CN101403676A

CN101403676A - Insulator hydrophobicity rank amalgamation judging method based on D-S evidence theory

Info

Publication number: CN101403676A
Application number: CNA200810225074XA
Authority: CN
Inventors: 唐良瑞; 祁兵; 杨秋霞; 张晶; 龚钢军; 孙毅
Original assignee: North China Electric Power University
Current assignee: North China Electric Power University
Priority date: 2008-10-28
Filing date: 2008-10-28
Publication date: 2009-04-08
Anticipated expiration: 2028-10-28
Also published as: CN101403676B

Abstract

The invention belongs to the technical field of insulation detection of high-voltage equipment in a power system, and discloses an insulator hydrophobic grade fusion judgment method based on D-S evidence theory. The technical solution is: use the water spray classification method to collect image information related to hydrophobicity, perform cropping, denoising and segmentation processing, and then extract three hydrophobicity characteristics based on the water drop/water mark optical characteristics and shape information of the hydrophobic image. Feature components; then use the extracted three feature components of area, shape and bright spot as the features of the three classifiers respectively, and use the fuzzy rules to find out the basic probability distribution function of each classifier, finally according to the synthesis of D-S evidence theory The rule fuses the results of the three classifiers for judgment. The invention can effectively improve the recognition accuracy and reliability of the hydrophobicity, overcome the defects of the existing methods in terms of accuracy, reliability and practicability, and meet the needs of on-line detection of the hydrophobicity of the insulator.

Description

Insulator Hydrophobic Grade Fusion Judgment Method Based on D-S Evidence Theory

技术领域 technical field

本发明属于电力系统高压设备绝缘检测的技术领域，尤其涉及一种基于D-S证据理论的绝缘子憎水性等级融合判决方法。The invention belongs to the technical field of insulation detection of high-voltage equipment in a power system, and in particular relates to an insulator hydrophobicity level fusion judgment method based on D-S evidence theory.

背景技术 Background technique

随着绝缘子污闪事故的不断发生，绝缘子安全运行已成为电力系统中的一个重要环节。绝缘子憎水性是影响绝缘子污闪的重要因素，所以研究绝缘子憎水性等级判决刻不容缓。传统憎水性检测方法多采用人工方式进行图像采集和判决，效率低下且浪费了大量人力资源；同时，由于人的主观性不可避免地会出现不一致性。基于图像分析的憎水性在线检测，使得自动判决成为可能。目前部分电力系统部门已采用在线检测来代替人工操作。但现有的方法对现场运行环境下憎水性检测准确性得不到保证，且现有的方法大多单纯提取一个方面的特征分量来识别憎水性等级，缺乏对多种信息的协同处理和综合利用。由于这些方法所利用的憎水性特征量单一，如果这些特征量受各种噪声干扰影响，可能降低憎水性识别精度和可靠性，同时在准确性、可靠性和实用性等方面都存在着不同程度的缺陷。With the continuous occurrence of insulator pollution flashover accidents, the safe operation of insulators has become an important link in the power system. The hydrophobicity of insulators is an important factor affecting the pollution flashover of insulators, so it is urgent to study the judgment of insulators' hydrophobicity level. Traditional hydrophobicity detection methods mostly use manual methods for image acquisition and judgment, which is inefficient and wastes a lot of human resources; at the same time, inconsistencies will inevitably occur due to human subjectivity. On-line detection of hydrophobicity based on image analysis makes automatic judgment possible. At present, some power system departments have adopted online detection instead of manual operation. However, the existing methods cannot guarantee the accuracy of hydrophobicity detection in the field operation environment, and most of the existing methods simply extract one characteristic component to identify the hydrophobicity level, and lack the collaborative processing and comprehensive utilization of various information . Because these methods use a single hydrophobic characteristic, if these characteristic quantities are affected by various noise interference, the accuracy and reliability of hydrophobic recognition may be reduced, and there are different degrees of accuracy, reliability and practicability. Defects.

发明内容 Contents of the invention

本发明的目的在于，提供一种基于D-S证据理论的绝缘子憎水性等级融合判决方法，通过提取三个方面的特征分量来识别憎水性等级，避免了由于某些特征分量受到各种噪声干扰时，影响憎水性识别精度、可靠性和准确性的缺陷，使绝缘子憎水性检测的准确性得到了极大地提高。The purpose of the present invention is to provide a method for fusion judgment of insulator hydrophobicity level based on D-S evidence theory, which can identify the hydrophobicity level by extracting three aspects of characteristic components, avoiding the interference of various noises due to certain characteristic components. The defects that affect the accuracy, reliability and accuracy of hydrophobicity identification have greatly improved the accuracy of insulator hydrophobicity detection.

本发明的技术方案是，一种基于D-S证据理论的绝缘子憎水性等级融合判决方法，其特征是所述方法包括下列步骤：The technical solution of the present invention is a method for judging the level fusion of insulator hydrophobicity based on D-S evidence theory, characterized in that the method includes the following steps:

步骤1：利用喷水分级法获取待测绝缘子憎水性图像；Step 1: Use the water spray classification method to obtain the hydrophobicity image of the insulator to be tested;

步骤2：在喷水结束后1-2秒的时间范围内，对待测绝缘子的憎水性图像进行裁剪；Step 2: Cut out the hydrophobicity image of the insulator to be tested within 1-2 seconds after the water spraying ends;

步骤3：对裁剪后的憎水性图像进行直方图均衡处理；Step 3: Perform histogram equalization processing on the cropped hydrophobic image;

步骤4：对均衡处理后的图像进行去噪处理；Step 4: denoising the equalized image;

步骤5：利用信息测度的方法，确定步骤1至步骤4处理后憎水性图像的三个信息测度分量，基于模糊熵的邻域有序性C(i，j)、基于梯度及模糊熵的方向性M(i，j)和基于相似度的分量R(i，j)；其中(i，j)是憎水性图像的像素点；Step 5: Using the method of information measurement, determine the three information measurement components of the hydrophobic image processed in steps 1 to 4, the neighborhood order C(i, j) based on fuzzy entropy, and the direction based on gradient and fuzzy entropy M(i, j) and similarity-based component R(i, j); where (i, j) is the pixel of the hydrophobic image;

