Overhead line porcelain insulator triple string image segmentation method

Abstract

The invention discloses an overhead line porcelain insulator triple string image segmentation method, which comprises the following steps: S1, shooting a target insulator at multiple angles by using an infrared camera of an unmanned aerial vehicle to obtain infrared images of the target insulator at multiple angles; preprocessing the plurality of infrared images of the target insulator to obtain a target insulator image data set; s2, based on a lightweight neural MobileNet network, improving a Clear network model, and obtaining a lightweight MN-Clear network model; s3, sending the target insulator image data set obtained in the step S1 into an MN-Clear network model for training, calling an unshielded infrared image as an excellent value, setting the excellent value as a reference value, and storing the reference value as a correction threshold value; s4, correcting the infrared images of the target insulator at multiple angles by using the excellent values obtained by training; and S5, carrying out image segmentation on the correction result to obtain a segmented image segmentation result. According to the method, the infrared image of the three-string insulator of the overhead line can be accurately segmented.

Description

Technical Field

[0001] The invention relates to the technical field of live inspection of power transmission lines by unmanned aerial vehicles, and in particular to an image segmentation method for a triple string of porcelain insulators of an overhead line. Background Art

[0002] Insulators are important basic insulation components in power systems, playing the role of electrical insulation and mechanical support. Their insulation performance directly affects the normal operation of the power grid. As the operating time increases, due to the combined influence of environmental factors and electromechanical loads, their insulation and mechanical properties will gradually decline, leading to aging and degradation of insulators, and even serious accidents such as explosions and string drops, causing large-scale power outages, threatening the safe and stable operation of the power grid.

[0003] The traditional insulator detection method is manual tower inspection, which has low detection efficiency, high work intensity and high risk factor. Infrared thermal imaging detection is currently the most effective alternative method. Automatic positioning, identification and segmentation of insulator areas in infrared images are the key to completing intelligent detection of insulator degradation status. In actual applications, due to the poor contrast of infrared images of triple strings of porcelain insulators on overhead lines, single channel features, complex background and image overlap, there is currently no more effective method to segment infrared images of triple strings of insulators on overhead lines. Summary of the invention

[0004] The object of the present invention is to provide an overhead line porcelain insulator triple string image segmentation method to solve the problems raised in the background technology.

[0005] To achieve the above object, the present invention provides an overhead line porcelain insulator triple string image segmentation method, comprising the following steps:

[0006] S1. Use the infrared camera of the drone to shoot the target insulator at multiple angles to obtain infrared images of the target insulator at multiple angles; pre-process the multiple infrared images of the target insulator to obtain a target insulator image data set;

[0007] S2. Based on the lightweight neural MobileNet network, the Clear network model is improved to obtain a lightweight MN-Clear network model; wherein the infrared image of the unobstructed insulator string is called the excellent value, and the excellent value is saved as the correction threshold;

[0008] S3, sending the target insulator image dataset obtained in step S1 into the lightweight MN-Clear network model for training, calling the unobstructed infrared image an excellent value, setting the excellent value as a reference value and saving the correction threshold (weight, hereinafter referred to as correction threshold);

[0009] S4, using the excellent values ​​obtained through training in step S3 to correct the infrared images of the target insulator at multiple angles;

[0010] S5. Perform image segmentation on the correction result obtained in step S4 to obtain a segmented image segmentation result.

[0011] Furthermore, the preprocessing of the multiple infrared images of the target insulator in step S1 is specifically as follows:

[0012] S1.1, manually select infrared images with simple background and no obstruction;

[0013] S1.2. Artificial intelligence identifies insulation strings and identifies the specific orientation of the porcelain insulator triple string through multiple sets of infrared images. When the drone is shooting, the built-in gyroscope can capture the azimuth and its horizontal angle respectively, thereby distinguishing them.

[0014] S1.3. Name or label the infrared images for subsequent classification.

[0015] Furthermore, in step S2, the structure of the lightweight MN-Clear network model includes an input layer for receiving image data, a convolution layer for extracting image features using convolution operations and pooling operations, a fully connected layer for mapping the features extracted by the convolution layer to the final output, and a Clear layer for regressing the output of the fully connected layer.

[0016] Furthermore, in step S3, the training process of the lightweight MN-Clear network model is as follows: using the MobileNet network to extract image features in the target insulator image data set; using the extracted image features as input to train a Clear regression model; and using the trained Clear regression model to perform regression prediction on new image data.

