A composite insulator defect detection method based on image analysis

Abstract

The invention discloses a composite insulator defect detection method based on image analysis, and relates to the technical field of composite insulator detection. The invention collects visible light images and red-ultraviolet images of defective composite insulators through unmanned aerial vehicles, pre-processes the collected composite insulator images; decomposes the pre-processed visible light images and red-ultraviolet images of composite insulators; fuses low-frequency sub-band images according to the matching degree of regional variance of low-frequency sub-band images; analyzes the fusion weight coefficient of high-frequency sub-band images according to a pulse coupled neural network model; evaluates the composite insulator fusion image obtained by fusion processing; and distinguishes the defect types of composite insulators according to defect recognition, abnormal temperature recognition, and ultraviolet discharge recognition. A new method is provided for composite insulator detection, which improves the degree of automation of detection and improves inspection efficiency.

Description

Technical Field

[0001] The invention relates to the technical field of composite insulator detection, and in particular to a composite insulator defect detection method based on image analysis. Background Art

[0002] Composite insulators are an important part of transmission lines and substations. Their main function is to suspend the conductors on the tower frame. Therefore, composite insulators are required to have not only excellent mechanical properties, but also excellent electrical properties. As the power system develops towards large capacity and ultra-high voltage, the safe and stable operation of composite insulators has received more and more attention. Any composite insulator will deteriorate to a certain extent after a long period of operation.

[0003] Since power transmission lines are often located tens of meters above the ground, with long lines and numerous components, and some transmission lines are even located in remote and difficult-to-reach places such as mountains and swamps, traditional manual inspection methods are very inefficient. Technicians often need to climb towers or hold detection equipment to inspect key components of composite insulators and record the status information of composite insulators, record equipment with potential safety hazards and notify maintenance personnel to repair them. In addition to efficiency issues, traditional inspection methods also bring great safety hazards to technicians.

[0004] Therefore, the present invention discloses a composite insulator defect detection method based on image analysis to solve the above problems. Summary of the invention

[0005] The object of the present invention is to provide a composite insulator defect detection method based on image analysis to solve the problems raised in the prior art.

[0006] To achieve the above object, the present invention provides the following technical solution: a composite insulator defect detection method based on image analysis, the method comprising the following steps:

[0007] S1: The visible light imager and red-ultraviolet imager on the UAV are used to collect visible light images and red-ultraviolet images of defective composite insulators, and the collected composite insulator images are preprocessed;

[0008] S2: Decompose the preprocessed composite insulator visible light image and infrared ultraviolet image to generate low-frequency sub-band image and high-frequency sub-band image; fuse the low-frequency sub-band images according to the matching degree of the regional variance of the low-frequency sub-band images; analyze the fusion weight coefficient of the high-frequency sub-band images according to the pulse coupled neural network model, and fuse the high-frequency sub-band images according to the fusion weight coefficient; evaluate the composite insulator fusion image obtained by the fusion process;

[0009] S3: Use the composite insulator fusion image to identify composite insulator crack defects, composite insulator abnormal temperature and composite insulator ultraviolet discharge; distinguish the defect types of composite insulators according to the defect identification, abnormal temperature identification and ultraviolet discharge identification.

[0010] Visible light imagers are often used to identify defects visible to the naked eye or to photograph targets. They can also be used in conjunction with target recognition algorithms to detect missing or defective composite insulators in power transmission lines. Red and ultraviolet imagers are used to perform infrared imaging of targets such as conductors and insulators in power lines. By analyzing the characteristics of heat distribution, it is determined whether the temperature of the target is within the normal range. Spectral principles are used to achieve corona detection.

[0011] Traditional pure manual tower climbing inspection and helicopters as inspection platforms require inspectors to take images of composite insulators multiple times, which has low inspection efficiency and poses great safety hazards. The present invention uses a drone to take inspection images of composite insulators at one time, and fuses the captured inspection images to form a composite insulator fusion image to identify defect types, realize automated inspection, and improve inspection efficiency.

[0012] According to the above scheme, in S1, the preprocessing includes grayscale, denoising, image normalization, edge enhancement and geometric transformation.

[0013] Denoising methods can reduce the noise in the image and suppress the image background, thereby improving the image quality and signal-to-noise ratio, and reducing the amount of data in subsequent image processing.