步骤6：利用模糊C均值聚类的图像分析方法，实现憎水性图像背景和物体的分割；Step 6: Utilize the image analysis method of fuzzy C-means clustering to realize the segmentation of the hydrophobic image background and objects;

步骤7：依据憎水性图像光学特性、形状信息，提取憎水性图像三个特征分量，面积特征

形状特征

亮点特征

Step 7: According to the optical characteristics and shape information of the hydrophobic image, extract three feature components of the hydrophobic image, the area feature

shape feature

Highlight Features

步骤8：将提取的面积特征、形状特征及亮点特征分量分别作为三个特征分类器；Step 8: Use the extracted area feature, shape feature and bright spot feature components as three feature classifiers respectively;

步骤9：构建憎水性隶属度函数，用隶属度函数表示命题的可信度，利用模糊规则求出各分类器的基本概率分配函数；Step 9: Construct the hydrophobic membership function, use the membership function to represent the credibility of the proposition, and use the fuzzy rules to find the basic probability distribution function of each classifier;

步骤10：利用D-S证据理论的合成法则，将三个分类器的基本概率分配函数进行融合，得到融合后的基本概率分配函数；Step 10: Using the composition rule of D-S evidence theory, fuse the basic probability distribution functions of the three classifiers to obtain the basic probability distribution function after fusion;

步骤11：利用等级判定的基本原则确定憎水性等级，实现憎水性等级融合判决。Step 11: Use the basic principles of grade judgment to determine the hydrophobicity grade, and realize the fusion judgment of hydrophobicity grade.

所述利用喷水分级法获取待测绝缘子憎水性图像，其过程是，对待测绝缘子进行喷水操作，喷水结束后，拍摄动态录像，并采用从拍摄的动态录像中截取静像的方式来获取绝缘子的喷水图像。The process of obtaining the hydrophobicity image of the insulator to be tested by using the water spray classification method is that the insulator to be tested is sprayed with water, and after the water spraying is completed, a dynamic video is taken, and a still image is intercepted from the captured dynamic video. Acquire a water spray image of an insulator.

所述对待测绝缘子的憎水性图像进行裁剪，其方法是，在图像中选取不包括复合绝缘子伞裙边缘，且喷水区内憎水性最差的部分，利用图像处理软件进行裁剪。The method of clipping the hydrophobicity image of the insulator to be tested is to select the part in the image that does not include the shed edge of the composite insulator and has the worst hydrophobicity in the spray area, and use image processing software to clip.

所述对裁剪后的憎水性图像进行直方图均衡处理，其方法是，对每个像素点(i，j)实现憎水性图像自适应直方图均衡： $T (f (i, j)) = 255 \times Σ_{r = 0}^{k} \frac{n_{r}}{M^{2}},$ 其中n_r是灰度级为r的像素在所选滑动窗口M×M中的数量，f(i，j)为(i，j)像素点灰度值，k为f(i，j)所对应灰度级。The method of performing histogram equalization processing on the cropped hydrophobic image is to realize adaptive histogram equalization of the hydrophobic image for each pixel point (i, j): $T (f (i, j)) = 255 \times Σ_{r = 0}^{k} \frac{{no}_{r}}{m^{2}},$ Among them, n _r is the number of pixels with gray level r in the selected sliding window M×M, f(i, j) is the gray value of (i, j) pixel, k is f(i, j) Corresponds to grayscale.

所述对均衡处理后的图像进行去噪处理，是采用中值滤波的方法，强迫将受到干扰的像素转变为其邻近的像素的灰度值，达到去除干扰，消除噪声的效果。The denoising process on the image after the equalization process is to use a median filter method to force the disturbed pixel to be transformed into the gray value of its adjacent pixel, so as to achieve the effect of removing interference and eliminating noise.

所述利用模糊C均值聚类的图像分析方法，实现憎水性图像背景和物体的分割，其过程是，首先：模糊C均值聚类的图像分析算法为 $J (U, c_{1}, . . ., c_{c}) = Σ_{i = 1}^{c} J_{i} = Σ_{i = 1}^{c} Σ_{j}^{n} u_{ij}^{m} d_{ij}^{2},$ 而 $c_{i} = \frac{Σ_{j = 1}^{n} u_{ij}^{m} x_{j}}{Σ_{j = 1}^{n} u_{ij}^{m}},$ $u_{ij} = \frac{1}{Σ_{k = 1}^{c} {(\frac{d_{ij}}{d_{kj}})}^{2 / (m - 1)}}$ 其中u_ij介于[01]间；c_i为模糊组i的聚类中心，d_ij为第i个聚类中心与第j个数据点间的欧几里德距离，且m∈[1，∞)是加权指数；然后再通过

进行计算，根据其是分类到c₁还是c₂来判断其是背景还是物体，其中C_ij、M_ij和R_ij分别为憎水性图像的三个信息测度分量C(i，j)、M(i，j)和R(i，j)。The image analysis method utilizing fuzzy C-means clustering is used to realize the segmentation of the hydrophobic image background and objects, and the process is, at first: the image analysis algorithm of fuzzy C-means clustering is

J (u, c_{1}, . . ., c_{c}) = Σ_{i = 1}^{c} J_{i} = Σ_{i = 1}^{c} Σ_{j}^{no} u_{ij}^{m} d_{ij}^{2},

and

c_{i} = \frac{Σ_{j = 1}^{no} u_{ij}^{m} x_{j}}{Σ_{j = 1}^{no} u_{ij}^{m}},

u_{ij} = \frac{1}{Σ_{k = 1}^{c} {(\frac{d_{ij}}{d_{kj}})}^{2 / (m - 1)}}

Where u _ij is between [01]; c _i is the cluster center of fuzzy group i, d _ij is the Euclidean distance between the i-th cluster center and the j-th data point, and m∈[1, ∞) is the weighting exponent; then by

Carry out calculations, and judge whether it is a background or an object according to whether it is classified into c ₁ or c ₂ , where C _ij , M _ij and R _ij are the three information measurement components C(i, j) and M( i,j) and R(i,j).

所述构建憎水性隶属度函数，其方法是，利用模糊规则，分别求出三个分类器的基本概率分配函数，对面积特征隶属度函数选取π型函数，对形状特征隶属度函数选取τ型函数，对亮点特征隶属度函数的选取依据泰勒级数展开式。The method of constructing the hydrophobic membership degree function is to use fuzzy rules to obtain the basic probability distribution functions of three classifiers respectively, select a π-type function for the area feature membership function, and select a τ-type function for the shape feature membership degree function. The selection of the membership function of bright spot features is based on the Taylor series expansion.

