CN111667473A - Insulator hydrophobicity grade judging method based on improved Canny algorithm - Google Patents

Abstract

The invention discloses a method for judging the hydrophobicity level of an insulator based on an improved Canny algorithm, comprising: collecting several original images of the insulator after simulating rain shower; converting the original image into a grayscale image; enhancing the image contrast and denoising; Use the canny algorithm to obtain the denoised image edges; connect the edges that cannot be effectively identified by the canny algorithm with breakpoints; perform contour morphological filling on the images after the breakpoints are connected; obtain the characteristic parameters of the filled images and normalize them. Normalization processing; send the normalized characteristic parameters of the training samples into the neural network model for training to obtain the insulator hydrophobicity judgment model; input the normalized characteristic parameters of each test sample into the insulator hydrophobicity judgment model to obtain The hydrophobicity grade of each test sample. The invention can more accurately identify the hydrophobicity level of the insulator, has high identification accuracy, can carry out live detection, greatly reduces the cost of electric equipment inspection and improves safety.

Description

Judgment method of insulator hydrophobicity grade based on improved Canny algorithm

technical field

The invention belongs to the field of running state detection of power transmission and transformation equipment, in particular to a method for judging the hydrophobicity level of an insulator based on an improved Canny algorithm.

Background technique

The hydrophobicity level of the surface of outdoor insulators will inevitably change, and the hydrophobicity of the surface of the silicone rubber insulators will decrease with the running time and the comprehensive factors such as corona, flashover, pollution, temperature, rain and snow, etc. even completely lost. The decline of the hydrophobicity and hydrophobic migration characteristics of silicone rubber insulators is manifested in the lower pollution flashover resistance, which poses a threat to the safety and stable operation of the power system.

Up to now, domestic and foreign experts and scholars have conducted a lot of research and experiments on the detection methods of insulator hydrophobicity. Although some detection methods have been applied in practice, most of the detection and evaluation methods of hydrophobicity are still in the immature stage. Relying on manual or cumbersome experimental steps to test its hydrophobicity, it is of great practical engineering value to invent a method for judging the hydrophobicity level of insulators.

Experts at home and abroad have done a lot of research on the detection methods of insulators. Among the existing insulator inspection methods, an improved form factor method is used for inspection. The method first extracts the binary image of the hydrophobic image, then extracts the largest water droplet (water trace) in the image, calculates its area ratio to the total image and its shape characteristics, and then analyzes these data through a large number of mathematical statistics , find the mathematical relationship between these two eigenvalues and the rank classification, so as to classify the image with the prediction. There is also the use of k-nearest neighbor method for classification. The method first uses image processing technology to process the image to obtain the edges of all water droplets (water traces) in the image, calculates all the eigenvalues of the water droplets, and classifies them by the k-nearest neighbor method. There is also an SVM decision tree method for detection. First, the image to be detected should be enhanced, and then it is segmented by a certain algorithm, and its feature values are extracted, and finally the SVM decision tree is used to classify and identify the image to be detected.

The statistics of the research results show that although these algorithms have some advantages, such as improving the accuracy of the hydrophobicity of composite insulators and reducing the errors caused by manual classification, there are still some defects. For example, the shape factor method extracts the relationship between the feature value and the classification. In practical engineering applications, there is a nonlinear relationship between the hydrophobicity level and the extracted eigenvalues in some levels. Therefore, this method cannot judge all hydrophobicity grades. The method of SVM decision tree has a great relationship with the amount of data and is not suitable for classifying big data.

SUMMARY OF THE INVENTION

The purpose of the present invention is to provide a method for judging the hydrophobicity level of insulators based on the improved Canny algorithm, aiming at the above-mentioned deficiencies of the prior art. Live detection greatly reduces the cost of electrical equipment inspection and improves safety.

