according to the pixel value of the discharge spot, that is, the discharge spot area Where H(x,y) is the pixel value of the pixel point (x,y) in the first binary image, then the light radiation intensity Wherein, λ is the calibration coefficient of the first camera unit, B1 is the average pixel value of the discharge spot in the first ultraviolet image, and G0 is the initial ultraviolet gain. In a simpler embodiment, the calibration coefficient λ can be verified by experiment. The physical meaning of the light intensity fluctuation ratio is the relationship between the light radiation intensity and the background noise. The intensity average value I1 and the intensity standard deviation σ of the light radiation intensity sequence are calculated, and the light intensity fluctuation ratio R=I1 / σ.

[0025] Step 3: Extract the characteristic parameters of the light radiation intensity sequence, and judge the discharge type according to the characteristic parameters. If the discharge light radiation is corona discharge, increase the UV gain until the light intensity fluctuation ratio is within the first threshold interval. If the discharge light radiation is arc discharge, reduce the UV gain until the light intensity fluctuation ratio is within the second threshold interval. Otherwise, do not adjust the UV gain. Figure 3 , the characteristic parameters include the intensity average value I1, intensity variance σ 2 And the proportion k of light radiation intensity less than 3I1. The samples with similar characteristic parameters are grouped into different clusters through clustering algorithm. Each cluster represents a discharge type. The optimal number of clusters is determined by the silhouette coefficient to establish a discharge classification model. For the current light radiation intensity sequence, its characteristic parameters are extracted, and the characteristic parameters are compared with the existing clusters in the discharge classification model. The distance between the characteristic parameters and the center of each cluster is calculated. The discharge type to which the cluster with the nearest distance belongs is the discharge type of the current discharge light radiation.

[0026] Arc discharge usually has a higher intensity and less volatility, while the pixel edges of corona discharge are blurred and have high volatility. In addition to the clustering algorithm adopted in this embodiment, the present invention can also adopt preset discharge judgment conditions. The discharge judgment conditions include: if I1 ≤ 10 μW / cm 2 , σ 2 ≥ 1 μW / cm 2 , and k ≥ 0.2, then the discharge light radiation is corona discharge. If I1 ≥ 100 μW / cm 2 , σ 2 ≤ 5 μW / cm 2 , then the discharge light radiation is arc discharge. The first threshold interval is [R1, R2], and the second threshold interval is [R3, R4], where R2 < R3. If the discharge light radiation is corona discharge, the adjustment step size ΔG of the ultraviolet gain = G min (1 + R / R1). If the discharge light radiation is arc discharge, the adjustment step size ΔG of the ultraviolet gain = G min (1 + R4 / R), where G min is the minimum adjustable gain. The minimum adjustable gain is related to the device itself and is usually 0.1 dB to 1 dB. Similarly, the first threshold interval can be [5, 10], and the second threshold interval can be [30, 40].

[0027] Step 4: The first imaging unit receives ultraviolet light based on the ultraviolet gain and generates a second ultraviolet video. At the same time, the second imaging unit receives visible light and generates a visible light video. Extract multiple second ultraviolet images of the second ultraviolet video, and generate a spot contour and spot pixel variance based on the multiple second ultraviolet images. Refer to Figure 4 , the second ultraviolet image is a grayscale image. Perform morphological operations on the second ultraviolet image, and remove noise points and internal holes in the second ultraviolet image through grayscale opening operation and grayscale closing operation, and smooth the spot boundary to generate a clean second ultraviolet image. Refer to Figure 5 , then use the average value method to fuse the pixel values of the discharge spots in multiple second ultraviolet images to generate an average ultraviolet image, and generate a second binary image of the average ultraviolet image based on the threshold segmentation method. That is, for any pixel point of the average ultraviolet image, if the pixel value of this pixel point is greater than B2, its pixel value in the second binary image is 1. If the pixel value of this pixel point is less than or equal to B2, its pixel value in the second binary image is 0, where B2 is the average pixel value of the average ultraviolet image.

