A method for visualizing insulator pollution levels based on hyperspectral images

Abstract

This invention discloses a method for visualizing the contamination level of insulators based on hyperspectral images, comprising: S1, dividing the naturally contaminated composite insulator surface into regions and acquiring hyperspectral images of them; S2, processing the hyperspectral full-band images and extracting several characteristic wavelengths from the bands included in the hyperspectral data; S3, establishing a correspondence between image grayscale values ​​and contamination levels; S4, dividing the composite insulator surface into regions with different contamination levels based on grayscale thresholds and calculating the area proportion of each contamination level region; S5, superimposing images of regions with different contamination levels and using color to distinguish the contamination level of the insulator surface. This invention utilizes hyperspectral image data to more accurately and efficiently detect the contamination status of insulator surfaces, providing guidance for insulator cleaning work and improving the reliability and safety of transmission line operation.

Description

Technical Field

[0001] This invention belongs to the technical field of insulator contamination detection, specifically relating to a method for visualizing the degree of insulator contamination based on hyperspectral images. Background Technology

[0002] Rapid and accurate detection of contamination on insulator surfaces is crucial for preventing flashover and maintaining the safe and stable operation of power systems. Traditional detection methods rely on offline testing, requiring operation under power-off conditions. Hyperspectral imaging, as a non-contact detection method, utilizes the differences in light reflectivity across areas of varying contamination levels on the insulator surface to achieve online detection of contamination distribution. Hyperspectral imaging can simultaneously acquire spectral and image information of the object under test. Hyperspectral spectral data reflects the light reflection intensity of the object across hundreds of wavebands, containing rich information. However, it also presents challenges in data processing, such as excessive redundant data and cumbersome workflows. Summary of the Invention

[0003] The purpose of this invention is to address the aforementioned shortcomings in the prior art by providing a method for visualizing the degree of contamination of insulators based on hyperspectral images, thereby solving the problems of data redundancy and cumbersome processing procedures in existing hyperspectral imaging methods for detecting the degree of contamination on the surface of insulators.

[0004] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0005] A method for visualizing the contamination level of insulators based on hyperspectral images, comprising the following steps:

[0006] S1. Divide the naturally contaminated composite insulator disc surface into several fan-shaped areas of equal area, acquire hyperspectral images of the composite insulator disc surface to obtain a hyperspectral full-band image of the composite insulator disc surface, and clean each fan-shaped area to obtain its actual degree of contamination.

[0007] S2. Perform dimensionality reduction processing on the hyperspectral full-band image and extract several characteristic wavelengths from the bands contained in the hyperspectral data.

[0008] S3. Calculate the average gray value of the sector region corresponding to each feature wavelength. Based on the average gray value, use the random forest algorithm to select the optimal feature wavelength to construct the correspondence between image gray value and degree of filth, and calculate the gray value threshold of the degree of filth.

[0009] S4. Divide the composite insulator disk surface into regions with different levels of contamination based on the grayscale threshold, and calculate the area ratio of each region with different levels of contamination.

[0010] S5. Overlay images of areas with different levels of contamination and use color to distinguish the degree of contamination on the insulator disc surface.

[0011] Furthermore, step S1 specifically includes:

[0012] S1.1. Under simulated natural lighting conditions, collect hyperspectral image information of composite insulator strings with natural pollution accumulation;

[0013] S1.2. Using the center of the composite insulator disc as the center, divide the composite insulator disc into several sector-shaped areas of equal area, and then divide each sector-shaped area into several sub-sector-shaped areas.

[0014] S1.3 Clean the dirt in each sub-sector area, collect the cleaned wastewater, and calculate the volumetric conductivity σ of the wastewater at room temperature θ. θ ;

[0015] S1.4. Calculate the salinity S of the wastewater based on its volumetric conductivity. a :

[0016] S a =(5.7σ) 20 ) 1.03

[0017] Where, σ 20 The volumetric conductivity of the wastewater at 20℃;

[0018] S1.5. Calculate the salt density (SDD) based on the salinity of the wastewater:

[0019] SDD = S a ·V / A

[0020] Where V is the volume of the waste liquid, and A is the area of ​​the cleaning zone;

[0021] S1.6. Based on the salt density, classify the degree of contamination in each sub-sector.