[0014] According to the above scheme, S2 includes the following contents:

[0015] S201: performing NSST decomposition on the preprocessed composite insulator visible light image and red-ultraviolet image, respectively, to generate a visible light low-frequency sub-band image, a visible light high-frequency sub-band image, a red-ultraviolet low-frequency sub-band image, and a red-ultraviolet high-frequency sub-band image;

[0016] S202: extracting significant features from the visible light low-frequency sub-band image and the red-ultraviolet low-frequency sub-band image, and calculating the regional variance of the significant features; calculating the matching degree of the regional variance of the significant features of the visible light low-frequency sub-band image and the red-ultraviolet low-frequency sub-band image, and fusing the low-frequency sub-band images according to the matching degree calculation result;

[0017] S203: using a pulse coupled neural network model to calculate a fusion weight coefficient of the high-frequency sub-band image for the visible light high-frequency sub-band image and the red-ultraviolet high-frequency sub-band image; and performing high-frequency sub-band image fusion according to the fusion weight coefficient;

[0018] S204: reconstructing the fused low-frequency and high-frequency sub-band images through an inverse NSST transform to obtain a final composite insulator fusion image;

[0019] S205: performing image quality evaluation on the composite insulator fusion image after fusion processing, and processing the composite insulator image before fusion according to the image quality evaluation result.

[0020] According to the above scheme, in S202, the significant feature extraction includes the following contents:

[0021] S20201: segmenting the visible light low-frequency sub-band image and the red-ultraviolet low-frequency sub-band image respectively;

[0022] S20202: Perform a comprehensive significance score S on the window R formed by the segmentation of S2021 R calculate:

[0023] S R =w C ×C R +w V ×V R +w G ×G R +w E ×E R ;

[0024] Among them, w C 、w V 、w G and w E are comprehensive significance coefficients, which are system preset constants; C R represents the regional contrast of window R; V R represents the regional variance of window R; G R represents the gradient amplitude of window R, E R represents the energy distribution of window R;

[0025] S20203: Set a threshold for the comprehensive significance score, and set the comprehensive significance score S R The window R that is larger than the set threshold is recorded as a significant window, and all significant windows are connected to form a significant feature.

[0026] Salient features usually reflect the global outline, brightness distribution and regional saliency of an image; they can distinguish important information to obtain a clearer and more detailed effect in the final fused image.

[0027] By calculating the comprehensive significance score of the segmented window of the low-frequency sub-band image and quantitatively analyzing the comprehensive significance of the window, the window with the threshold set by the comprehensive significance score can be found more efficiently, providing data support for the subsequent significance feature analysis system.

[0028] According to the above solution, in S202, the low-frequency sub-band image fusion is performed according to the matching degree calculation result, including the following contents:

[0029] S20211: calculating regional variances of significant features of the visible light low-frequency sub-band image and the red-ultraviolet low-frequency sub-band image, and calculating the matching degree of the regional variances of the significant features; the matching degree of the regional variances of the significant features is equal to the regional variance of the significant features of the visible light low-frequency sub-band image divided by the regional variance of the significant features of the red-ultraviolet low-frequency sub-band image;

[0030] S20212: setting a threshold for the matching degree of the regional variance of the significant features; if the regional variance matching degree of the visible light low-frequency sub-band image and the red-ultraviolet low-frequency sub-band image is greater than or equal to the set threshold, the visible light low-frequency sub-band image and the red-ultraviolet low-frequency sub-band image are fused by the weighted average method; if the regional variance matching degree of the visible light low-frequency sub-band image and the red-ultraviolet low-frequency sub-band image is less than the set threshold, the visible light low-frequency sub-band image and the red-ultraviolet low-frequency sub-band image are fused by the grayscale value maximum method;

[0031] The weighted average method will fuse information according to the weight ratio to reduce redundancy; the gray value selection method will directly select pixels with larger gray values ​​to retain more significant details. Reducing redundancy can effectively reduce the amount of data for subsequent calculation and analysis, and improve analysis efficiency; retaining significant details can improve the accuracy of defect judgment for subsequent analysis.

