Insulator partial discharge detection method and system based on multiband image fusion
Through multi-band image fusion technology, the spatial and spectral characteristics of local discharge of insulators are extracted, and the convolutional neural network is used for detection, which solves the accuracy and safety of local discharge detection of insulators in the existing technology, and realizes the accuracy of surface discharge of insulators Positioning and early signal capture.
Patent Information
- Application Number
- CN202411355137.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-27
- Publication Date
- 2024-11-12
AI Technical Summary
The existing insulator partial discharge detection methods have intervening measurements that cause distortion on the surface electric field distribution, high labor intensity and high risk, and the non-contact method has low accuracy and sensitivity, so it is impossible to effectively and promptly detect early local discharge phenomena.
Using multi-band image fusion technology, the multi-band image of local discharge of insulators is obtained, spatial and spectral multi-modal features are extracted, and feature fusion and discharge target detection are used to achieve accurate positioning of insulator surface discharge and Early signal capture.
It improves the accuracy and accuracy of local discharge detection of insulators, can accurately capture early signals, and realizes accurate positioning of insulator surface discharge, reducing the risk and labor intensity of the detection process.
Smart Images

Figure CN118937932A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of partial discharge detection, and in particular relates to a method and system for detecting partial discharge of an insulator based on multi-band image fusion. Background Art
[0002] In order to prevent large-scale surface flashover of insulators and thus threaten the safe and stable operation of power systems, it is necessary to effectively monitor the current partial discharge on the surface of insulators. At present, the contact detection method is mainly used to detect partial discharge of insulators in power systems, including pulse current method, leakage current method and voltage distribution measurement method, but these detection methods are all intrusive measurements. The introduction of measurement equipment will distort the original surface electric field distribution of the insulator. The measurement process is labor-intensive and has certain risks. At present, there are also some non-contact measurement methods, including visual method, ultrasonic detection method and infrared imaging method. However, the visual method observes strong discharge in a weak light environment or identifies it through the sound of strong discharge. The measurement results are more dependent on subjective experience, and the measurement results cannot be accurately characterized and data management; the ultrasonic partial discharge detection method has low sensitivity and a small measurement range; and the infrared imaging method mainly relies on heat for partial discharge detection, which cannot effectively and timely detect early partial discharge phenomena. Therefore, the effect of non-contact technology currently applied to partial discharge detection of insulators is still not ideal. Summary of the invention
[0003] Considering that partial discharge on the surface of insulators is visible to the human eye, and the corona generated by partial discharge can radiate light energy, heat and ultraviolet rays, and certain physical structural changes occur at the same time, multi-band images can reflect the spatial structural characteristics and spectral reflectivity information of the observed scene, thereby accurately capturing partial discharge features. Therefore, the present invention proposes an insulator partial discharge detection technology based on multi-band image fusion, obtains multi-band images of insulator partial discharge, extracts its spatial and spectral multimodal features, realizes feature fusion and discharge target detection, which can not only accurately capture the early signals of insulator partial discharge, but also realize the precise positioning of insulator surface discharge.
[0004] The technical solution of the present invention is as follows:
[0005] An insulator partial discharge detection method based on multi-band image fusion, the method comprising:
[0006] Acquire multi-band images of the discharge area;
[0007] Preprocessing the multi-band image;
[0008] The preprocessed multi-band image is detected to obtain the target position and category information of the discharge area.
[0009] Furthermore, the multi-band image is preprocessed as follows:
[0010] Firstly, the empirical linear method is used to invert the multi-band image to obtain more accurate and realistic spectral reflectance data of the discharge scene. Then, according to the sparse representation theory, the local and global correlations of the multi-band image in the spatial dimension and the local redundant correlations in the spectral dimension are utilized, and the image low-rank constraint is introduced in the sparse representation framework to remove the noise in the multi-band image.