[0022] Further, in step S1.3, the volumetric conductivity σ of the wastewater at room temperature θ is calculated. θ ,include:

[0023] σ 20 =σ θ [1-b(θ-20)]

[0024] Where b is the temperature coefficient.

[0025] Furthermore, the degree of contamination in step S1.6, which divides each sub-sector into different areas, includes:

[0026] When SDD is 0 mg / cm 2 ~0.03mg / cm2 When the SDD is 0.03 mg / cm³, it is classified as level I dirtiness; 2 ~0.06mg / cm 2 When the SDD is 0.06 mg / cm³, it is classified as dirtiness level II; 2 ~0.10mg / cm 2 When the SDD is >0.10 mg / cm³, it is classified as dirtiness level III; 2 If the level is low, it is classified as level IV of filth.

[0027] Furthermore, step S2 specifically includes:

[0028] S2.1 Select n sets of hyperspectral spectral line information, each set of hyperspectral spectral line information includes 256 wavelength data variables;

[0029] S2.2. The Monte Carlo sampling method is adopted, and the number of sampling times is set to N. Each time, 70% of the data is randomly extracted as the training set and the remaining 30% is used as the prediction set to construct the PLS model. The absolute value weight of the regression coefficient in the PLS model is calculated in each sampling process.

[0030] S2.3. Use the exponential decay function to remove wavelength variables whose absolute values ​​in the regression coefficient weights are less than the weight threshold;

[0031] S2.4. During each sampling, an adaptive reweighted sampling method is used to select a number of variables, R. i A PLS model is constructed using ×n wavelength variables, and the root mean square of the cross-validation is calculated.

[0032] S2.5 After N Monte Carlo samplings, N candidate feature wavelength subsets and their corresponding cross-validation root mean square values ​​are obtained. The feature wavelength subset corresponding to the minimum cross-validation root mean square value is selected as the feature wavelength.

[0033] Further, step S2.2 calculates the absolute value weights of the regression coefficients in the PLS model for each sampling process, including:

[0034]

[0035] Among them, |b i | represents the absolute value of the regression coefficient of the i-th variable, w i is the absolute value weight of the regression coefficient of the i-th variable, and m is the number of variables remaining in each sampling.

[0036] Further, in step S2.3, the proportion of wavelength variables retained, calculated based on the exponential decay function, includes:

[0037] R i =μe-ki

[0038]

[0039] Where μ and k i R is a constant, where n is the number of original wavelength variables, taken as 256; i When establishing the PLS model based on Monte Carlo sampling for the i-th time, the proportion of wavelength variables retained is obtained from the EDF.

[0040] Furthermore, step S3 specifically includes:

[0041] S3.1 Calculate the average gray value of each sub-sector region under each characteristic wavelength;

[0042] S3.2 Obtain the actual degree of pollution of M regions and the average gray value of M regions under N characteristic wavelengths to obtain M sets of vectors. Each set of vectors consists of N average gray value data and the corresponding actual degree of pollution. Among them, the degree of pollution I, degree of pollution II, degree of pollution III, and degree of pollution IV are denoted as C1, C2, C3, and C4, respectively.

[0043] S3.3. Use M sets of vectors as input to the random forest model, set the number of decision trees to 100, and the test set to account for 20% to train the random forest classification model.

[0044] S3.4 Calculate the prediction error rate of the random forest model using out-of-bag data to obtain the out-of-bag error, denoted as OOBerr1;

[0045] S3.5. Add noise interference to specific features of all samples in the out-of-bag data, that is, randomly change the value of the sample at the wavelength of that feature, and recalculate the out-of-bag error, denoted as OOBerr2;

[0046] S3.6. Calculate the importance P of the interfered characteristic wavelength based on OOBerr1 and OOBerr2;

[0047]

[0048] Where N0 is the number of decision trees in the random forest;

[0049] S3.7 Calculate the mean of the gray values ​​of C1, C2, C3 and C4 under the features with the highest importance, and use the four mean values ​​as the gray thresholds for dirtiness level I, dirtiness level II, dirtiness level III and dirtiness level IV, respectively.

[0050] Furthermore, step S4 specifically includes:

[0051] S4.1. Using the binary segmentation method, the regions D1, D2, D3 and D4 are obtained by segmenting according to the gray level thresholds corresponding to the degree of filth I, degree of filth II, degree of filth III and degree of filth IV.