所述等级判定的基本原则包括：The basic principles of the grade determination include:

1)目标等级具有最大的信任度值；1) The target level has the largest trust value;

2)目标等级与其它等级的信任度值之差大于最大信任度值的25％；2) The difference between the trust value of the target level and other levels is greater than 25% of the maximum trust value;

3)不确定性区间长度小于最大信任度值的25％；3) The length of the uncertainty interval is less than 25% of the maximum confidence value;

4)目标等级的信任度值大于模糊性基本概率分配函数的值。4) The trust value of the target level is greater than the value of the fuzzy basic probability distribution function.

本发明的效果在于，采用三个特征分量对憎水性等级进行检测，与常用方法相比，检测参数多，反映憎水性变化全面；同时，采用变量的隶属度函数表示命题的可信度，表征憎水性特征分量的不确定性，具有较强的稳健性，最终提高了复合绝缘子憎水性在线检测的精度和准度。The effect of the present invention is that three characteristic components are used to detect the level of hydrophobicity. Compared with the common method, there are many detection parameters, which reflect the comprehensive change of hydrophobicity; at the same time, the membership function of the variable is used to represent the credibility of the proposition. The uncertainty of the characteristic component of hydrophobicity has strong robustness, which finally improves the accuracy and accuracy of online detection of hydrophobicity of composite insulators.

附图说明 Description of drawings

图1是基于D-S证据理论的绝缘子憎水性等级融合判决方法的流程图。Figure 1 is a flow chart of the insulator hydrophobicity level fusion judgment method based on the D-S evidence theory.

图2是憎水性特征分量提取的示意图。Fig. 2 is a schematic diagram of hydrophobic feature component extraction.

图3是基于D-S证据理论的憎水性等级融合判决框图。Fig. 3 is a block diagram of hydrophobicity level fusion judgment based on D-S evidence theory.

具体实施方式 Detailed ways

下面结合附图，对优选实施例作详细说明。应该强调的是，下述说明仅仅是示例性的，而不是为了限制本发明的范围及其应用。The preferred embodiments will be described in detail below in conjunction with the accompanying drawings. It should be emphasized that the following description is only exemplary and not intended to limit the scope of the invention and its application.

图1是基于D-S证据理论的绝缘子憎水性等级融合判决方法的流程图。通过分析，发现憎水性图像获取受喷水角度、拍摄角度、现场运行环境等影响，因此本方法采用下述规则来定位憎水性图像区域。图1中，步骤101利用喷水分级法获取待测绝缘子憎水性图像。在此采用拍摄动态录像，并从动态录像中截取静像的方式来获取。Figure 1 is a flow chart of the insulator hydrophobicity level fusion judgment method based on the D-S evidence theory. Through analysis, it is found that the acquisition of hydrophobic images is affected by water spray angle, shooting angle, and on-site operating environment, so this method adopts the following rules to locate hydrophobic image regions. In FIG. 1 , step 101 uses the water spray classification method to obtain an image of the hydrophobicity of the insulator to be tested. Here, a dynamic video is shot and a static image is captured from the dynamic video.

步骤102在喷水结束后1-2秒的时间范围内，对待测绝缘子的憎水性图像进行裁剪。在图像的裁剪过程中，目标区域的选取不能包括复合绝缘子伞裙的边缘，且必须是喷水区内憎水性最差的部分；裁剪通过自编图像处理软件实现。Step 102: Clipping the hydrophobicity image of the insulator to be tested within a time range of 1-2 seconds after the water spraying ends. In the process of cropping the image, the selection of the target area cannot include the edge of the shed of the composite insulator, and must be the part with the worst hydrophobicity in the water spray area; the cropping is realized by self-edited image processing software.

由于复合绝缘子所处环境复杂性，不可避免会受绝缘表面污秽影响，加上水的透明性导致的水珠/水迹与背景灰度差较小和水对光的反射导致的对光一侧的边界极为模糊等原因，致使图像信息获取困难。为此，需要对憎水性图像进行预处理，即步骤103对裁剪后的憎水性图像进行直方图均衡处理。方法是对每个像素点(i，j)实现憎水性图像自适应直方图均衡： $T (f (i, j)) = 255 \times Σ_{r = 0}^{k} \frac{n_{r}}{M^{2}},$ 其中n_r是灰度级为r的像素在所选滑动窗口M×M中的数量，f(i，j)为(i，j)像素点灰度值，k为f(i，j)所对应灰度级。Due to the complexity of the environment in which composite insulators are located, it is inevitable that they will be affected by the pollution of the insulating surface, coupled with the small difference between the water drops/water traces and the background gray level caused by the transparency of water, and the reflection of water on the light. The boundary is extremely blurred and other reasons, making it difficult to obtain image information. To this end, it is necessary to preprocess the hydrophobic image, that is, step 103 performs histogram equalization processing on the cropped hydrophobic image. The method is to implement adaptive histogram equalization of the hydrophobic image for each pixel point (i, j): $T (f (i, j)) = 255 \times Σ_{r = 0}^{k} \frac{{no}_{r}}{m^{2}},$ Among them, n _r is the number of pixels with gray level r in the selected sliding window M×M, f(i, j) is the gray value of (i, j) pixel, k is f(i, j) Corresponds to grayscale.

之后，执行步骤104对均衡后的图像进行去噪处理，即采用中值滤波的方法，强迫将受到干扰的像素转变为其邻近的像素的灰度值，达到去除干扰，消除噪声的效果。Afterwards, step 104 is executed to perform denoising processing on the equalized image, that is, the method of median filtering is used to forcibly transform the disturbed pixel into the gray value of its adjacent pixel, so as to achieve the effect of removing interference and noise.

由于憎水性图像所处环境复杂、水光相互作用的影响等，综合考虑图像的灰度、梯度及相似度的信息测度，有效地提取水珠/水迹与背景不同信息的测度，并引入基于模糊C均值聚类(FCM)的图像分析技术分割水珠/水迹与背景信息，实现憎水性图像分割。Due to the complex environment of hydrophobic images and the influence of water-light interaction, etc., comprehensively consider the information measurement of image grayscale, gradient and similarity, effectively extract the measurement of different information between water drops/water traces and background, and introduce the measurement based on The image analysis technology of fuzzy C-means clustering (FCM) segments water droplets/water traces and background information to realize hydrophobic image segmentation.