[0011] Furthermore, the preprocessed multi-band image is detected to obtain the target position and category information of the discharge area:
[0012] Firstly, the spatial and spectral features of the image are extracted by multi-band respectively;
[0013] The extracted spatial features and spectral features are fused to obtain fused features;
[0014] The YOLO detection algorithm based on the convolutional neural network is used to perform discharge detection on the fused features to obtain the target position and category information of the discharge area.
[0015] Furthermore, the spatial features reflect the shape and texture in the image; and the spectral features reflect the spectral response of the image in different bands.
[0016] The present invention also provides an insulator partial discharge detection system based on multi-band image fusion, the system comprising:
[0017] A multi-band partial discharge image acquisition module is used to obtain multi-band images of the discharge area;
[0018] An image preprocessing module, used for preprocessing the multi-band image;
[0019] The feature fusion and detection module is used to detect the preprocessed multi-band image and obtain the target position and category information of the discharge area.
[0020] Furthermore, the multi-band partial discharge image acquisition module is composed of a tunable filter and a CCD camera;
[0021] Wherein, the tunable filter is used to selectively pass light of a specific wavelength band and filter out unnecessary spectral components;
[0022] The CCD camera is used to receive the image data processed by the filter and convert it into an electrical signal.
[0023] Furthermore, the adjustment range of the tunable filter is 10nm-100nm.
[0024] Furthermore, the feature fusion and detection module consists of an input layer, a two-branch convolutional neural network, a fully connected layer and an output layer;
[0025] Among them, the input layer is used to first receive the multi-band discharge image;
[0026] A two-branch convolutional neural network is used to extract spatial and spectral features;
[0027] The fully connected layer is used to process the fusion features and obtain the final detection results;
[0028] The output layer is used to output the detection results of multi-band discharge images.
[0029] Furthermore, the dual-branch convolutional neural network includes a spatial branch and a spectral branch;
[0030] In the spatial branch, a combination of 24 two-dimensional convolutional layers and pooling layers is used to extract spatial features, where the convolution kernel size is 3*3;
[0031] In the spectral branch, 18 one-dimensional convolutional layers and pooling layers are combined to realize spectral feature extraction, and the convolution kernel size is 1*1.
[0032] Compared with the prior art, the present invention has the following advantages:
[0033] (1) The spectral reflectance features contained in the multi-band image described in the present invention can effectively identify the changes in physical structure and chemical properties caused by local discharge on the surface of the insulator, and detect and identify the discharge area by extracting and fusing spatial-spectral features, thereby improving the detection accuracy;
[0034] (2) The main acquisition components of the multi-band partial discharge image acquisition module designed by the present invention can be fixed on the tripod platform, without the need for an external power supply, and the acquisition module is small in size, light in weight, easy to carry, and can be applied to data acquisition of outdoor power equipment;
[0035] (3) The image preprocessing module designed by the present invention fully utilizes the global / local redundant correlation characteristics of the multi-band image space-spectrum to effectively remove the photon noise in the acquisition process and provide accurate and reliable data for the subsequent detection module;
[0036] (4) The feature fusion and detection module designed by the present invention constructs a dual-branch convolutional neural network to extract the spatial and spectral features of multi-band images respectively, and introduces the more efficient YOLO detection algorithm to achieve feature fusion and discharge target detection. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] The accompanying drawings generally illustrate various embodiments by way of example and not limitation, and together with the description and claims, serve to illustrate the embodiments of the invention. Where appropriate, the same reference numerals are used throughout the drawings to refer to the same or similar parts. Such embodiments are illustrative and are not intended to be exhaustive or exclusive embodiments of the present apparatus or method.
[0038] Figure 1 A schematic diagram of the framework of the present invention is shown;
[0039] Figure 2 A schematic diagram of the structure of the image acquisition module of the present invention is shown;
[0040] Figure 3 A schematic diagram of the structure of the feature fusion and detection module of the present invention is shown. DETAILED DESCRIPTION
[0041] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.