[0052] S4.2. Based on the ratio of the total number of pixels in each region to the total number of pixels in the composite insulator disk image, obtain the area proportion P of each region. D1 P D2 P D3 P D4 ;

[0053] Among them, the area proportion of the region with pollution level I is P1 = P D1 -P D2 The area of ​​pollution level II accounts for P2 = P D2 -P D3 The area of ​​pollution level II accounts for P3 = P D3 -P D4 The area proportion of contamination level IV is P4 = P D4 .

[0054] Furthermore, step S5 specifically includes: assigning different weights to pixels with gray values ​​greater than different pollution level gray thresholds on the grayscale image of the composite insulator disk surface, and then performing pseudo-colorization processing on the image, that is, the degree of pollution in the region is proportional to the color depth.

[0055] The insulator pollution level visualization method based on hyperspectral images provided by this invention has the following beneficial effects:

[0056] This invention comprehensively utilizes hyperspectral image data to more accurately and efficiently detect surface contamination on insulators. It reduces data processing volume and simplifies visualization processes. In terms of detection results, it achieves non-contact online detection of insulator contamination distribution, providing guidance for insulator cleaning and improving the reliability and safety of transmission line operation. Attached Figure Description

[0057] Figure 1 This is a flowchart of a method for visualizing the pollution level of insulators based on hyperspectral images. Detailed Implementation

[0058] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0059] Example 1

[0060] refer to Figure 1 The insulator contamination visualization method based on hyperspectral images in this embodiment utilizes hyperspectral spectral lines and image information, and visualizes the contamination distribution on the insulator surface through image features. Specifically, it includes the following steps:

[0061] Step S1: Divide the naturally contaminated composite insulator disc surface into several fan-shaped areas of equal area, acquire hyperspectral images of the composite insulator disc surface to obtain a hyperspectral full-band image of the composite insulator disc surface, and clean each fan-shaped area to obtain its actual degree of contamination.

[0062] Step S2: Perform dimensionality reduction data processing on the hyperspectral full-band image and extract several feature wavelengths from the bands contained in the hyperspectral data.

[0063] Step S3: Calculate the average gray value of the sector region corresponding to each feature wavelength. Based on the average gray value, use the random forest algorithm to select the optimal feature wavelength to construct the correspondence between image gray value and degree of filth, and calculate the gray value threshold of the degree of filth.

[0064] Step S4: Divide the composite insulator disk surface into regions with different levels of contamination based on the grayscale threshold, and calculate the area ratio of each contamination level region.

[0065] Step S5: Overlay images of areas with different levels of contamination and use color to distinguish the degree of contamination on the insulator disc surface.

[0066] This embodiment makes full use of the information contained in hyperspectral images, selects feature wavelengths using spectral line data, and then selects the optimal feature wavelengths using a random forest algorithm to determine the most representative grayscale image. The information contained in the grayscale image is then used to detect and visualize the degree of contamination.

[0067] Example 2

[0068] This embodiment is a further explanation of Embodiment 1, and based on hyperspectral spectral lines and image information, it realizes the visualization of contamination distribution on the insulator surface through image features. Specifically, it includes the following steps:

[0069] Step S1: Divide the naturally contaminated composite insulator disc surface into several fan-shaped areas of approximately equal area, ensuring that the degree of contamination is roughly the same in each area. Obtain the actual degree of contamination in each area through cleaning, and then acquire hyperspectral images of the composite insulator disc surface. The specific steps include:

[0070] Step S1.1: Using a portable hyperspectral imager with a tripod, select a suitable angle and use a halogen lamp to simulate natural lighting environment in a dark room to collect hyperspectral image information of the composite insulator string with natural pollution accumulation.

[0071] Step S1.2: Using the center of the composite insulator disc as the center, the composite insulator disc is approximately divided into several sector-shaped areas of equal area. Each area is then divided into several smaller sector-shaped areas, i.e., sub-sector-shaped areas, so that the degree of pollution in each sub-sector-shaped area is roughly the same.