提取水珠/水迹与背景不同信息的测度，通过步骤105实现。步骤105利用信息测度的方法，确定步骤101至步骤104处理后憎水性图像的三个信息测度分量，基于模糊熵的邻域有序性C(i，j)、基于梯度及模糊熵的方向性M(i，j)和基于相似度的分量R(i，j)；其中(i，j)是憎水性图像的像素点。The measure of extracting the different information of the water drops/water traces and the background is realized through step 105 . Step 105 uses the method of information measurement to determine the three information measurement components of the hydrophobic image processed in steps 101 to 104, the neighborhood order C(i, j) based on fuzzy entropy, and the directionality based on gradient and fuzzy entropy M(i, j) and the similarity-based component R(i, j); where (i, j) is the pixel of the hydrophobic image.

步骤106利用模糊C均值聚类的图像分析方法，实现憎水性图像背景和物体的分割。其过程是，首先：模糊C均值聚类的图像分析算法为 $J (U, c_{1}, . . ., c_{c}) = Σ_{i = 1}^{c} J_{i} = Σ_{i = 1}^{c} Σ_{j}^{n} u_{ij}^{m} d_{ij}^{2},$ 而 $c_{i} = \frac{Σ_{j = 1}^{n} u_{ij}^{m} x_{j}}{Σ_{j = 1}^{n} u_{ij}^{m}},$ $u_{ij} = \frac{1}{Σ_{k = 1}^{c} {(\frac{d_{ij}}{d_{kj}})}^{2 / (m - 1)}}$ 其中u_ij介于[0 1]间；c_i为模糊组i的聚类中心，d_ij为第i个聚类中心与第j个数据点间的欧几里德距离，且m∈[1，∞)是加权指数；然后通过

进行计算，根据其是分类到c₁还是c₂来判断其是背景还是物体，其中C_ij、M_ij和R_ij分别为憎水性图像的三个信息测度分量C(i，j)、M(i，j)和R(i，j)。Step 106 utilizes the image analysis method of fuzzy C-means clustering to realize the segmentation of the background and object of the hydrophobic image. The process is, firstly: the image analysis algorithm of fuzzy C-means clustering is

J (u, c_{1}, . . ., c_{c}) = Σ_{i = 1}^{c} J_{i} = Σ_{i = 1}^{c} Σ_{j}^{no} u_{ij}^{m} d_{ij}^{2},

and

c_{i} = \frac{Σ_{j = 1}^{no} u_{ij}^{m} x_{j}}{Σ_{j = 1}^{no} u_{ij}^{m}},

u_{ij} = \frac{1}{Σ_{k = 1}^{c} {(\frac{d_{ij}}{d_{kj}})}^{2 / (m - 1)}}

Where u _ij is between [0 1]; c _i is the cluster center of fuzzy group i, d _ij is the Euclidean distance between the i-th cluster center and the j-th data point, and m∈[1 , ∞) is the weighted exponent; then by

图1中，步骤107依据憎水性图像光学特性、形状信息等，提取憎水性图像三个特征分量，面积特征

形状特征

亮点特征

图2是本发明的憎水性特征分量提取的示意图。图2中，201为所提取的三个憎水性特征分量。分析各特征分量与憎水性等级间的关系：随着憎水性等级变大(即由HC₁向HC₇变化)，面积特征K值逐渐变大，而不同等级间的形状特征f_c及亮点特征f_p均遵循一定的变化规律，表明这三个特征分量可以用于后续憎水性等级的判定。In Figure 1, step 107 extracts three feature components of the hydrophobic image based on the optical characteristics and shape information of the hydrophobic image, and the area feature

shape feature

Highlight Features

Fig. 2 is a schematic diagram of the hydrophobic feature component extraction of the present invention. In FIG. 2 , 201 is the extracted three hydrophobic feature components. Analyze the relationship between each characteristic component and the hydrophobicity grade: as the hydrophobicity grade increases (that is, from HC ₁ to HC ₇ ), the value of the area characteristic K gradually increases, while the shape characteristics f _c and bright spot characteristics between different grades f _p follows a certain change rule, indicating that these three characteristic components can be used to determine the subsequent hydrophobicity level.

图3是基于D-S证据理论的憎水性等级融合判决框图。图3中，301为三个分类器，302为模糊规则。憎水性等级融合判决使用所提取的面积、形状及亮点三个特征分量分别作为三个分类器的特征，因此，步骤108将提取的面积特征、形状特征及亮点特征分量分别作为三个特征分类器。Fig. 3 is a block diagram of hydrophobicity level fusion judgment based on D-S evidence theory. In Fig. 3, 301 is three classifiers, 302 is a fuzzy rule. Hydrophobic grade fusion judgment uses the extracted three feature components of area, shape and bright spot as the features of the three classifiers respectively, therefore, step 108 uses the extracted area feature, shape feature and bright spot feature component as the three feature classifiers respectively .

由于所处理的憎水性图像中提取的特征分量往往不确定，为表述与处理这种不确定性，本方法采用变量的隶属度函数表示命题的可信度，利用模糊规则求出各分类器的基本概率分配函数，最后利用D-S证据理论的合成法则将各分类器的结果结合起来进行统一判决，实现憎水性等级融合判决。Since the feature components extracted from the processed hydrophobic image are often uncertain, in order to express and deal with this uncertainty, this method uses the membership function of variables to represent the credibility of propositions, and uses fuzzy rules to obtain the The basic probability distribution function, and finally use the composition rule of D-S evidence theory to combine the results of each classifier to make a unified judgment, and realize the fusion judgment of hydrophobicity level.

D-S证据理论通过对不确定信息的描述采用区间估计的方法，在区分不知道与不确定方面显示出了很大的灵活性。当不同的分类器所提供的关于目标的报告发生冲突时，它可以通过“悬挂”在所有目标集上共有的概念(可信度)使得发生的冲突获得解决，并保障原来高可信度的结果比低可信度的结果加权要大。D-S evidence theory shows great flexibility in distinguishing unknown and uncertain by adopting the method of interval estimation for the description of uncertain information. When the reports about the target provided by different classifiers conflict, it can resolve the conflict by "hanging" the common concept (credibility) on all target sets, and guarantee the original high-confidence Results are weighted more heavily than those with low confidence.