[0032] According to the above solution, in S203, performing high frequency sub-band image fusion according to the fusion weight coefficient includes the following contents:

[0033] S20301: Use the pulse coupled neural network model to extract features and define the extraction layer as U i，k j =L{H j ，l}; Among them, H j，l represents the high-frequency sub-band image, j represents the sequence number of the high-frequency sub-band image, l is the number of channels of the j-th high-frequency sub-band image; L{} represents the feature extraction layer of the pulse coupled neural network model, U i，k j represents the i-th feature layer obtained after the j-th high-frequency sub-band image is extracted through L{}, where i∈[1,I], I represents the total number of feature extraction layers; k represents the number of channels in the i-th feature layer;

[0034] S20302: Perform L2 regularization on the extracted features; the result R obtained by performing L2 regularization on the i-th feature layer of the j-th high-frequency sub-band image i j It can be expressed as: R ij =||U i，k j (x, y)||2; where U i，k j (x, y) represents the point (x, y) in the feature layer U i，k j+ Corresponding points; ||||2 represents regularization operation;

[0035] S20303: Perform convolution calculation on the feature layer after regularization operation, R i j Calculate ^R i j ; The corresponding weight calculation method is:

[0036]

[0037] Among them, ^R i j (x, y) represents the value of the j-th high-frequency sub-band image at the i-th feature layer after the convolution operation at the point (x, y), W i j (x, y) represents the weight value of the i-th feature layer of the j-th high-frequency sub-band image at the point (x, y); J represents the number of high-frequency sub-band images; the weight value corresponding to the visible light high-frequency sub-band image is recorded as W j，i A (x, y); the weight value corresponding to the red-ultraviolet high-frequency sub-band image is recorded as W j，i B (x, y);

[0038] S20304: Calculate the maximum weight: W j A (x, y) = [W j，1：I A (x, y)] max ; W j B (x, y) = [W j，1：I B (x, y)] max ;W j，1：I A (x, y) and W j，1：I B (x, y) represents a 1-dimensional vector; [] max Represents the maximum value calculation function.

[0039] The pulse coupled neural network model can accurately extract significant features, reasonably allocate fusion weights, enhance detail expression and suppress noise interference, effectively retain the significant features of multi-source images, and make the fused image have better performance in terms of visual effects and information retention.

[0040] According to the above solution, in S205, the following contents are included:

[0041] S20501: analyzing image quality evaluation indicators between the composite insulator fusion image and the preprocessed composite insulator visible light image and red-ultraviolet image; normalizing the image quality evaluation indicators; the image quality evaluation indicators include peak signal-to-noise ratio, structural similarity index, information entropy and mean square error;

[0042] S20502: Evaluate image quality according to the image quality evaluation system:

[0043] IQA = [log(PSNR A-F +PSNR B-F +1)+(SSIM A-F +SSIM B-F )α+H F ]÷(1+MSE A-F +MSE B-F );

[0044] Among them, IQA represents the image quality score; PSNR A-F PSNR represents the peak signal-to-noise ratio of the composite insulator fusion image and the composite insulator visible light image. B-F SSIM represents the peak signal-to-noise ratio of the composite insulator fusion image and the composite insulator red-ultraviolet image; A-F SSIM represents the structural similarity index between the composite insulator fusion image and the composite insulator visible light image; B-F represents the structural similarity index between the composite insulator fusion image and the composite insulator red-ultraviolet image; H F Information entropy representing the composite insulator fusion image; MSE A-F MSE represents the mean square error between the composite insulator fusion image and the composite insulator visible light image; B-F represents the mean square error between the composite insulator fusion image and the composite insulator red-ultraviolet image; α represents a constant preset by the system;

[0045] S20503: Setting a threshold for the image quality score; if the image quality score is greater than or equal to the set threshold, no processing is performed; if the image quality score is less than the set threshold, deleting the corresponding composite insulator visible light image and red-ultraviolet image from the database.

[0046] By evaluating the image quality of the composite insulator fusion image, the image fusion effect can be directly quantified to ensure that the final image quality meets the ideal requirements. The visible light images and red-ultraviolet images of the composite insulator that do not meet the requirements can be deleted, which can effectively reduce the required memory capacity and increase the accuracy of subsequent defect identification.