[0072] Step S1.3: Use a soft brush and distilled water to collect dirt from each area of ​​the insulator panel surface, and measure the volume conductivity σ of the cleaned dirt solution at room temperature θ. θ (S / m):

[0073] σ 20 =σ θ [1-b(θ-20)]

[0074] The volumetric conductivity σ of the wastewater at 20℃ was obtained. 20 In the formula, b is the temperature coefficient, which can be obtained from a table;

[0075] Step S1.4: Calculate the salinity S of the wastewater based on its volumetric conductivity. a (kg / m 3 ):

[0076] S a =(5.7σ) 20 ) 1.03

[0077] Step S1.5: Calculate the salt density SDD (mg / cm³) based on the salinity of the wastewater. 2 ):

[0078] SDD = S a ·V / A

[0079] Wherein, V(cm) 3 A(cm³) represents the volume of the waste liquid. 2 () represents the area of ​​the cleaning zone;

[0080] Step S1.6: Based on the salt density, classify the degree of pollution in each area, specifically as follows:

[0081] When SDD is 0 mg / cm 2 ~0.03mg / cm 2 When the SDD is 0.03 mg / cm³, it is classified as level I dirtiness; 2 ~0.06mg / cm 2When the SDD is 0.06 mg / cm³, it is classified as dirtiness level II; 2 ~0.10mg / cm 2 When the SDD is >0.10 mg / cm³, it is classified as dirtiness level III; 2 If the level is low, it is classified as level IV of filth.

[0082] This step involves dividing each insulator into eight sector-shaped regions along its radius, and then dividing each of these eight sectors in half along a direction perpendicular to the radius, resulting in 16 approximately equal-sized regions on each insulator. The boundaries of the selected regions are marked on the insulator surface. Clean cotton or a small brush is then used to apply deionized water to the selected regions on the insulator surface. The dirt is collected, and the surface is wiped repeatedly, with the cotton or brush immersed in deionized water for 30-60 seconds each time to ensure the dirt is completely submerged.

[0083] This step shields the interference of other light sources and uses halogen lamps to better simulate the effect of natural light under outdoor conditions. The actual degree of pollution in each area of ​​the insulator disc is determined by the equivalent salt density method, which serves as the input for the subsequent classification model and is used to train the classification model.

[0084] Step S2: Perform dimensionality reduction on the acquired hyperspectral full-band images. Use the Competitive Adaptive Reweighting (CARS) algorithm to extract several feature wavelengths from the 256 bands contained in the hyperspectral data, retaining the information corresponding to the feature wavelengths and removing redundant information from other wavelengths. This specifically includes the following steps:

[0085] Step S2.1: Select several sets of hyperspectral spectral line information, denoted as n, with each set containing 256 wavelength data variables;

[0086] Step S2.2: Use Monte Carlo sampling method, set the number of samplings to N, and randomly select 70% of the data from n groups each time as the training set, and the remaining 30% as the prediction set to build a partial least squares regression (PLS) model.

[0087] Calculate the absolute value weights of the regression coefficients in the PLS model for each sampling process:

[0088]

[0089] Among them, |b i | represents the absolute value of the regression coefficient of the i-th variable, w i is the absolute value weight of the regression coefficient of the i-th variable, and m is the number of variables remaining in each sampling;

[0090] Step S2.3: Remove w using the exponential decay function (EDF). iA relatively small wavelength variable; when establishing the PLS model based on Monte Carlo sampling for the i-th time, the proportion R of the wavelength variable retained is obtained from the EDF. i :

[0091] R i =μe -ki

[0092]

[0093] Where n is the number of original wavelength variables, which is 256; μ and k are constants;

[0094] Step S2.4: During each sampling, select a number of variables, R, from the number of variables in the previous sampling using Adaptive Reweighted Sampling (ARS). i PLS modeling is performed using ×n wavelength variables, and the cross-validation root mean square (RMSECV) is calculated.

[0095] Step S2.5: After N Monte Carlo samplings, N candidate feature wavelength subsets and their corresponding RMSECV values ​​are obtained. The wavelength variable subset corresponding to the minimum RMSECV value is selected as the feature wavelength.

[0096] In one embodiment of the present invention, the acquired hyperspectral image is first preprocessed, including black-and-white correction and multivariate scattering correction. The black-and-white correction is as follows:

[0097]

[0098] Where R is the corrected hyperspectral data, R S For the raw hyperspectral data of the sample, R D Hyperspectral data for calibration in complete black was obtained by capturing images with the lens covered. R W To calibrate hyperspectral data with a white background, a standard calibration white plate is placed when photographing the sample.