D-S证据理论用辨别框架Ω＝{S₁，S₂，S₃}表示感兴趣的命题集，并定义了一个集函数m：2^Ω→[0，1]为辨别框架上的基本概率分配函数，该函数满足以下两个条件：DS evidence theory uses the discriminative frame Ω={S ₁ , S ₂ , S ₃ } to represent the proposition set of interest, and defines a set function m: 2 ^Ω →[0, 1] as the basic probability distribution function on the discriminative frame , the function satisfies the following two conditions:

$\{\begin{matrix} \underset{HC HC &SubsetEqual; &SubsetEqual; Ω Ω}{Σ Σ} m m ((HC HC)) = = 11 \\ m m ((φ φ)) = = 00 \end{matrix}$

对于任何命题集，信任函数Bel()和似真函数pl()定义为：For any proposition set, the belief function Bel() and plausibility function pl() are defined as:

$\{\begin{matrix} Bel bel ((HC HC)) = = \underset{B B &SubsetEqual; &SubsetEqual; HC HC}{Σ Σ} m m ((B B)) & &ForAll; &ForAll; HC HC &SubsetEqual; &SubsetEqual; Ω Ω \\ pl pl ((HC HC)) = = 11 - - Bel bel ((\overset{&OverBar; &OverBar;}{HC HC})) & &ForAll; &ForAll; HC HC &SubsetEqual; &SubsetEqual; Ω Ω \end{matrix} - - - - - - ((11))$

即HC的信任函数为HC中每个子集的信任度之和，表示命题成立的最小的不确定性支持程度；似真函数表示不否定命题HC成立的程度。[Bel(HC)，pl(HC)]表示了HC的不确定空间。That is, the trust function of HC is the sum of the trust degrees of each subset in HC, which indicates the minimum degree of uncertainty support for the establishment of the proposition; the plausibility function indicates the degree of non-denial of the establishment of the proposition HC. [Bel(HC), pl(HC)] represents the uncertainty space of HC.

图1中，步骤109在憎水性隶属度函数的构建中，对面积特征选取π型函数，对形状特征选取τ型函数，对亮点特征隶属度函数的选取依据泰勒级数展开式。In Fig. 1, in step 109, in the construction of the hydrophobic membership function, a π-type function is selected for the area feature, a τ-type function is selected for the shape feature, and the selection of the membership function for the bright spot feature is based on the Taylor series expansion.

建立憎水性七个等级所对应的面积特征隶属度函数，如下式所示：Establish the area feature membership function corresponding to the seven levels of hydrophobicity, as shown in the following formula:

${u u}_{K K} (({HC HC}_{11})) = = \{\begin{matrix} 11 - - 22 {((((K K - - 0.005 0.005)) / / 0.01 0.01))}^{22} & 00 \leq \leq K K \leq \leq 0.01 0.01 \\ 22 {((((K K - - 0.015 0.015)) / / 0.01 0.01))}^{22} & 0.01 0.01 < < K K \leq \leq 0.015 0.015 \\ 00 & else else \end{matrix}$

${u u}_{K K} ((H h {C C}_{22})) = = \{\begin{matrix} 22 ((((K K + + 0.005 0.005)) / / 0.03 0.03 {))}^{22} & 00 \leq \leq K K \leq \leq 0.01 0.01 \\ 11 - - 22 {((((K K - - 0.25 0.25)) / / 0.03 0.03))}^{22} & 0.01 0.01 < < K K \leq \leq 0.04 0.04 \\ 22 {((((K K - - 00 . . 055055)) / / 0.03 0.03))}^{22} & 0.04 0.04 < < K K \leq \leq 0.055 0.055 \\ 00 & else else \end{matrix}$

${u u}_{K K} (({HC HC}_{33})) = = \{\begin{matrix} 22 {((((K K + + 0.04 0.04)) / / 0.16 0.16))}^{22} & 00 \leq \leq K K \leq \leq 0.04 0.04 \\ 11 - - 22 {((((K K - - 0.12 0.12)) / / 0.16 0.16))}^{22} & 0.04 0.04 < < K K \leq \leq 0.2 0.2 \\ 22 {((((K K - - 0.28 0.28)) / / 0.16 0.16))}^{22} & 0.2 0.2 < < K K \leq \leq 0.28 0.28 \\ 00 & else else \end{matrix}$

${u u}_{K K} (({HC HC}_{44})) = = \{\begin{matrix} 22 {((((K K + + 0.05 0.05)) / / 00 . . 33))}^{22} & 00 \leq \leq K K \leq \leq 00 . . 11 \\ 11 - - 22 {((((K K - - 0.25 0.25)) / / 00 . . 33))}^{22} & 00 . . 11 < < K K \leq \leq 00 . . 44 \\ 22 {((((K K - - 00 . . 5555)) / / 00 . . 33))}^{22} & 0.4 0.4 < < K K \leq \leq 00 . . 5555 \\ 00 & else else \end{matrix}$

${u u}_{K K} (({HC HC}_{55})) = = \{\begin{matrix} 22 {((((K K - - 00 . . 22)) / / 00 . . 44))}^{22} & 0.2 0.2 \leq \leq K K \leq \leq 0.4 0.4 \\ 11 - - 22 {((((K K - - 00 . . 66)) / / 00 . . 44))}^{22} & 0.4 0.4 < < K K \leq \leq 0.8 0.8 \\ 22 {((((K K - - 11)) / / 00 . . 44))}^{22} & 0.8 0.8 < < K K \leq \leq 11 \\ 00 & else else \end{matrix}$

${u u}_{K K} (({HC HC}_{66})) = = \{\begin{matrix} 22 {((((K K - - 00 . . 725725)) / / 0.15 0.15))}^{22} & 0.725 0.725 < < K K \leq \leq 00 . . 88 \\ 11 - - 22 {((((K K - - 00 . . 875875)) / / 0.15 0.15))}^{22} & 00 . . 88 < < K K \leq \leq 00 . . 9595 \\ 22 {((((K K - - 11 . . 025025)) / / 0.15 0.15))}^{22} & 00 . . 9595 < < K K \leq \leq 11 \\ 00 & else else \end{matrix}$