[0028] In this embodiment, for example, a minimum ellipse fitting method is used, and an edge detection algorithm is used to extract the contour of the mean discharge spot from the second binary image, and all pixel points on the contour are obtained. A target fitting function is constructed according to the principle of least squares method, pixel points and target fitting graph. The target fitting graph is an ellipse. The partial derivatives of the target fitting function are calculated and set to zero to generate a set of equations. Finally, the set of equations is solved to find the optimal parameters of the target fitting function. The spot contour is generated according to the optimal parameters and the ellipse equation. The specific implementation method refers to Example 2. The present invention does not limit the fitting method of the spot contour. The mean discharge spot can be fitted as a minimum ellipse by the least squares method, or the mean discharge spot can be fitted as a smooth spline curve by the B-spline curve fitting. The spot pixel variance is the variance of the pixel values ​​in the mean ultraviolet image.

[0029] Step 5: Calculate the actual spot area of ​​the spot profile based on the ultraviolet gain, and determine the ideal sampling time of the sampling period according to the similarity between the discharge spot of the multiple second ultraviolet images and the spot profile. In this embodiment, the spot profile is a minimum ellipse, and the spot profile area E1 is calculated according to the major axis L1 and the minor axis L2 of the spot profile, E1 = πL1 L2 / 4. Obtain a sample group with different ultraviolet gains, spot profile areas E1 and actual spot areas S2, take the ultraviolet gain and spot profile area E1 in the sample group as input, and the actual spot area S2 as output, use the linear regression method to determine the quantitative relationship between the input and output, establish an actual spot area prediction model, and predict the actual spot area S2 through the ultraviolet gain, spot profile area E1 and actual spot area prediction model.

[0030] The discharge spot area E2 is calculated according to the number of pixels occupied by the discharge spot, and the area similarity D1 = |E1-E2| / |E1+E2|. The pixels on the border of the discharge spot and the spot outline are extracted to generate the pixel point set O of the discharge spot and the pixel point set P of the spot outline respectively, and the one-way Hausdorff distance h(O, P) from the pixel point set O to the pixel point set P is calculated = max o∈O min p∈P ||op||, and the one-way Hausdorff distance from pixel set P to pixel set O h(P, O) = max p∈P min o∈O ||po||, then calculate the bidirectional Hausdorff distance H(O,P)=max(h(O,P),h(P,O)), normalize the bidirectional Hausdorff distance H(O,P) to get the contour similarity Then the similarity between the discharge spot and the spot contour is D3 = w1(1-D1)+w2(1-D2), where w1 and w2 are weight factors of area similarity and contour similarity respectively. In this embodiment, w1 = w2 = 0.5, and the shooting time of the second ultraviolet image corresponding to the maximum similarity D3 is selected as the ideal sampling time.

[0031] Step 6: Extract the visible light image at the ideal sampling time in the visible light video, match the light spot contour with the visible light image, and calculate the relative discharge position of the light spot contour. Figure 6 , insulators usually have a metal conductor, a high-voltage side and a low-voltage side, and the metal conductor is wrapped with insulating ceramic. The same degree of contamination will produce a larger discharge spot on the high-voltage side. Therefore, when a larger discharge spot is observed on the high-voltage side, the degree of contamination is relatively low. Similarly, when a larger discharge spot is observed on the low-voltage side, the degree of contamination is relatively high. The present invention presets an insulator image template, extracts the insulator pattern of the image template from the visible light image, and obtains the insulator contour of the insulator pattern through an edge detection method. The coordinates (x', y') of the high-voltage side of the insulator are obtained according to the insulator contour. The center coordinates of the spot contour are (x0, y0), and the relative discharge position

[0032] Step 7: Calculate the pixel height of the insulator in the visible light image, and calculate the actual discharge area based on the insulator's calibrated height, pixel height, and actual spot area. Obtain the high-voltage side boundary point and low-voltage side boundary point of the insulator according to the insulator contour. The pixel distance between the two boundary points is the pixel height. The actual discharge area S = S2H 2 / h 2 , where h is the pixel height of the insulator and H is the calibrated height of the insulator.