[0017] Furthermore, in step S3, the lightweight MN-Clear network model extracts image features of the target insulator image dataset through convolution operations and pooling operations, and uses a fully connected layer for classification or regression.

[0018] Furthermore, in step S3, the lightweight MN-Clear network model continuously optimizes the algorithm through the following convolution formula:

[0019] picture = R + G + B (1);

[0020] pur = α × picture (2);

[0021] yellow = β × picture (3);

[0022] new=pur+yellow (4);

[0023] new = (α + β) (R + G + B) (5);

[0024] In the above formulas (1)-(5), α and β represent correction threshold coefficients, which are values ​​between 0 and 1.

[0025] Compared with the prior art, the present invention has the following beneficial effects:

[0026] The invention discloses an overhead line porcelain insulator triple string image segmentation method. A special image analysis method is made for the triple string existing in the high-voltage transmission line. By shooting the target insulator at multiple angles, an excellent value with the least obstruction, the cleanest background and the clearest shooting in a certain direction is obtained. The excellent value is used as a reference and set as a calibration base. Other pictures shot in the same direction are calibrated and segmented. The image segmentation method of the invention can accurately segment the infrared image of the overhead line triple string insulator, improve the detection efficiency and reduce the work intensity.

[0027] In addition to the above-described purposes, features and advantages, the present invention has other purposes, features and advantages. The present invention will be further described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the following specific implementations, they are used to explain the embodiments of the present invention, but do not constitute a limitation on the embodiments of the present invention. In the accompanying drawings:

[0029] Figure 1 It is an infrared image of a triple string of porcelain insulators with obstructions used in an image segmentation method of a triple string of porcelain insulators for overhead lines of the present invention;

[0030] Figure 2 It is a schematic diagram of the infrared image excellence value of a triple string of porcelain insulators used in an overhead line porcelain insulator triple string image segmentation method of the present invention;

[0031] Figure 3 It is the deviation between the predicted value and the true value calculated by the convolution formula; the smaller the deviation, the smaller the segmentation error;

[0032] Figure 4 is a schematic diagram of a target insulator with excellent values ​​of the azimuth as a set. DETAILED DESCRIPTION

[0033] The present invention will be described in detail below in conjunction with the various embodiments shown in the accompanying drawings, but it should be noted that these embodiments are not limitations of the present invention, and any equivalent transformations or substitutions in functions, methods, or structures made by ordinary technicians in the field based on these embodiments are all within the scope of protection of the present invention.

[0034] See also Figures 1 to 4 This embodiment provides an overhead line porcelain insulator triple string image segmentation method, comprising the following steps:

[0035] S1. Use the infrared camera of the drone to shoot the target insulator at multiple angles to obtain infrared images of the target insulator at multiple angles; pre-process the multiple infrared images of the target insulator to obtain a target insulator image data set;

[0036] The specific preprocessing is:

[0037] S1.1. Select high-quality images with simple background and no obstruction through manual screening;

[0038] S1.2, artificial intelligence identifies insulation strings, and identifies the specific orientation of the porcelain insulator triple string through multiple sets of infrared images;

[0039] S1.3. Name or label the infrared images for subsequent classification.

[0040] S2. Based on the lightweight neural MobileNet network, the Clear network model is improved to obtain a lightweight MN-Clear network model; the lightweight MN-Clear network model includes an input layer for receiving image data, a convolutional layer for extracting image features using convolution and pooling operations, a fully connected layer for mapping the features extracted by the convolutional layer to the final output, and a Clear layer for regressing the output of the fully connected layer.

[0041] S3. Send the target insulator image dataset obtained in step S1 to the lightweight MN-Clear network model for training. The unobstructed infrared image is called the excellent value, set as the reference value and save (weight, hereinafter referred to as the correction threshold); the training process of the lightweight MN-Clear network model is as follows: use the MobileNet network to extract image features in the target insulator image dataset; use the extracted image features as input to train a Clear regression model; use the trained Clear regression model to perform regression prediction on other image data taken in the same orientation. The lightweight MN-Clear network model extracts image features of the target insulator image dataset through convolution operations and pooling operations, and uses a fully connected layer for classification or regression. The lightweight MN-Clear network model continuously optimizes the algorithm through the following convolution formula:

[0042] picture = R + G + B (1);

[0043] pur = α × picture (2);

[0044] yellow = β × picture (3);

[0045] new=pur+yellow (4);

[0046] new = (α + β) (R + G + B) (5);

[0047] In the above formulas (1)-(5), α and β represent the correction threshold coefficients, which are values ​​between 0 and 1. This formula can be used to represent the edge of the image by a numerical value and perform correction processing. Formula (1) is the coordinate value in the two-dimensional image. The deviation between the predicted value and the true value calculated by the convolution formula is as follows: Figure 3 As shown in the figure, it can be seen that the predicted value obtained by the method of the present invention is close to the true value.