[0047] According to the above scheme, S3 includes the following:

[0048] S301: Composite insulator crack defect identification based on convolutional neural network model;

[0049] S302: Collect infrared image fault samples of abnormal heating of composite insulators and ultraviolet discharge images of different discharge states to build a standard target recognition image set; use Faster-RCNN and YOLO-V3 models to train and test abnormal temperature recognition of composite insulators based on the standard target recognition image set; build a deep convolutional neural network model based on the AlexNet structure, and train and test ultraviolet discharge recognition of composite insulators based on the standard target recognition image set;

[0050] Both Faster-RCNN and YOLO-V3 algorithm models can identify and locate abnormal heating points of composite insulators in infrared images with high accuracy in a relatively complex detection background, and have good detection effects. Based on the AlexNet structure, a deep convolutional neural network model is built, which has the advantages of high accuracy and strong generalization ability, can meet the actual detection needs in engineering, and can provide a basis for intelligent classification methods for flashover of composite insulators.

[0051] S303: using the composite insulator fusion image to identify composite insulator crack defects, composite insulator abnormal temperature and composite insulator ultraviolet discharge; distinguishing defect types of the composite insulator according to defect identification conditions, abnormal temperature identification conditions and ultraviolet discharge identification conditions.

[0052] According to the above scheme, the defect types of the composite insulator include zero-value composite insulator, damaged composite insulator and surface-contaminated composite insulator.

[0053] The difference between zero-value composite insulators and general defective composite insulators is that general defective composite insulators have a higher temperature than normal composite insulators and a higher brightness when collecting infrared images, while zero-value composite insulators have a lower temperature than normal composite insulators and a lower brightness when collecting infrared images. Zero-value composite insulators are composite insulators whose temperature is significantly lower than that of surrounding composite insulators when in working condition.

[0054] According to the inspection regulations for composite insulators, when there is damage on the surface of a composite insulator, it is determined to be a damaged composite insulator. The insulation performance of a damaged composite insulator is low, and the resistance value of the insulation resistance is low, and a certain leakage current is generated. Therefore, the temperature of a damaged composite insulator is higher than that of a normal composite insulator. Damaged composite insulators include composite insulator shed breakage, chip fall-off, and cracks. When a damaged composite insulator is in operation, due to the increase in leakage current, the surface temperature of the composite insulator will increase to a certain extent, and there may also be obvious damage or cracks in the damaged composite insulator.

[0055] The dirt layer attached to the surface of the contaminated composite insulator will form a conductive channel. When the air humidity is high, leakage current will be generated on the surface of the composite insulator, resulting in an increase in the heat power on the surface of the composite insulator, and the temperature of the composite insulator itself will increase to a certain extent.

[0056] The surface contaminated composite insulators are classified into one category for inspection. The main reason is that the fault handling methods are different. The surface contamination problem of composite insulators can be removed by manual wiping, while other defect types must be handled by replacing the composite insulators.

[0057] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention classifies different defects of composite insulators through multi-spectral fusion images, and the image fusion method can accurately detect the defects and types of composite insulators. It solves the problems existing in manual inspection and helicopter inspection, and the use of image information for composite insulator detection is intuitive, and the detection information is more accurate, which provides a new way for composite insulator detection, improves the degree of automation of detection, and achieves the purpose of accurately judging composite insulator defects. It can not only greatly improve the inspection efficiency, but also effectively reduce the impact of environmental conditions on the inspection process, and has good economic and social benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0059] Figure 1 The present invention is a schematic flow chart of a composite insulator defect detection method based on image analysis. DETAILED DESCRIPTION

[0060] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0061] See also Figure 1 The present invention provides a technical solution: a composite insulator defect detection method based on image analysis, characterized in that the method comprises the following steps:

[0062] S1: The visible light imager and red-ultraviolet imager on the UAV are used to collect visible light images and red-ultraviolet images of defective composite insulators, and the collected composite insulator images are preprocessed;

[0063] In S1, preprocessing includes grayscale, denoising, image normalization, edge enhancement, and geometric transformation.