[0042] The embodiment of the present invention provides a method for detecting partial discharge of insulators based on multi-band image fusion, comprising:
[0043] To obtain multi-band images of the discharge area, a tunable filter with a band range of 400nm-1300nm and a high-gain CCD camera are used. The operation of the filter and the CCD camera is controlled by relevant multi-band acquisition software, so that the images are automatically acquired, displayed and saved at a certain band interval. The band interval can be manually set according to the actual situation, and the adjustable range is 10nm-100nm.
[0044] Preprocessing the multi-band image;
[0045] The process of collecting discharge images will inevitably be affected by factors such as atmospheric molecular scattering, target surface factors, and solar radiation in the environment. Therefore, there is a certain difference between the spectral reflectance of the collected multi-band image and the true value. Therefore, it is necessary to eliminate these effects through spectral inversion to obtain target spectral reflectance data that is closer to the actual situation. The present invention uses the empirical linear method in the image preprocessing module to implement the inversion process. Assume that the linear expression of the relationship between the brightness value (DN) and the reflectance R of each pixel in the collected discharge image is:
[0046] DN=k×R+b
[0047] By knowing the pixel brightness values of two or more substances and the corresponding spectral reflectance in the standard spectral library, the values of coefficients k and b can be calculated by the least squares method, thereby obtaining the mapping relationship between the brightness value and the actual spectral reflectance and realizing spectral inversion.
[0048] In addition, the present invention also takes into account that during the acquisition of multi-band discharge images, factors such as imaging devices and imaging mechanisms may affect the generation of a certain degree of noise, which is not conducive to subsequent feature extraction and accurate detection of the discharge area. Therefore, image denoising is performed in the image preprocessing module. Since general signals or images have sparseness on some fixed basis, the noise in the image does not have such a property. Therefore, using image sparsity as a priori information for image denoising can achieve noise removal. Assume and They represent multi-band discharge images with and without noise respectively. In order to facilitate model representation, the multi-band image is represented in two-dimensional form, M represents the number of pixels, and L represents the number of bands. The noise model is expressed as:
[0049] Y=X+N
[0050] Where N represents noise. In the sparse representation framework, the local / global redundant correlation of the spatial dimension of the multi-band image and the global redundant correlation of the spectral dimension are fully utilized to construct an image denoising model. Since the rank of the noise-free multi-band image is much smaller than the size of the image itself, it has a low-rank characteristic, while the noisy multi-band spectral image is affected by the randomness of the noise, and its rank is much larger than the noise-free image, which does not meet the low-rank characteristic. Therefore, the rank of the multi-band image is used to constrain the global correlation of the spectral dimension. The objective function optimization model of the denoising process is expressed as:
[0051]
[0052] The first term on the right side of the function is the data fidelity term, which represents the difference between the image before and after denoising. The second and third terms represent the use of sparse representation theory to extract the spatial local / global redundant correlation of multi-band images to achieve image denoising. D represents an overcomplete spectral dictionary, and A represents the corresponding sparse representation coefficient matrix, in which only a small number of elements are non-zero elements, so the 1 norm is used to constrain its sparsity. The fourth term indicates that the rank of the denoised image should be as low as possible, thereby constraining the global correlation of the spectrum. The uniqueness of the objective function optimization solution is guaranteed by the nuclear norm ||·|| * Approximately represents the rank of the noise-free image. By solving the variable separation and alternating optimization method, the denoised multi-band discharge image X can be iteratively obtained.
[0053] The preprocessed multi-band image is detected to obtain the target position and category information of the discharge area.