[0033] Step 8: Repeat steps 1 to 7, generate a spot timing matrix according to the spot pixel variance, actual discharge area and relative discharge position of multiple sampling periods, and predict the insulator contamination data according to the spot timing matrix. The pixels, area and region of the insulator discharge spot are related to the insulator contamination. The insulator contamination data includes contamination components and contamination levels. The contamination classification model is trained according to the spot pixel variance, actual discharge area and relative discharge position of different contamination components and contamination levels, and the spot timing matrix is ​​input into the contamination classification model to predict the insulator contamination data. The insulator contamination data can also be contamination accumulation. The difference matrix is ​​obtained by subtracting the spot pixel variance, actual discharge area and relative discharge position of adjacent sampling periods in the spot timing matrix, and the contamination accumulation is predicted according to the size of the difference matrix. The present invention is not limited to this.

[0034] In this embodiment, the pollution data is, for example, the pollution level, and the pollution classification model is, for example, the pollution level prediction model. Pair and combine training samples of n pollution levels to generate n(n-1) / 2 training sample sets, each of which contains samples of two different pollution levels. Train a binary classifier for each training sample set to generate n(n-1) / 2 sub-classifiers capable of binary classification. These n(n-1) / 2 sub-classifiers constitute the pollution level prediction model. The light spot timing matrix is ​​input to the pollution level prediction model. Each sub-classifier classifies the pollution level based on the input light spot timing matrix and outputs the pollution level it considers to be correct. The pollution level predicted the most times among all sub-classifiers is the prediction result of the pollution level prediction model.

[0035] Embodiment 2: Reference Figure 7 , a spectrum detection method for insulator contamination, the principle and implementation method of which are basically the same as those of Example 1, except that, in step 4, the method for generating the spot profile is to use the minimum ellipse fitting method to perform profile fitting on the mean discharge spot and generate the spot profile, specifically: Step 401: The pixel value range of each pixel in the mean ultraviolet image is between 0 and 255. The average pixel value B2 of the mean ultraviolet image is calculated, and each pixel in the grayscale image is checked one by one. If the grayscale value of a pixel exceeds B2, the pixel value of the pixel is set to 1, otherwise, it is set to 0 to generate a second binary image.

[0036] Step 402: Extract the contour of the mean discharge spot from the second binary image, and obtain the coordinates (x m ,y m ), where m∈[1,M], M is the total number of pixels on the contour, and the target fitting function is constructed according to the principle of least squares method. Calculate the partial derivative of the target fitting function f with respect to each parameter and set it to zero to generate a linear equation system. Use numerical methods to solve the above linear equation system, that is, solve Find the optimal parameters A1, A2, A3, A4, A5, A6.

[0037] Step 403: The coordinates of the center point of the minimum ellipse (x0, y0) are The deflection angle of the minimum ellipse θ = -arctan[A1 / (A2-A3)] / 2, the major axis of the minimum ellipse and short axis The light spot profile is the minimum ellipse.

[0038] The minimum ellipse fitting can effectively process graphs with complex and irregular shapes, and the minimum ellipse fitting has high compactness, which can use the fewest parameters to describe the geometric characteristics of the mean discharge spot while ensuring the fitting accuracy.

[0039] Embodiment 3: A method for spectral detection of insulator contaminants, which is different from Embodiment 1 in that it further discloses an optimization method for penalty factors and kernel function parameters, establishes a contamination level prediction model based on the optimal penalty factors and kernel function parameters, and a method for predicting the insulator contamination level based on the contamination level prediction model and the spot time series matrix.