[0064] S2: Decompose the preprocessed composite insulator visible light image and infrared ultraviolet image to generate low-frequency sub-band image and high-frequency sub-band image; fuse the low-frequency sub-band images according to the matching degree of the regional variance of the low-frequency sub-band images; analyze the fusion weight coefficient of the high-frequency sub-band images according to the pulse coupled neural network model, and fuse the high-frequency sub-band images according to the fusion weight coefficient; evaluate the composite insulator fusion image obtained by the fusion process;

[0065] In S2, the following are included:

[0066] S201: performing NSST decomposition on the preprocessed composite insulator visible light image and red-ultraviolet image, respectively, to generate a visible light low-frequency sub-band image, a visible light high-frequency sub-band image, a red-ultraviolet low-frequency sub-band image, and a red-ultraviolet high-frequency sub-band image;

[0067] S202: extracting significant features from the visible light low-frequency sub-band image and the red-ultraviolet low-frequency sub-band image, and calculating the regional variance of the significant features; calculating the matching degree of the regional variance of the significant features of the visible light low-frequency sub-band image and the red-ultraviolet low-frequency sub-band image, and fusing the low-frequency sub-band images according to the matching degree calculation result;

[0068] In S202, the significant feature extraction includes the following contents:

[0069] S20201: segmenting the visible light low-frequency sub-band image and the red-ultraviolet low-frequency sub-band image respectively;

[0070] S20202: Perform a comprehensive significance score S on the window R formed by the segmentation of S2021 R calculate:

[0071] S R =w C ×C R +w V ×V R +w G ×G R +w E ×E R ;

[0072] Among them, w C 、w V 、w G and w E All are comprehensive significance coefficients, which are system preset constants; C R represents the regional contrast of window R; V R represents the regional variance of window R; G R represents the gradient amplitude of window R, E R represents the energy distribution of window R;

[0073] S20203: Set a threshold for the comprehensive significance score, and set the comprehensive significance score S R The window R that is larger than the set threshold is recorded as a significant window, and all significant windows are connected to form a significant feature.

[0074] In S202, the low-frequency sub-band image fusion is performed according to the matching degree calculation result, including the following contents:

[0075] S20211: calculating regional variances of significant features of the visible light low-frequency sub-band image and the red-ultraviolet low-frequency sub-band image, and calculating the matching degree of the regional variances of the significant features; the matching degree of the regional variances of the significant features is equal to the regional variance of the significant features of the visible light low-frequency sub-band image divided by the regional variance of the significant features of the red-ultraviolet low-frequency sub-band image;

[0076] S20212: setting a threshold for the matching degree of the regional variance of the significant features; if the regional variance matching degree of the visible light low-frequency sub-band image and the red-ultraviolet low-frequency sub-band image is greater than or equal to the set threshold, the visible light low-frequency sub-band image and the red-ultraviolet low-frequency sub-band image are fused by the weighted average method; if the regional variance matching degree of the visible light low-frequency sub-band image and the red-ultraviolet low-frequency sub-band image is less than the set threshold, the visible light low-frequency sub-band image and the red-ultraviolet low-frequency sub-band image are fused by the grayscale value maximum method;

[0077] S203: using a pulse coupled neural network model to calculate a fusion weight coefficient of the high-frequency sub-band image for the visible light high-frequency sub-band image and the red-ultraviolet high-frequency sub-band image; and performing high-frequency sub-band image fusion according to the fusion weight coefficient;

[0078] In S203, performing high frequency sub-band image fusion according to the fusion weight coefficient includes the following contents:

[0079] S20301: Use the pulse coupled neural network model to extract features and define the extraction layer as U i，k j =L{H j ，l}; Among them, H j，l represents the high-frequency sub-band image, j represents the sequence number of the high-frequency sub-band image, l is the number of channels of the j-th high-frequency sub-band image; L{} represents the feature extraction layer of the pulse coupled neural network model, U i，k j represents the i-th feature layer obtained after the j-th high-frequency sub-band image is extracted through L{}, where i∈[1,I], I represents the total number of feature extraction layers; k represents the number of channels in the i-th feature layer;

[0080] S20302: Perform L2 regularization on the extracted features; the result R obtained by performing L2 regularization on the i-th feature layer of the j-th high-frequency sub-band image i j It can be expressed as: R i j =||U i，k j (x, y)||2; where U i，k j (x, y) represents the point (x, y) in the feature layer U i，k j+ Corresponding points; ||||2 represents regularization operation;

[0081] S20303: Perform convolution calculation on the feature layer after regularization operation, R i j Calculate to get ^R i j ; The corresponding weight calculation method is:

[0082]

[0083] Among them, ^R i j (x, y) represents the value of the jth high-frequency sub-band image at the i-th feature layer after the convolution operation at the point (x, y), W i j (x, y) represents the weight value of the i-th feature layer of the j-th high-frequency sub-band image at the point (x, y); J represents the number of high-frequency sub-band images; the weight value corresponding to the visible light high-frequency sub-band image is recorded as W j，i A (x, y); the weight value corresponding to the red-ultraviolet high-frequency sub-band image is recorded as W j，iB (x, y);

[0084] S20304: Calculate the maximum weight: W j A (x, y) = [W j，1：I A (x, y)] max ; W j B (x, y) = [W j，1：I B (x, y)] max ;W j，1：I A (x, y) and W j，1：I B (x, y) represents a 1-dimensional vector; [] max Represents the maximum value calculation function.