For solving the above-mentioned technical problems, the technical scheme adopted in the present invention is:

A method for judging the hydrophobicity level of insulators based on the improved Canny algorithm, which is characterized by including the following steps:

Step 1, collecting several original images of insulators of various hydrophobic grades after simulating rain shower, wherein M images are used as training samples, and N images are used as test samples;

Step 2: Convert the original images of the training samples and test samples into grayscale images;

Step 3, enhancing the image contrast of the grayscale image and denoising it;

Step 4, use the canny algorithm to obtain the edge of the denoised image, and obtain a binary image that only retains edge information;

Step 5: Connect the edges that cannot be effectively identified by the canny algorithm to breakpoints through the edge connection algorithm;

Step 6, perform contour morphological filling on the image after the breakpoint connection, to obtain a binary image of the outer contour of the water trace and the surface of the insulator;

Step 7, obtain the characteristic parameters of the image after filling, the characteristic parameters include the area of the largest water trace, the proportion of the largest water trace in the graphic area, and the shape factor; the characteristic parameters are normalized;

Step 8, sending the normalized characteristic parameters of the training samples into the neural network model for training to obtain the insulator hydrophobicity judgment model;

Step 9: Input the normalized characteristic parameters of each test sample into the insulator hydrophobicity judgment model to obtain the hydrophobicity level corresponding to each test sample.

As a preferred manner, in the step 3, a multi-scale retinex algorithm is used to enhance the image contrast of the grayscale image.

As a preferred way, the step 3 includes:

Calculate the enhanced image at a single scale;

Weighted summation of each enhanced image at multiple scales;

The enhanced image after weighted summation is used as the guide image for filtering and denoising.

As a preferred way, in the step 5, the edge connection algorithm includes:

Connect to the surrounding 8 neighborhoods starting from the endpoints of all broken edges;

According to the location of the endpoint and the size of the gradient magnitude matrix, the pixels that need to be connected are determined until they are connected to another edge or boundary.

As a preferred way, in the canny algorithm, the maximum inter-class variance method is used to obtain the threshold.

Compared with the prior art, the present invention can more accurately identify the hydrophobicity level of the insulator, improve the accuracy of hydrophobicity identification, and can perform live detection, greatly reducing the cost of electrical equipment inspection and improving safety.

Description of drawings

Fig. 1 is the flow chart of the method of the present invention.

Figure 2 is an image obtained after grayscale processing of the original color image.

Figure 3 is an image obtained by using the retinex algorithm and guided filtering.

Figure 4 is an image obtained by using the traditional canny algorithm or the improved canny algorithm.

Figure 5 is a flowchart of the improved canny algorithm edge connection algorithm.

Figure 6 is the image after contour filling.

Figure 7 is a diagram of the training process of the neural network model.

FIG. 8 is an example of the result of determination of the water repellency level.

Detailed ways

The working principle of the present invention: water will show different shapes on the surface of composite insulators with different hydrophobicity levels. The shape is attached to the surface of the insulator. On the surface of the composite insulator with a hydrophobicity level of HC7, it is almost impossible to form circular water droplets, which adhere to the surface of the insulator in the form of a large-area water film. At the same time, due to environmental factors, there are many kinds of dirt on the surface. According to this feature, the hydrophobicity grade can be judged.

As shown in Figure 1, the method for judging the hydrophobicity level of insulators based on the improved Canny algorithm includes the following steps:

Step 1: Collect several original images (color) of insulators of various hydrophobic grades after simulated rain spraying through the electronic camera on the high-voltage tower, wherein M images are used as training samples, and N images are used as test samples. ; transfer the original image back to the computer for subsequent processing of the original image.

In step 2, the original images of the training samples and the test samples are preprocessed and converted into grayscale images; an image as shown in FIG. 2 is obtained.

Step 3, use the multi-scale retinex algorithm to enhance the image contrast of the grayscale image (that is, enhance the contrast between the edge and the background of the image) and use the guided filtering algorithm to denoise it; an image as shown in Figure 3 is obtained.

The step 3 specifically includes:

Calculate the enhanced image at a single scale;

Weighted summation of each enhanced image at multiple scales;

The enhanced image after weighted summation is used as the guide image for filtering and denoising.