[0099] The formula for multivariate scattering correction is:

[0100]

[0101]

[0102]

[0103] The average value of the hyperspectral data is calculated according to formula (1), where n represents the number of sample points in the image data, f represents the number of samples, f = 1, 2, ..., n; d represents the d-th band, d = 1, 2, ..., 256; A f,d This represents the reflectance value at the d-th band of the f-th sample data in the original hyperspectral data matrix. The matrix represents the average values ​​of f sample data; a univariate regression is performed on the hyperspectral data according to equation (2), where A f Let m represent the original hyperspectral data of the f-th sample. f b represents the relative offset coefficient after linear regression between the original hyperspectral data and the average spectral data. f The translation amount is given; multivariate scattering correction is performed on the hyperspectral data according to equation (3), where A f(MSC) This represents the data matrix after multivariate scattering correction.

[0104] This step filters the 256 bands contained in the hyperspectral image, retaining the data corresponding to the characteristic wavelengths and removing other redundant data, which greatly reduces the amount of data processing and has a negligible impact on the detection accuracy.

[0105] Step S3: Take the grayscale image at the feature wavelength, obtain the average grayscale value of each region, select the optimal feature using the random forest algorithm, and construct the correspondence between image grayscale values ​​and the degree of dirtiness using the grayscale image features at the optimal feature wavelength. This specifically includes the following steps:

[0106] Step S3.1: The hyperspectral image contains grayscale images under 256 wavelengths. Take the grayscale image corresponding to the characteristic wavelength determined in step S2, and calculate the average grayscale value of each region under each characteristic wavelength according to the region divided in step S1.

[0107] Step S3.2: Obtain the actual degree of contamination of M regions and the average gray value of each region at N characteristic wavelengths to obtain M sets of vectors. Each set of vectors consists of N average gray value data and the corresponding actual degree of contamination. The degree of contamination I, degree of contamination II, degree of contamination III, and degree of contamination IV are denoted as C1, C2, C3, and C4, respectively.

[0108] Step S3.3: Use the original vector set as input to the random forest model, set the number of decision trees to 100, and the test set to account for 20% to train the random forest classification model.

[0109] Step S3.4: In a single sampling, approximately 1 / 3 of the data is not utilized and is called out-of-bag (OOB) data. The prediction error rate of the model is calculated using the out-of-bag data to obtain the out-of-bag error, denoted as OOBerr1.

[0110] Step S3.5: Add noise interference to specific features of all samples in the out-of-bag data, that is, randomly change the value of the sample at the wavelength of that feature, and recalculate the out-of-bag error, denoted as OOBerr2;

[0111] Step S3.6: Calculate the importance P of the interference features based on OOBerr1 and OOBerr2.

[0112]

[0113] Where N0 is the number of decision trees in the random forest, which is set to 100;

[0114] Step S3.7: Based on the calculation results of step S3.6, denote the feature with the highest importance as x. m All C1 class regions in x m The average of the grayscale values ​​is then taken as the grayscale threshold for the degree of dirtiness I.

[0115] Similarly, calculate the mean of the gray values ​​of C2, C3 and C4 under the features with the highest importance, and use the three mean values ​​as the gray thresholds for dirtiness level II, dirtiness level III and dirtiness level IV, respectively.

[0116] This step uses the change in out-of-bag data error to calculate the impact of changes in each feature on the detection accuracy, thereby ranking all features by importance to select the optimal feature wavelength; the grayscale image corresponding to the optimal feature wavelength is more representative; the average grayscale value of each level of contamination region at the optimal feature wavelength is used as the grayscale threshold of that level of contamination, thus establishing the correspondence between the level of contamination and the grayscale value of the hyperspectral image.

[0117] Step S4: Determine the grayscale values ​​corresponding to different levels of contamination. By dividing the grayscale value thresholds, the extraction of regions with different levels of contamination on the insulator disc surface and the calculation of the area ratio of each level of contamination region are realized. This specifically includes the following steps:

[0118] Step S4.1: Using the binary segmentation method, based on the grayscale threshold determined in step S3, segment the insulator disk surface in the x... m The grayscale image below is processed; it is segmented using grayscale thresholds for dirtiness levels I, II, III, and IV to obtain regions D1, D2, D3, and D4.