[0085] S204: reconstructing the fused low-frequency and high-frequency sub-band images through an inverse NSST transform to obtain a final composite insulator fusion image;

[0086] S205: performing image quality evaluation on the composite insulator fusion image after fusion processing, and processing the composite insulator image before fusion according to the image quality evaluation result.

[0087] In S205, the following contents are included:

[0088] S20501: Analyze the image quality evaluation indexes between the composite insulator fusion image and the pre-processed composite insulator visible light image and red-ultraviolet image; perform normalization processing on the image quality evaluation indexes; the image quality evaluation indexes include peak signal-to-noise ratio, structural similarity index, information entropy and mean square error;

[0089] S20502: Evaluate image quality according to the image quality evaluation system:

[0090] IQA = [log(PSNR A-F +PSNR B-F +1)+(SSIM A-F +SSIM B-F )α+H F ]÷(1+MSE A-F +MSE B-F );

[0091] Among them, IQA represents the image quality score; PSNR A-F PSNR represents the peak signal-to-noise ratio of the composite insulator fusion image and the composite insulator visible light image. B-F SSIM represents the peak signal-to-noise ratio of the composite insulator fusion image and the composite insulator red-ultraviolet image;A-F SSIM represents the structural similarity index between the composite insulator fusion image and the composite insulator visible light image; B-F represents the structural similarity index between the composite insulator fusion image and the composite insulator red-ultraviolet image; H F Information entropy representing the composite insulator fusion image; MSE A-F MSE represents the mean square error between the composite insulator fusion image and the composite insulator visible light image; B-F represents the mean square error between the composite insulator fusion image and the composite insulator red-ultraviolet image; α represents a constant preset by the system;

[0092] S20503: Setting a threshold for the image quality score; if the image quality score is greater than or equal to the set threshold, no processing is performed; if the image quality score is less than the set threshold, deleting the corresponding composite insulator visible light image and red-ultraviolet image from the database.

[0093] S3: Use the composite insulator fusion image to identify composite insulator crack defects, composite insulator abnormal temperature and composite insulator ultraviolet discharge; distinguish the defect types of composite insulators according to the defect identification, abnormal temperature identification and ultraviolet discharge identification.

[0094] In S3, include the following:

[0095] S301: Composite insulator crack defect identification based on convolutional neural network model;

[0096] S302: Collect infrared image fault samples of abnormal heating of composite insulators and ultraviolet discharge images of different discharge states to build a standard target recognition image set; use Faster-RCNN and YOLO-V3 models to train and test abnormal temperature recognition of composite insulators based on the standard target recognition image set; build a deep convolutional neural network model based on the AlexNet structure, and train and test ultraviolet discharge recognition of composite insulators based on the standard target recognition image set;

[0097] S303: using the composite insulator fusion image to identify composite insulator crack defects, composite insulator abnormal temperature and composite insulator ultraviolet discharge; distinguishing defect types of the composite insulator according to defect identification conditions, abnormal temperature identification conditions and ultraviolet discharge identification conditions.

[0098] The defect types of composite insulators include zero-value composite insulators, damaged composite insulators and surface-contaminated composite insulators.

[0099] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0100] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

Claims

1. A composite insulator defect detection method based on image analysis, characterized in that: The method comprises the following steps: S1: The visible light imager and red-ultraviolet imager on the UAV are used to collect visible light images and red-ultraviolet images of defective composite insulators, and the collected composite insulator images are preprocessed; S2: Decompose the preprocessed composite insulator visible light image and infrared ultraviolet image to generate low-frequency sub-band image and high-frequency sub-band image; fuse the low-frequency sub-band images according to the matching degree of the regional variance of the low-frequency sub-band images; analyze the fusion weight coefficient of the high-frequency sub-band images according to the pulse coupled neural network model, and fuse the high-frequency sub-band images according to the fusion weight coefficient; evaluate the composite insulator fusion image obtained by the fusion process; S3: Using composite insulator fusion images to identify composite insulator crack defects, composite insulator abnormal temperature, and composite insulator UV discharge; The defect types of composite insulators are distinguished according to the defect identification conditions, abnormal temperature identification conditions and ultraviolet discharge identification conditions.