[0040] Step 801: For n contamination levels N1, N2, …, N n , each contamination level has corresponding training samples. Pair and combine the training samples of these n contamination levels to generate n(n - 1) / 2 training sample sets.

[0041] Step 802: For each training sample set, use the sparrow search algorithm to determine the optimal input parameters of the sub-classifier. The input parameters are the penalty factor β and the kernel function parameter γ. First, define the value ranges of the penalty factor β and the kernel function parameter γ. For example, β is between [0.1, 100], and γ is between [0.001, 10]. Randomly generate V groups of initial combinations of β and γ. Then define an evaluation criterion. The evaluation criterion is, for example, the accuracy rate. For each combination of β and γ, use this combination to train an intermediate classifier, use the cross-validation method to evaluate the performance of the intermediate classifier, obtain the accuracy rate, and sort the V groups of combinations of β and γ in descending order of accuracy rate. Select the combinations of β and γ with the top 20% accuracy rate as the discoverers, randomly select some combinations of β and γ as the joiners and the vigilants respectively. During the update process, the positions of the discoverers, joiners, and vigilants are continuously adjusted. If the accuracy rate of the new position is higher than the original one, update the current position to the new position. Through the above iterative process, continuously update the combinations of β and γ. When the preset number of iterations is reached or the change in the accuracy rate tends to be stable, stop the search and output the optimal input parameters.

[0042] Step 803: For each pair of contamination levels (C i , C j ), generate a binary classifier, where C i is the i-th contamination level, C j is the j-th contamination level, i < j. Based on the optimal input parameters, train a binary classifier for each training sample set to generate n(n - 1) / 2 binary-classifiable sub-classifiers. The task of each sub-classifier is to distinguish its corresponding two contamination levels. Combine all the sub-classifiers to form a contamination level prediction model.

[0043] Step 804: for the light spot timing matrix, use each sub-classifier to perform classification, each sub-classifier will output a classification result, count the number of times each pollution level is selected by the sub-classifier, and the pollution level with the most selected times is the prediction result of the pollution level prediction model.

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

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

Claims

1. A spectrum detection method for insulator contamination, characterized in that: The following steps are involved: Step 1: a spectrometer collects discharge light radiation generated on the surface of the insulator, and divides the discharge light radiation into ultraviolet light and visible light. A first camera unit receives the ultraviolet light based on an initial ultraviolet gain and generates a first ultraviolet video. Step 2, extracting the first ultraviolet image of the first ultraviolet video, extracting the discharge light spot of each ultraviolet image, calculating the light radiation intensity according to the discharge light spot and the initial ultraviolet gain, multiple groups of light radiation intensities constitute a light radiation intensity sequence, and calculating the light intensity fluctuation ratio of the light radiation intensity sequence; Step 3, extracting characteristic parameters of the light radiation intensity sequence, and judging the discharge type according to the characteristic parameters, if the discharge light radiation is corona discharge, increasing the ultraviolet gain until the light intensity fluctuation ratio is within the first threshold interval, if the discharge light radiation is arc discharge, reducing the ultraviolet gain until the light intensity fluctuation ratio is within the second threshold interval, if the discharge radiation is neither corona discharge nor arc discharge, not adjusting the ultraviolet gain; Step 4, the first camera unit receives ultraviolet light based on ultraviolet gain and generates a second ultraviolet video, extracts multiple second ultraviolet images of the second ultraviolet video, and generates a light spot profile and a light spot pixel variance according to the multiple second ultraviolet images; Step 5, calculating the actual spot area of ​​the spot profile based on the ultraviolet gain, and determining the ideal sampling time of the sampling period according to the similarity between the discharge spots of the plurality of second ultraviolet images and the spot profile; Step 6: The second camera unit receives visible light to generate a visible light video, extracts a visible light image at an ideal sampling time in the visible light video, matches the light spot contour with the visible light image, and calculates a relative discharge position of the light spot contour; Step 7, calculating the pixel height of the insulator in the visible light image, and calculating the actual discharge area based on the calibrated height of the insulator, the pixel height and the actual spot area; Step 8, repeating steps 1 to 7, generating a light spot timing matrix according to the light spot pixel variance, actual discharge area and relative discharge position of multiple sampling cycles, and predicting the insulator contamination data according to the light spot timing matrix.