${u u}_{K K} (({HC HC}_{77})) = = \{\begin{matrix} 22 {((((K K - - 0.925 0.925)) / / 0.05 0.05))}^{22} & 0.925 0.925 < < K K \leq \leq 0.95 0.95 \\ 11 - - 22 {((((K K - - 0.975 0.975)) / / 0.05 0.05))}^{22} & 0.95 0.95 < < K K \leq \leq 11 \\ 00 & else else \end{matrix}$

u_K(HC_j)(j＝1，2，3，4，5，6，7)七个值的大小决定了该憎水性图像隶属于相应等级的可能性的大小。例如u_K(HC₁)越大，说明该憎水性隶属于HC₁的概率越大，否则隶属于HC₁的概率越小。将它们七个值归一化后分别作为HC₁-HC₇的基本概率分配函数值：The seven values of u _K (HC _j ) (j=1, 2, 3, 4, 5, 6, 7) determine the possibility of the hydrophobic image belonging to the corresponding grade. For example, the larger the u _K (HC ₁ ), the greater the probability that the hydrophobicity belongs to HC ₁ , otherwise, the lower the probability that the hydrophobicity belongs to HC ₁ . Normalize these seven values as the basic probability distribution function values of HC ₁ -HC ₇ :

${m m}_{11} (({HC HC}_{j j})) = = \frac{{u u}_{K K} (({HC HC}_{j j}))}{{Σ Σ}_{j j = = 11}^{77} {u u}_{K K} (({HC HC}_{j j}))},, j j = = 1,2,3,4,5,6,7 1,2,3,4,5,6,7$

建立憎水性七个等级所对应的形状特征隶属度函数，如下式所示：Establish the membership function of the shape feature corresponding to the seven levels of hydrophobicity, as shown in the following formula:

${u u}_{{f f}_{c c}} (({HC HC}_{i i})) = = \{\begin{matrix} {e e}^{55 ln ln ((22 (({f f}_{c c} - - 0.7 0.7))))} & 00 \leq \leq {f f}_{c c} < < 0.7 0.7 \\ {e e}^{- - 1010 ln ln ((22 (({f f}_{c c} - - 0.7 0.7))))} & 0.7 0.7 \leq \leq {f f}_{c c} < < 0.8 0.8 \\ {e e}^{55 ln ln ((22 (({f f}_{c c} - - 11))))} & 0.8 0.8 \leq \leq {f f}_{c c} \leq \leq 11 \end{matrix},, i i = = 1,2,3 1,2,3$

${u u}_{{f f}_{c c}} (({HC HC}_{i i})) = = {e e}^{- - 22 ln ln ((22 {f f}_{c c}))},, i i = = 4,5,6 4,5,6$

${u u}_{{f f}_{c c}} (({HC HC}_{i i})) = = \{\begin{matrix} {e e}^{44 ln ln ((22 (({f f}_{c c} - - 0.75 0.75))))} & 00 \leq \leq {f f}_{c c} < < 0.75 0.75 \\ {e e}^{- - 44 ln ln ((22 (({f f}_{c c} - - 0.75 0.75))))} & 0.75 0.75 \leq \leq {f f}_{c c} \leq \leq 11 \end{matrix},, i i = = 77$

(j＝1，2，3，4，5，6，7)七个值的大小决定了该憎水性图像隶属于相应等级的可能性的大小。例如

越大，说明该憎水性隶属于HC₁的概率越大，否则隶属于HC₁的概率越小。将它们七个值归一化后分别作为HC₁-HC₇的基本概率分配函数值：

(j=1, 2, 3, 4, 5, 6, 7) The size of the seven values determines the possibility of the hydrophobic image belonging to the corresponding level. For example

The larger the , the greater the probability that the hydrophobicity belongs to HC ₁ , otherwise the lower the probability of belonging to HC ₁ . Normalize these seven values as the basic probability distribution function values of HC ₁ -HC ₇ :

${m m}_{22} (({HC HC}_{j j})) = = \frac{{u u}_{{f f}_{c c}} (({HC HC}_{j j}))}{{Σ Σ}_{j j = = 11}^{77} {u u}_{{f f}_{c c}} (({HC HC}_{j j}))},, j j = = 1,2,3,4,5,6,7 1,2,3,4,5,6,7$

建立憎水性七个等级所对应的亮点特征隶属度函数，如下式所示：Establish the membership function of bright spot features corresponding to the seven levels of hydrophobicity, as shown in the following formula:

${u u}_{{f f}_{p p}} (({HC HC}_{11})) = = \{\begin{matrix} \frac{{f f}_{p p}^{55}}{55!!} & 00 \leq \leq {f f}_{p p} < < 0.1 0.1 \\ \frac{{f f}_{p p} + + 1.4 1.4}{3030} & 0.1 0.1 \leq \leq {f f}_{p p} < < 0.7 0.7 \\ \frac{{f f}_{p p}^{22} / / 22 - - {f f}_{p p}^{44}}{44} & 0.7 0.7 \leq \leq {f f}_{p p} < < 0.8 0.8 \\ {f f}_{p p} - - \frac{{f f}_{p p}^{22}}{22} & 0.8 0.8 \leq \leq {f f}_{p p} \leq \leq 11 \end{matrix}$

${u u}_{{f f}_{p p}} (({HC HC}_{22})) = = \{\begin{matrix} \frac{{f f}_{p p}^{55}}{55!!} & 00 \leq \leq {f f}_{p p} < < 0.1 0.1 \\ \frac{{f f}_{p p} - - 2.2 2.2}{3030} & 0.1 0.1 \leq \leq {f f}_{p p} < < 0.7 0.7 \\ \frac{cos cos (({f f}_{p p}))}{33} & 0.7 0.7 \leq \leq {f f}_{p p} < < 0.8 0.8 \\ \frac{{e e}^{- - {f f}_{p p}}}{22} - - \frac{{f f}_{p p}^{33}}{33 \times \times 33!!} & 0.8 0.8 \leq \leq {f f}_{p p} \leq \leq 11 \end{matrix}$