2. The composite insulator defect detection method based on image analysis according to claim 1, characterized in that: In S1, the preprocessing includes grayscale, denoising, image normalization, edge enhancement and geometric transformation.

3. The composite insulator defect detection method based on image analysis according to claim 1, characterized in that: In S2, the following are included: S201: performing NSST decomposition on the preprocessed composite insulator visible light image and red-ultraviolet image, respectively, to generate a visible light low-frequency sub-band image, a visible light high-frequency sub-band image, a red-ultraviolet low-frequency sub-band image, and a red-ultraviolet high-frequency sub-band image; S202: extracting significant features from the visible light low-frequency sub-band image and the red-ultraviolet low-frequency sub-band image, and calculating the regional variance of the significant features; calculating the matching degree of the regional variance of the significant features of the visible light low-frequency sub-band image and the red-ultraviolet low-frequency sub-band image, and fusing the low-frequency sub-band images according to the matching degree calculation result; S203: using a pulse coupled neural network model to calculate fusion weight coefficients of the high-frequency sub-band images for the visible light high-frequency sub-band images and the red-ultraviolet high-frequency sub-band images; and performing fusion of the high-frequency sub-band images according to the fusion weight coefficients; S204: reconstructing the fused low-frequency and high-frequency sub-band images through an inverse NSST transform to obtain a final composite insulator fusion image; S205: performing image quality evaluation on the composite insulator fusion image after fusion processing, and processing the composite insulator image before fusion according to the image quality evaluation result.

4. The composite insulator defect detection method based on image analysis according to claim 3 is characterized in that: In S202, the significant feature extraction includes the following contents: S20201: segmenting the visible light low-frequency sub-band image and the red-ultraviolet low-frequency sub-band image respectively; S20202: Perform a comprehensive significance score S on the window R formed by the segmentation of S2021 R calculate: S R =w C ×C R +w V ×V R +w G ×G R +w E ×E R ; Among them, w C 、w V 、w G and w E are comprehensive significance coefficients, which are system preset constants; C R represents the regional contrast of window R; V R represents the regional variance of window R; G R represents the gradient amplitude of window R, E R represents the energy distribution of window R; S20203: Set a threshold for the comprehensive significance score, and set the comprehensive significance score S R The window R that is larger than the set threshold is recorded as a significant window, and all significant windows are connected to form a significant feature.

5. The composite insulator defect detection method based on image analysis according to claim 3 is characterized in that: In S202, the low-frequency sub-band image fusion is performed according to the matching degree calculation result, including the following contents: S20211: calculating regional variances of significant features of the visible light low-frequency sub-band image and the red-ultraviolet low-frequency sub-band image, and calculating the matching degree of the regional variances of the significant features; the matching degree of the regional variances of the significant features is equal to the regional variance of the significant features of the visible light low-frequency sub-band image divided by the regional variance of the significant features of the red-ultraviolet low-frequency sub-band image; S20212: setting a threshold for the matching degree of the regional variance of the significant features; if the regional variance matching degree of the visible light low-frequency sub-band image and the red-ultraviolet low-frequency sub-band image is greater than or equal to the set threshold, the visible light low-frequency sub-band image and the red-ultraviolet low-frequency sub-band image are fused by the weighted average method; if the regional variance matching degree of the visible light low-frequency sub-band image and the red-ultraviolet low-frequency sub-band image is less than the set threshold, the grayscale value maximum method is used for the visible light low-frequency sub-band image and the red-ultraviolet low-frequency sub-band image.