[0054] The introduced YOLO detection method first divides the original image into 7*7 grids. If the center of the discharge target is in a certain grid, the grid is used to detect the target. Each grid detects 2 corresponding bounding boxes. Each bounding box predicts 5 parameters, including the coordinates of the center point of the bounding box (x, y), width w, height h and confidence C. The confidence indicates the possibility of the current bounding box containing the discharge target and the accuracy of its position. The confidence calculation is expressed as:
[0055] C=P o ×IOU
[0056] Where P o Indicates the probability that the current bounding box contains the discharge target. If it does, P o =1, otherwise P o = 0. IOU represents the intersection-over-union ratio between the bounding box and the true bounding box when the bounding box contains the discharge target. The loss function of the detection network includes target coordinate loss, position loss, confidence loss and category probability loss. The loss function is expressed as:
[0057] Loss=L1((x,y),(x',y'))+L2((w,h),(w',w'))+L3(c,c')+L4((x,y),(p c ,p c '))
[0058] The embodiment of the present invention also provides a method for detecting partial discharge of insulators based on multi-band image fusion, comprising:
[0059] The system consists of three modules: multi-band partial discharge image acquisition, image preprocessing, and feature fusion and detection. It aims to use multi-band images to effectively capture the changes in physical and chemical properties caused by partial discharge on the insulator surface, obtain accurate and reliable multi-band discharge images, make full use of the contextual information of multi-band images in the discharge area, and use the convolutional neural network structure to achieve background error suppression and spatial-spectral feature fusion of the discharge area, determine the category and location information of the discharge target, and achieve end-to-end accurate detection.
[0060] The main components and functions of the partial discharge multi-band image acquisition module are as follows:
[0061] Imaging lens: As the front end of the system, it is responsible for accurately capturing the image information of the target object and providing high-quality raw data for subsequent processing.
[0062] Tunable filter: Immediately following the imaging lens, this filter can be flexibly adjusted to selectively pass light in a specific band, effectively filtering out unnecessary spectral components, and improving the accuracy and pertinence of image data.
[0063] Connecting lens: It plays a bridging role in the system, ensuring the smooth transmission of image data from the imaging lens to the filter, and then from the filter to the CCD camera, thus ensuring the integrity and continuity of the data.
[0064] CCD camera: As the core component of image conversion, the CCD camera receives image data processed by filters and converts it into electrical signals for subsequent digital processing, storage and analysis.
[0065] Portable spectrometer: It not only receives image data from the CCD camera, but also directly receives external spectral information through incident light, performs spectral analysis, provides additional spectral dimension information for image data, and enhances the comprehensiveness and depth of image analysis.
[0066] Laptop: As the center of data processing and display, it receives analytical data from the portable spectrometer, performs real-time display, storage and further advanced analysis, and provides researchers with an intuitive and convenient data processing platform.
[0067] In the acquisition module, the light I entering the module passes through the imaging lens and the connecting lens to form a parallel beam entering the tunable filter, and then forms an image in the CCD camera through a group of connecting lenses. The camera can be directly connected to a portable laptop, and the tunable filter is connected to the USB interface of the laptop through a controller to realize power supply. In order to invert the collected multi-band images, the module also connects the laptop to a portable spectrometer to directly obtain the relative spectral reflectance curve of the scene target. The multi-band acquisition software includes functions such as electrically adjustable filter port query, band selection, band automatic scanning, CCD parameter setting, image display, acquisition and automatic storage. When the multi-band image acquisition is performed by the software, the electrically adjustable filter is first initialized, and then the CCD parameters are adjusted according to the grayscale value of the imaging in different bands, and finally the image is acquired and stored. The main acquisition components of the multi-band partial discharge image acquisition module designed by the present invention can be installed and fixed on the tripod of the pan-tilt, and it is simple to build in practical applications, without the need for an additional external power supply, and the overall volume is small, and the weight is light and thus easy to carry, and it can be applied to the partial discharge multi-band data acquisition of outdoor power equipment.