${u u}_{{f f}_{p p}} (({HC HC}_{33})) = = \{\begin{matrix} \frac{{f f}_{p p}^{55}}{55!!} & 00 \leq \leq {f f}_{p p} < < 0.1 0.1 \\ \frac{{f f}_{p p} - - 77}{6060} & 0.1 0.1 \leq \leq {f f}_{p p} < < 0.7 0.7 \\ \frac{cos cos (({f f}_{p p}))}{22} & 0.7 0.7 \leq \leq {f f}_{p p} < < 0.8 0.8 \\ \frac{{e e}^{- - {f f}_{p p}}}{22} & 0.8 0.8 \leq \leq {f f}_{p p} \leq \leq 11 \end{matrix}$

${u u}_{{f f}_{p p}} (({HC HC}_{44})) = = \{\begin{matrix} {f f}_{p p} & 00 \leq \leq {f f}_{p p} < < 0.1 0.1 \\ \frac{{f f}_{p p} + + 1.125 1.125}{33} & 0.1 0.1 \leq \leq {f f}_{p p} < < 0.7 0.7 \\ \frac{cos cos (({f f}_{p p}))}{66} & 0.7 0.7 \leq \leq {f f}_{p p} < < 0.8 0.8 \\ \frac{{f f}_{p p}^{33}}{33 \times \times 33!!} & 0.8 0.8 \leq \leq {f f}_{p p} \leq \leq 11 \end{matrix}$

${u u}_{{f f}_{p p}} (({HC HC}_{55})) = = \{\begin{matrix} {f f}_{p p} & 00 \leq \leq {f f}_{p p} < < 0.1 0.1 \\ \frac{1.125 1.125 - - {f f}_{p p}}{33} & 0.1 0.1 \leq \leq {f f}_{p p} < < 0.7 0.7 \\ \frac{{f f}_{p p}^{22} / / 22 - - {f f}_{p p}^{44} / / 44}{22} & 0.7 0.7 \leq \leq {f f}_{p p} < < 0.8 0.8 \\ \frac{{f f}_{p p}^{33}}{44!!} & 0.8 0.8 \leq \leq {f f}_{p p} \leq \leq 11 \end{matrix}$

${u u}_{{f f}_{p p}} (({HC HC}_{66})) = = \{\begin{matrix} \frac{11 - - {f f}_{p p}}{22} - - 33 \times \times \frac{{f f}_{p p}^{55}}{55!!} & 00 \leq \leq {f f}_{p p} < < 0.1 0.1 \\ \frac{15.8 15.8 - - {f f}_{p p}}{120120} & 0.1 0.1 \leq \leq {f f}_{p p} < < 0.7 0.7 \\ {f f}_{p p}^{44} - - \frac{{f f}_{p p}^{44}}{44} & 0.7 0.7 \leq \leq {f f}_{p p} < < 0.8 0.8 \\ \frac{{f f}_{p p}^{33}}{33!!} - - \frac{{f f}_{p p}^{33}}{44!!} & 0.8 0.8 \leq \leq {f f}_{p p} \leq \leq 11 \end{matrix}$

${u u}_{{f f}_{p p}} ((H h {C C}_{77})) = = \{\begin{matrix} \frac{11 - - 33 \times \times {f f}_{p p}}{22} & 00 \leq \leq {f f}_{p p} < < 0.1 0.1 \\ \frac{15.8 15.8 - - {f f}_{p p}}{120120} & 0.1 0.1 \leq \leq {f f}_{p p} < < 0.7 0.7 \\ \frac{{f f}_{p p}^{55}}{55!!} & 0.7 0.7 \leq \leq {f f}_{p p} < < 0.8 0.8 \\ \frac{{f f}_{p p}^{55}}{55!!} & 0.8 0.8 \leq \leq {f f}_{p p} \leq \leq 11 \end{matrix}$

${m m}_{33} (({HC HC}_{j j})) = = \frac{{u u}_{{f f}_{p p}} (({HC HC}_{j j}))}{{Σ Σ}_{j j = = 11}^{77} {u u}_{{f f}_{p p}} (({HC HC}_{j j}))},, j j = = 1,2,3,4,5,6,7 1,2,3,4,5,6,7$

图1中，步骤110利用D-S证据理论的合成法则，将三个分类器的结果进行融合判决，得到融合后的基本概率分配函数。由于憎水性三个特征分量之间相关性较小，因此本方法假设三个分类器产生的证据结果是独立的，这样即可利用下式的D-S合成规则来进行融合，即：In Fig. 1, step 110 utilizes the combination rule of the D-S evidence theory to fuse and judge the results of the three classifiers to obtain the basic probability distribution function after fusion. Since the correlation between the three characteristic components of hydrophobicity is small, this method assumes that the evidence results produced by the three classifiers are independent, so that the following D-S composition rules can be used for fusion, namely:

$m m ((HC HC)) = = \frac{\underset{\cap \cap {HC HC}_{j j} = = HC HC}{Σ Σ} {Π Π}_{i i = = 11}^{33} {m m}_{i i} (({HC HC}_{j j}))}{11 - - \underset{\cap \cap {HC HC}_{j j} = = φ φ}{Σ Σ} {Π Π}_{i i = = 11}^{33} {m m}_{i i} (({HC HC}_{j j}))},, j j = = 1,2,3,4,5,6 1,2,3,4,5,6,, 77,, i i = = 1,2,3 1,2,3$

由上式可以得到融合后的基本概率分配函数m(HC)，在得到融合后的基本概率分配函数后，由式(1)计算信度区间

针对每一幅憎水性图像，定义以下规则来确定该憎水性等级：The fused basic probability distribution function m(HC) can be obtained from the above formula. After the fused basic probability distribution function is obtained, the reliability interval can be calculated by formula (1)

For each hydrophobicity image, define the following rules to determine the hydrophobicity level:

图1中，步骤111利用等级判定的基本原则确定憎水性等级，实现憎水性等级融合判决。在具体判定时，按所列规则的顺序依次判定，即若只有唯一的最大信任度值，则后面的规则不需要检验，否则的话再判定第二个规则，依此类推。In Figure 1, step 111 uses the basic principles of grade determination to determine the hydrophobicity grade, and realizes the fusion judgment of the hydrophobicity grade. When making a specific judgment, judge according to the order of the listed rules, that is, if there is only a unique maximum trust value, then the following rules do not need to be checked, otherwise, judge the second rule, and so on.