6. The composite insulator defect detection method based on image analysis according to claim 3 is characterized in that: In S203, performing high frequency sub-band image fusion according to the fusion weight coefficient includes the following contents: S20301: Use the pulse coupled neural network model to extract features and define the extraction layer as U i，k j =L{H j，l }; Among them, H j，l represents the high-frequency sub-band image, j represents the sequence number of the high-frequency sub-band image, l is the number of channels of the j-th high-frequency sub-band image; L{} represents the feature extraction layer of the pulse coupled neural network model, U i，k j represents the i-th feature layer obtained after the j-th high-frequency sub-band image is extracted through L{}, where i∈[1,I], I represents the total number of feature extraction layers; k represents the number of channels in the i-th feature layer; S20302: Perform L2 regularization on the extracted features; the result obtained after L2 regularization of the i-th feature layer of the j-th high-frequency sub-band image It can be expressed as: Among them U i，k j (x, y) represents the point (x, y) in the feature layer U i，k j+ Corresponding points; ||||2 represents regularization operation; S20303: Perform convolution calculation on the feature layer after regularization operation. Calculation The corresponding weight calculation method is: in, It represents the value of the point (x, y) at the i-th feature layer of the j-th high-frequency sub-band image after the convolution operation. represents the weight value of the i-th feature layer of the j-th high-frequency sub-band image at point (x, y); J represents the number of high-frequency sub-band images; the weight value corresponding to the visible light high-frequency sub-band image is recorded as W j，i A (x, y); the weight value corresponding to the red-ultraviolet high-frequency sub-band image is recorded as W j，i B (x, y); S20304: Calculate the maximum weight: W j A (x, y) = [W j，1：I A (x, y)] max ; W j B (x, y) = [W j，1：I B (x, y)] max ;W j，1：I A (x, y) and W j，1：I B (x, y) represents a 1-dimensional vector; [] max Represents the maximum value calculation function.

7. The composite insulator defect detection method based on image analysis according to claim 3 is characterized in that: In S205, the following contents are included: S20501: analyzing image quality evaluation indicators between the composite insulator fusion image and the preprocessed composite insulator visible light image and red-ultraviolet image; normalizing the image quality evaluation indicators; the image quality evaluation indicators include peak signal-to-noise ratio, structural similarity index, information entropy and mean square error; S20502: Evaluate image quality according to the image quality evaluation system: AQI=[log(PSNR A-F +PSNR B-F +1)+(SSIM A-F +SSIM B-F )α+H F ]÷(1+MSE A-F +MSE B-F ); Among them, IQA represents the image quality score; PSNR A-F PSNR represents the peak signal-to-noise ratio of the composite insulator fusion image and the composite insulator visible light image. B-F SSIM represents the peak signal-to-noise ratio of the composite insulator fusion image and the composite insulator red-ultraviolet image; A-F SSIM represents the structural similarity index between the composite insulator fusion image and the composite insulator visible light image; B-F represents the structural similarity index between the composite insulator fusion image and the composite insulator red-ultraviolet image; H F Information entropy representing the composite insulator fusion image; MSE A-F MSE represents the mean square error between the composite insulator fusion image and the composite insulator visible light image; B-F represents the mean square error between the composite insulator fusion image and the composite insulator red-ultraviolet image; α represents a constant preset by the system; S20503: Setting a threshold for the image quality score; if the image quality score is greater than or equal to the set threshold, no processing is performed; if the image quality score is less than the set threshold, deleting the corresponding composite insulator visible light image and red-ultraviolet image from the database.

8. The composite insulator defect detection method based on image analysis according to claim 1, characterized in that: In S3, Includes the following: S301: Composite insulator crack defect identification based on convolutional neural network model; S302: Collect infrared image fault samples of abnormal heating of composite insulators and ultraviolet discharge images of different discharge states to build a standard target recognition image set; use Faster-RCNN and YOLO-V3 models to train and test abnormal temperature recognition of composite insulators based on the standard target recognition image set; build a deep convolutional neural network model based on the AlexNet structure, and train and test ultraviolet discharge recognition of composite insulators based on the standard target recognition image set; S303: using the composite insulator fusion image to identify composite insulator crack defects, composite insulator abnormal temperature, and composite insulator ultraviolet discharge; The defect types of composite insulators are distinguished according to the defect identification conditions, abnormal temperature identification conditions and ultraviolet discharge identification conditions.

9. The composite insulator defect detection method based on image analysis according to claim 8, characterized in that: The defect types of the composite insulator include zero-value composite insulator, damaged composite insulator and surface-contaminated composite insulator.

Images