[0119] Step S4.2: Based on the ratio of the total number of pixels in each region to the total number of pixels in the composite insulator disk image, obtain the area proportion P of each region. D1 P D2 P D3 P D4 ;

[0120] Among them, the area proportion of the region with pollution level I is P1 = P D1 -P D2 The area of ​​pollution level II accounts for P2 = P D2-P D3 The area of ​​pollution level II accounts for P3 = P D3 -P D4 The area proportion of contamination level IV is P4 = P D4 .

[0121] Step S5: Overlay images of areas with different levels of contamination, using color depth to differentiate the severity of contamination, thus visually displaying the contamination distribution on the insulator panel. This specifically includes:

[0122] Different weights are assigned to pixels with gray values ​​greater than gray thresholds for different levels of contamination on the grayscale image of the insulator disc surface. Then, pseudo-colorization processing is performed on the image, and the darker the color is displayed in areas with heavier contamination, thus realizing a visual display of the contamination distribution on the insulator disc surface.

[0123] This step segments regions with different levels of contamination based on grayscale thresholds, calculates the area ratio of each region, and visualizes the distribution of regions with different levels of contamination, providing a more intuitive and clear reflection of the contamination distribution on the insulator surface.

[0124] Although specific embodiments of the invention have been described in detail with reference to the accompanying drawings, this should not be construed as limiting the scope of protection of this patent. Various modifications and variations that can be made by a person skilled in the art without inventive effort within the scope described in the claims still fall within the scope of protection of this patent.

Claims

1. A method for visualizing the degree of contamination of an insulator based on hyperspectral images, characterized by, The method comprises the following steps: S1, divide the natural pollution composite insulator disc surface into several equal area sectors, collect the hyperspectral image of the composite insulator disc surface, obtain the hyperspectral full wave band image of the composite insulator disc surface, and clean each sector to obtain the actual pollution degree; S2, perform data dimension reduction processing on the hyperspectral full wave band image, and extract several characteristic wavelengths from the wave bands contained in the hyperspectral data; S3, calculate the average value of the gray value of each sector corresponding to each characteristic wavelength, select the optimal characteristic wavelength based on the average value of the gray value by using the random forest algorithm, construct the corresponding relationship between the image gray value and the pollution degree, and calculate the gray threshold value of the pollution degree; the step S3 specifically comprises: S3.1, calculate the average value of the gray value of each sub-sector at each characteristic wavelength; S3.2, obtain the actual pollution degree of M regions and the average value of the gray value of M regions at N characteristic wavelengths, obtain M groups of vectors, each group of vectors is composed of N average value data of the gray value and the corresponding actual pollution degree, wherein the pollution degree I, the pollution degree II, the pollution degree III and the pollution degree IV are respectively denoted as C1, C2, C3 and C4; S3.3, take the M groups of vectors as the input of the random forest model, set the number of decision trees to 100, and set the test set ratio to 20%, train the random forest classification model; S3.4, calculate the prediction error rate of the random forest model by using the out-of-bag data, obtain the out-of-bag data error, denoted as OOBerr1; S3.5, add noise interference to the specific features of all samples of the out-of-bag data, that is, randomly change the value of the sample at the characteristic wavelength, calculate the out-of-bag error again, denoted as OOBerr2; S3.

6. Calculate the importance of the disturbed characteristic wavelength according to OOBerrl and OOBerr2 P； wherein, is the number of decision trees in the random forest; S3.7, calculate the average value of the average value of the gray value of C1, C2, C3 and C4 at the most important characteristic, and take the four average values obtained by calculation as the gray threshold values of the pollution degree I, the pollution degree II, the pollution degree III and the pollution degree IV respectively; S4, divide the composite insulator disc surface into different pollution degree regions according to the gray threshold value, and calculate the area ratio of each pollution degree region; S5, superimpose the images of different pollution degree regions, and use color to distinguish the pollution degree of the insulator disc surface.