The main principle of step 3 is as follows:

The retinex algorithm can regard the image as the superposition of the incident image and the reflected image. The incident light shines on the reflective object, and the reflected light enters the human eye through the reflection of the reflective object. The resulting image can be represented by the following formula:

Among them, R(x,y) represents the reflection property of the object, that is, the intrinsic property of the image, which we should preserve to the greatest extent; and L(x,y) represents the incident light image, which determines the dynamic range that the image pixels can achieve. We should be removed as much as possible.

Generally, we assume that the illumination image is estimated to be a spatially smooth image, the original image is S(x,y), the reflection image is R(x,y), and the luminance image is L(x,y), we can obtain formula (1) , and formula (2):

where r(x,y) is the output image. F(x,y) is the center wrap function, expressed as:

In formula (3), c is the Gaussian surround scale, λ is a scale, and its value must satisfy the following formula:

∫∫F(x,y)dxdy=1 (4)

The above is the retinex algorithm in a single scale, and the multi-scale retinex algorithm adopted in the present invention needs to perform weighted summation of r(x, y) in each single scale. The formula is as formula (5):

When conducting guided filtering on an image, the output of a pixel in the image is as follows:

where q is the output image, I is the guide image, and α and b are the invariant coefficients of this linear function when the center of the window is located. The assumption of this method is that there is a local linear relationship between q and I in the window centered on pixel k. In order to solve the coefficients a and b in (6), it is assumed that p is the result before q filtering, and the difference between q and p is minimized. According to the method of unconstrained image restoration, it can be transformed into an optimization problem, and its value function is (7).

The solution obtained by the least squares method is:

是I在窗口ω_k中的方差，|ω|是窗口ω_k中像素的数量，

是待滤波图像p在窗口ω_k中的均值。where μ _k is the average value of I in the window ω _k ,

is the variance of I in window ω _k , |ω| is the number of pixels in window ω _k ,

is the mean of the image p to be filtered in the window ω _k .

In the end, each pixel will be contained by multiple windows, so the weighted average can be obtained (9):

Step 4, using the improved canny algorithm to obtain the edge of the image after denoising, to obtain a binary image that only retains edge information.

The degree of automation of the traditional canny algorithm is not high, and the threshold value needs to be manually specified; and in the improved canny algorithm used in the present invention, the maximum inter-class variance method is used to obtain the threshold value, and its formula is:

In step 5, the edges captured by the traditional canny algorithm are often broken after completion, which is very difficult for the next contour filling. The invention uses the edge connection algorithm to connect the edges that cannot be effectively identified by the canny algorithm to breakpoints; after the breakpoints are connected, a complete edge is obtained. As shown in the right figure of Figure 4.

By comparing the left and right images in Figure 4, it can be found that the edges captured by the original canny algorithm (the original canny algorithm on the left) will have edge fractures, as marked by the small gray circles in the left image of Figure 4. This loses part of the edge information, which has a great impact on the subsequent identification of hydrophobicity grades. The one on the right is the improved canny algorithm. It can be found that the broken edges in the left picture can be captured in the improved canny algorithm, and the information of the edges is almost not lost, which is very helpful for the subsequent identification of the hydrophobicity level.

As shown in Figure 5, in step 5, the edge connection algorithm includes:

Starting from each broken edge, starting from the end point, the connection direction of the edge is determined according to the direction of the end point and the magnitude of the gradient, and at each connection step, it is judged whether it is connected to another edge. until the end of each broken edge connection.

Step 6: Perform contour morphological filling on the image after the breakpoints are connected to obtain a binary image of the outer contour of the water trace and the surface of the insulator; the image after contour filling is shown in Figure 6.

Step 7, use morphological algorithm to obtain the characteristic parameters of the filled image, the characteristic parameters include the area of the largest water trace, the proportion of the largest water trace in the area of the figure, and the shape factor; in order to avoid numerical problems and the appearance of differences , normalize the feature parameters.