假设，有如下运算结果，Assume that the following operation results are obtained,

其憎水性等级融合判决为：在行1中，显然HC1所对应的信任度值最大，此时利用规则1)将其判断为HC1；在行2中，HC1及HC3具有相同的信任度值，无法利用1)和2)规则判定，则利用3)规则，由于不确定性区间长度应小于最大可信任值的25％，即0.088，所以将其判断为HC1，而非HC3。The fusion judgment of hydrophobicity level is: in line 1, obviously the trust value corresponding to HC1 is the largest, at this time, it is judged as HC1 by using rule 1); in line 2, HC1 and HC3 have the same trust value, If the rules 1) and 2) cannot be used, then the rule 3) should be used. Since the length of the uncertainty interval should be less than 25% of the maximum credible value, that is, 0.088, it is judged as HC1 instead of HC3.

以上所述，仅为本发明较佳的具体实施方式，但本发明的保护范围并不局限于此，任何熟悉本技术领域的技术人员在本发明揭露的技术范围内，可轻易想到的变化或替换，都应涵盖在本发明的保护范围之内。因此，本发明的保护范围应该以权利要求的保护范围为准。The above is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed in the present invention can easily think of changes or Replacement should be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be determined by the protection scope of the claims.

Claims

1. A method for insulator hydrophobic grade fusion judgment based on D-S evidence theory, characterized in that said method comprises the following steps:

Step 1: Use the water spray classification method to obtain the hydrophobicity image of the insulator to be tested;

Step 2: Cut out the hydrophobicity image of the insulator to be tested within 1-2 seconds after the water spraying ends;

Step 3: Perform histogram equalization processing on the cropped hydrophobic image;

Step 4: denoising the equalized image;

Step 5: Using the method of information measurement, determine the three information measurement components of the hydrophobic image processed in steps 1 to 4, the neighborhood order C(i, j) based on fuzzy entropy, and the direction based on gradient and fuzzy entropy M(i, j) and similarity-based component R(i, j); where (i, j) is the pixel of the hydrophobic image;

Step 6: Utilize the image analysis method of fuzzy C-means clustering to realize the segmentation of the hydrophobic image background and objects;

shape feature Highlight Features

Step 8: Use the extracted area feature, shape feature and bright spot feature components as three feature classifiers respectively;

Step 9: Construct the hydrophobic membership function, use the membership function to represent the credibility of the proposition, and use the fuzzy rules to find the basic probability distribution function of each classifier;

Step 10: Using the composition rule of D-S evidence theory, fuse the basic probability distribution functions of the three classifiers to obtain the basic probability distribution function after fusion;

Step 11: Use the basic principles of grade judgment to determine the hydrophobicity grade, and realize the fusion judgment of hydrophobicity grade.

2. A method for judging the level of insulator hydrophobicity fusion based on D-S evidence theory according to claim 1, characterized in that the water spray classification method is used to obtain the image of the hydrophobicity of the insulator to be tested, and the process is that the insulator to be tested is Water spraying operation, after the water spraying is over, take a dynamic video, and use the method of intercepting still images from the captured dynamic video to obtain the water spraying image of the insulator.

3. According to claim 1, an insulator hydrophobic grade fusion judgment method based on D-S evidence theory is characterized in that the hydrophobicity image of the insulator to be tested is cut out, and the method is to select in the image that does not include composite The edge of the shed of the insulator and the part with the worst hydrophobicity in the water spray area are cut by image processing software.

4. A method for judging the level of insulator hydrophobicity fusion based on DS evidence theory according to claim 1, characterized in that the histogram equalization process is performed on the cropped hydrophobicity image, and the method is that each pixel Point (i, j) realizes adaptive histogram equalization of hydrophobic image:

T (f (i, j)) = 255 \times Σ_{r = 0}^{k} \frac{{no}_{r}}{m^{2}},

Among them, n _r is the number of pixels with gray level r in the selected sliding window M×M, f(i, j) is the gray value of (i, j) pixel, k is f(i, j) Corresponds to grayscale.

5. According to claim 1, a method for judging insulator hydrophobicity grade fusion based on D-S evidence theory, is characterized in that said denoising processing of the image after equalization processing is a method of median filtering, forcing the The disturbed pixel is transformed into the gray value of its adjacent pixel to achieve the effect of removing interference and noise.

6. According to claim 1, a method for judging insulator hydrophobicity level fusion based on DS evidence theory, characterized in that the image analysis method using fuzzy C-means clustering realizes the segmentation of hydrophobic image background and objects, The process is, firstly: the image analysis algorithm of fuzzy C-means clustering is

J (u, c_{1}, . . ., c_{c}) = Σ_{i = 1}^{c} J_{i} = Σ_{i = 1}^{c} Σ_{j}^{no} u_{ij}^{m} d_{ij}^{2},

and

c_{i} = \frac{Σ_{j = 1}^{no} u_{ij}^{m} x_{j}}{Σ_{j = 1}^{no} u_{ij}^{m}},

u_{ij} = \frac{1}{Σ_{k = 1}^{c} {(\frac{d_{ij}}{d_{kj}})}^{2 / (m - 1)}}

Where u _ij is between [0 1]; c _i is the cluster center of fuzzy group i, d _ij is the Euclidean distance between the i-th cluster center and the j-th data point, and m∈[1 , ∞) is the weighted exponent; then pass

7. According to claim 1, a D-S evidence theory-based insulator hydrophobic grade fusion judgment method is characterized in that the construction of the hydrophobic membership function is carried out by using fuzzy rules to obtain three classifications respectively. The basic probability distribution function of the device, the π-type function is selected for the membership function of the area feature, the τ-type function is selected for the membership function of the shape feature, and the selection of the membership function of the bright spot feature is based on the Taylor series expansion.

8. A method for insulator hydrophobic grade fusion judgment based on D-S evidence theory according to claim 1, characterized in that the basic principles of the grade judgment include:

1) The target level has the largest trust value;

2) The difference between the trust value of the target level and other levels is greater than 25% of the maximum trust value;

3) The length of the uncertainty interval is less than 25% of the maximum confidence value;

4) The trust value of the target level is greater than the value of the fuzzy basic probability distribution function.