[0048] S4, using the excellent value obtained by training in step S3 to correct the infrared images of the target insulator at multiple angles; specifically: comparing the target insulator and the excellent value at the same angle can be understood as a segmentation of a set, such as Figure 4 , respectively name the green, blue, and gray areas A, B, and C, the excellent value is A∪B, the target value is B∪C, there is an intersection B between the two, and there are also correction deviations A and C, so the corrected result should be A∪B. That is, the correction value of the target image should be infinitely close to the excellent value.

[0049] S5. Perform image segmentation on the correction result obtained in step S4 to obtain a segmented image segmentation result. Compare and analyze the corrected result (which can be divided into two areas, excellent value area and non-excellent value area) with the captured infrared image, and retain the rows (excellent value area) within the segmented area, and cut the rest (non-excellent value area).

[0050] In the present invention, the three strings of insulators are parallel and in the same plane, but shooting the insulator strings requires shooting at multiple angles, and some angles may be blocked. Therefore, pictures taken from some unobstructed angles are needed as a reference for segmentation. For example, when overlap occurs, the segmentation plan when no overlap occurs can be referred to.

[0051] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for segmenting an image of a triple string of porcelain insulators of an overhead line, characterized in that: The following steps are involved: S1. Use the infrared camera of the drone to shoot the target insulator at multiple angles to obtain infrared images of the target insulator at multiple angles; pre-process the multiple infrared images of the target insulator to obtain a target insulator image data set; S2. Based on the lightweight neural MobileNet network, the Clear network model is improved to obtain a lightweight MN-Clear network model; S3, sending the target insulator image dataset obtained in step S1 into the lightweight MN-Clear network model for training; wherein the infrared image of the unobstructed insulator string is called the excellent value, and the excellent value is saved as the correction threshold; S4, using the excellent values ​​obtained through training in step S3 to correct the infrared images of the target insulator at multiple angles; S5. Perform image segmentation on the correction result obtained in step S4 to obtain a segmented image segmentation result.

2. The overhead line porcelain insulator triple string image segmentation method according to claim 1, characterized in that: The preprocessing of the multiple infrared images of the target insulator in step S1 is specifically as follows: S1.1, manually screen out infrared images with simple background and no obstruction; S1.2, artificial intelligence identifies insulation strings, and identifies the specific orientation of the porcelain insulator triple string through multiple sets of infrared images; S1.

3. Name or label the infrared images for subsequent classification.

3. The overhead line porcelain insulator triple string image segmentation method according to claim 1, characterized in that: In step S2, the structure of the lightweight MN-Clear network model includes an input layer for receiving image data, a convolution layer for extracting image features using convolution operations and pooling operations, a fully connected layer for mapping the features extracted by the convolution layer to the final output, and a Clear layer for regressing the output of the fully connected layer.

4. The overhead line porcelain insulator triple string image segmentation method according to claim 3, characterized in that: In step S3, the training process of the lightweight MN-Clear network model is as follows: using a lightweight neural MobileNet network to extract image features in the target insulator image data set; using the extracted image features as input to train a Clear regression model; and using the trained Clear regression model to perform regression prediction on new image data.

5. The overhead line porcelain insulator triple string image segmentation method according to claim 3, characterized in that: In step S3, the lightweight MN-Clear network model extracts image features of the target insulator image dataset through convolution operations and pooling operations, and uses a fully connected layer for classification or regression.

6. The overhead line porcelain insulator triple string image segmentation method according to claim 3, characterized in that: In step S3, the lightweight MN-Clear network model continuously optimizes the algorithm through the following convolution formula: picture = R + G + B (1); pur = α × picture (2); yellow = β × picture (3); new=pur+yellow (4); new = (α + β) (R + G + B) (5); In the above formulas (1)-(5), α and β represent correction threshold coefficients, which are values ​​between 0 and 1.

Images