In step 8, the normalized characteristic parameters of the training samples are sent to the neural network model for training. The training process is shown in Figure 7, and the insulator hydrophobicity judgment model is obtained.

The neural network model includes a pooling layer, a depthwise separable convolutional layer, a fully connected layer, and a pooling layer.

Step 9: Input the three characteristic parameters after normalized processing of each test sample into the insulator hydrophobicity judgment model to obtain the hydrophobicity level corresponding to each test sample. An example of the water repellency level judgment result is shown in Figure 8.

In this example, 1132 insulator images of various levels are taken as the research object. Among them, 755 are used as training set and 377 are used as validation set. Obtain the maximum area, maximum area ratio, and shape factor of all training set images. Fit the data into a 1132*3 matrix and send it to the neural network model for training. Extract the maximum area, maximum area ratio, and shape factor of all validation stages and input them into the trained model to obtain a model with a validation accuracy of 97.8%.

The present invention is a method for judging the hydrophobicity level of an insulator based on the improved canny algorithm. The biggest difference from other detection technologies is that the improved canny algorithm can capture relatively complete edge information of water traces. The second is the high accuracy of the trained model. The power transmission and transformation staff only need to capture the insulator picture on the high-voltage tower and input it into the computer to obtain the corresponding hydrophobicity level, which is convenient and quick and has practical engineering application value.

The embodiments of the present invention have been described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific embodiments, which are merely illustrative rather than limiting. Under the inspiration of the present invention, without departing from the scope of protection of the spirit of the present invention and the claims, many forms can be made, which all fall within the protection scope of the present invention.

Claims (5)

1. An insulator hydrophobicity grade judging method based on an improved Canny algorithm is characterized by comprising the following steps:

step 1, collecting a plurality of original images of insulators with various hydrophobic grades sprayed by simulated rainwater, wherein M images are used as training samples, and N images are used as inspection samples;

step 2, converting original images of the training sample and the test sample into gray level images;

step 3, enhancing the image contrast of the gray level image and carrying out denoising treatment on the image contrast;

step 4, obtaining the edges of the image after denoising by using a canny algorithm to obtain a binary image only retaining edge information;

step 5, performing breakpoint connection on the edges which cannot be effectively identified by the canny algorithm through an edge connection algorithm;

step 6, carrying out contour morphological filling on the image after the breakpoint connection to obtain a water mark external contour and a binary image of the surface of the insulator;

step 7, solving the characteristic parameters of the filled image, wherein the characteristic parameters comprise the area of the maximum water mark, the proportion of the maximum water mark in the area of the image and a shape factor; normalizing the characteristic parameters;

step 8, sending the characteristic parameters after the training sample normalization processing into a neural network model for training to obtain an insulator hydrophobicity judgment model;

and 9, inputting the characteristic parameters after normalization processing of each test sample into the insulator hydrophobicity judgment model to obtain the hydrophobicity grade corresponding to each test sample.

2. The method for judging the hydrophobicity grade of the insulator based on the modified Canny algorithm in claim 1, wherein in the step 3, a multi-scale retinex algorithm is used for enhancing the image contrast of the gray scale image.

3. The insulator hydrophobicity grade judging method based on the modified Canny algorithm in claim 2, wherein the step 3 comprises the following steps:

calculating an enhanced image under a single scale;

performing weighted summation on each enhanced image under multiple scales;

and performing filtering and denoising by taking the weighted and summed enhanced image as a guide image.

4. The insulator hydrophobicity rating method according to claim 1, wherein in the step 5, the edge connection algorithm comprises:

starting to connect to 8 surrounding neighborhood by starting from the end points of all the fracture edges;

and determining the pixel points needing to be connected according to the positions of the end points and the size of the gradient amplitude matrix until the pixel points are connected to another edge or boundary.

5. The method for determining the hydrophobicity grade of an insulator based on the modified Canny algorithm in claim 1, wherein in the Canny algorithm, the threshold is obtained by using a variance method between maximum classes.

Images