2. The method for visualizing the contamination level of an insulator based on hyperspectral images according to claim 1, characterized in that, The step S1 specifically comprises: S1.1, collect the hyperspectral spectral information of the natural pollution composite insulator string in the simulated natural light environment; S1.2, divide the composite insulator disc surface into several equal area sectors with the center of the composite insulator disc surface as the center, and then divide each sector into several sub-sectors; S1.3, washing the contamination in each sub-fan-shaped area, collecting the washed contamination liquid, and calculating the volume conductivity of the contamination liquid at room temperature θ ;​ S1.

4. Calculate the salinity of the contaminated liquid from the volume conductivity of the contaminated liquid : wherein, is the volume conductivity of the contaminated liquid at 20°C; S1.

5. Calculate the density of the attached salt according to the salinity of the contaminated liquid SDD : wherein V is the volume of the contaminated liquid, A is the area of the cleaning zone; S1.6, divide the pollution degree of each sub-sector according to the salt deposit density.

3. The method for visualizing the contamination level of an insulator based on hyperspectral images according to claim 2, characterized in that, In step S1.3, the temperature of the wastewater at room temperature is calculated. θ The volume conductivity of the wastewater below ,include: wherein b is the temperature coefficient.

4. The method for visualizing the contamination level of the insulator based on the hyperspectral image according to claim 2, wherein, The step S1.6 of dividing the pollution degree of each sub-sector comprises: When SDD is 0 mg / cm 2 ~0.03 mg / cm 2 then it is classified as degree of contamination I; when SDD is 0.03 mg / cm 2 ~0.06 mg / cm 2 then it is classified as degree of contamination II; when SDD is 0.06 mg / cm 2 ~0.10 mg / cm 2 then it is classified as degree of contamination III; when SDD > 0.10 mg / cm 2 then it is classified as degree of contamination IV.

5. The method for visualizing the contamination level of an insulator based on hyperspectral images according to claim 2, wherein, The step S2 specifically comprises: S2.1, select n groups of hyperspectral spectral line information, each group of hyperspectral spectral line information comprising 256 wavelength data variables; S2.2, using Monte Carlo sampling method, setting the sampling number as N, randomly extracting 70% from the data as the training set and the remaining 30% as the prediction set each time to construct the PLS model and calculate the absolute value weight of the regression coefficient in the PLS model in each sampling process; S2.3, using exponential decay function to remove the wavelength variables less than the weight threshold in the absolute value weight of the regression coefficient; S2.4, at each sampling, the adaptive reweighted sampling is used to select the number of × n wavelength variables for PLS model construction, and the cross-validation root mean square is calculated; wherein, is the i EDF is used to obtain the proportion of retained wavelength variables when the PLS model is established based on the Monte Carlo sampling for the th time. S2.5, after N times of Monte Carlo sampling, N groups of candidate characteristic wavelength subsets and the corresponding cross-validation root mean square values are obtained, and the characteristic wavelength subset corresponding to the minimum cross-validation root mean square value is selected as the characteristic wavelength.

6. The method for visualizing the contamination level of an insulator based on hyperspectral images according to claim 5, wherein, The step S2.2 includes: wherein, is the absolute value of the regression coefficient of the i th variable, is the absolute value of the regression coefficient of the i th variable weight, m is the number of remaining variables in each sampling.

7. The method for visualizing the contamination level of an insulator based on hyperspectral images according to claim 5, wherein, The step S2.3 includes: wherein, μ and k i is a constant, n is a raw wavelength variable number, taken as 256.

8. The method for visualizing the contamination level of an insulator based on hyperspectral images according to claim 7, characterized in that, The step S4 specifically includes: S4.1, using binary segmentation method, segmenting according to the gray threshold values corresponding to the contamination degree I, the contamination degree II, the contamination degree III and the contamination degree IV to obtain the regions D1, D2, D3 and D4; S4.2, according to the proportion of the total number of pixels in the region and the total number of pixels of the composite insulator disc image, the area proportion of each region is obtained ; The area proportion of the region with the dirt degree I is The area proportion of the region with the dirt degree II is The area proportion of the region with the dirt degree II is The area proportion of the region with the dirt degree IV is .

9. The method for visualizing the contamination level of an insulator based on hyperspectral images according to claim 8, characterized in that, The step S5 specifically includes: assigning different weights to the pixel points with gray values greater than the gray threshold values of different contamination degrees on the composite insulator disc surface gray scale image, and then performing pseudo-color processing on the image, that is, the contamination degree in the region is proportional to the color depth.

Images