2. The method for detecting insulator contamination spectra according to claim 1, characterized in that: In the step 2, the first ultraviolet image of the first ultraviolet video is extracted, the discharge spot of each ultraviolet image is extracted, and the light radiation intensity is calculated according to the discharge spot and the initial ultraviolet gain, specifically: Obtain the first binary image of the first ultraviolet image and extract the discharge spot of the first binary image, calculate the discharge spot area S1 according to the pixel value of the discharge spot, and then calculate the light radiation intensity Wherein, G0 is the initial ultraviolet gain, λ is the calibration coefficient of the first camera unit, and B1 is the average pixel value of the discharge light spot in the first ultraviolet image.

3. The method for detecting insulator contamination spectra according to claim 2, characterized in that: In step 2, the light intensity fluctuation ratio is R=I1 / σ, where I1 is the intensity average value of the light radiation intensity sequence, and σ is the intensity standard deviation.

4. The method for detecting insulator contamination spectrum according to claim 1, characterized in that: In the step 3, the characteristic parameters of the light radiation intensity sequence include the intensity average value I1, the intensity variance σ2, and the proportion k of the light radiation intensity less than 3I1.

5. The method for detecting insulator contamination spectrum according to claim 4, characterized in that: In step 3 described above, the first threshold interval is [R1, R2], and the second threshold interval is [R3, R4], where R2 < R3. If the discharge optical radiation is corona discharge, the adjustment step size ΔG of the ultraviolet gain = G min (1 + R / R1). If the discharge optical radiation is arc discharge, the adjustment step size ΔG of the ultraviolet gain = G min (1 + R4 / R), where G min is the minimum adjustable gain.

6. The method for detecting insulator contamination spectrum according to claim 1, characterized in that: In step 4, the second ultraviolet image is a grayscale image. Multiple second ultraviolet images are fused to generate a mean ultraviolet image. The mean ultraviolet image is converted into a second binary image based on a threshold segmentation method. The mean discharge spot in the second binary image is fitted to generate a spot contour.

7. The method for detecting insulator contamination spectra according to claim 1, characterized in that: The specific step 5 is: the similarity between the discharge spot and the spot contour D3 = w1(1-D1)+w2(1-D2), wherein D1 is the area similarity between the discharge spot and the spot contour, D2 is the contour similarity between the discharge spot and the spot contour, w1 and w2 are weight factors of the area similarity and contour similarity respectively, and the shooting time of the second ultraviolet image corresponding to the maximum similarity D3 is selected as the ideal sampling time.

8. The method for detecting insulator contamination spectrum according to claim 1, characterized in that: In step 6, the relative discharge position of the spot profile is calculated, specifically: the insulator profile of the visible light image is obtained, the coordinates (x', y') of the high-voltage side of the insulator are extracted, and the center coordinates of the spot profile are (x0, y0), then the relative discharge position 9. The method for detecting insulator contamination spectrum according to claim 1, characterized in that: In step 7, the actual discharge area S = S2H 2 / h 2 , where S2 is the actual spot area, h is the pixel height of the insulator, and H is the calibrated height of the insulator.

10. The method for detecting insulator contamination spectrum according to claim 1, characterized in that: In step 8, the insulator contamination data includes contamination components and contamination levels. The contamination classification model is trained according to the light spot pixel variance, actual discharge area and relative discharge position of different contamination components and contamination levels, and the light spot timing matrix is ​​input into the contamination classification model to predict the insulator contamination data.

Images