[0068] Feature fusion and detection module
[0069] Since convolutional neural networks can automatically learn multi-level image features through a combination of convolutional layers, pooling layers, and fully connected layers, they can achieve tasks such as image classification, recognition, and detection. In image detection, convolutional neural networks can accurately give the location and category information of different targets in the scene. The Yolo target detection algorithm (You Only Look Once) is an efficient detection method based on convolutional neural networks. Its core idea is to convert target detection into a regression problem, use the entire image as the input of the network, and obtain the location of the target and its category information. In order to make full use of the multimodal features of space and spectrum in multi-band images and improve detection accuracy, the present invention constructs a dual-branch convolutional neural network, respectively extracts the two-dimensional spatial and one-dimensional spectral features of multi-band images, and introduces the YOLO detection framework to realize feature fusion and discharge target detection. Its network structure is as follows: Figure 3 As shown. After the image preprocessing module, the high signal-to-noise ratio multi-band partial discharge image is used as the input of the dual-branch convolutional neural network. The two branches are used for spatial and spectral feature extraction respectively. The spatial branch uses two-dimensional convolution and the spectral branch uses one-dimensional convolution. In the spatial branch, a combination of 24 two-dimensional convolution layers and pooling layers is used to extract spatial features, where the convolution kernel size is 3*3. In the spectral branch, 18 one-dimensional convolution layers and pooling layers are used to extract spectral features, and the convolution kernel size is 1*1. Then, the spatial-spectral features extracted by the two fully connected layers are fused, and the category probability and position coordinates of the detected object are given.
[0070] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.
Claims
1. A method for detecting partial discharge of insulators based on multi-band image fusion, characterized in that: The method comprises: Acquire multi-band images of the discharge area; Preprocessing the multi-band image; The preprocessed multi-band image is detected to obtain the target position and category information of the discharge area.
2. The insulator partial discharge detection method based on multi-band image fusion according to claim 1 is characterized in that: The preprocessing of the multi-band image is specifically as follows: Firstly, the empirical linear method is used to invert the multi-band image to obtain more accurate and realistic spectral reflectance data of the discharge scene. Then, according to the sparse representation theory, the local and global correlations of the multi-band image in the spatial dimension and the local redundant correlations in the spectral dimension are utilized, and the image low-rank constraint is introduced in the sparse representation framework to remove the noise in the multi-band image.
3. The insulator partial discharge detection method based on multi-band image fusion according to claim 1 is characterized in that: The preprocessed multi-band image is detected to obtain the target position and category information of the discharge area: Firstly, the spatial and spectral features of the image are extracted by multi-band respectively; The extracted spatial features and spectral features are fused to obtain fused features; The YOLO detection algorithm based on the convolutional neural network is used to perform discharge detection on the fused features to obtain the target position and category information of the discharge area.
4. The insulator partial discharge detection method based on multi-band image fusion according to claim 3 is characterized in that: The spatial features reflect the shape and texture in the image; the spectral features reflect the spectral response of the image in different bands.
5. An insulator partial discharge detection system based on multi-band image fusion, characterized in that: The system comprises: A multi-band partial discharge image acquisition module is used to obtain multi-band images of the discharge area; An image preprocessing module, used for preprocessing the multi-band image; The feature fusion and detection module is used to detect the preprocessed multi-band image and obtain the target position and category information of the discharge area.
6. The insulator partial discharge detection system based on multi-band image fusion according to claim 5 is characterized in that: The multi-band partial discharge image acquisition module is composed of a tunable filter and a CCD camera; Wherein, the tunable filter is used to selectively pass light of a specific wavelength band and filter out unnecessary spectral components; The CCD camera is used to receive the image data processed by the filter and convert it into an electrical signal.
7. The insulator partial discharge detection system based on multi-band image fusion according to claim 6, characterized in that: The adjustment range of the tunable filter is 10nm-100nm.
8. The insulator partial discharge detection system based on multi-band image fusion according to claim 5, characterized in that: The feature fusion and detection module consists of an input layer, a two-branch convolutional neural network, a fully connected layer and an output layer; Among them, the input layer is used to first receive the multi-band discharge image; A two-branch convolutional neural network is used to extract spatial and spectral features; The fully connected layer is used to process the fusion features and obtain the final detection results; The output layer is used to output the detection results of multi-band discharge images.
9. The insulator partial discharge detection system based on multi-band image fusion according to claim 5, characterized in that: The dual-branch convolutional neural network includes a spatial branch and a spectral branch; In the spatial branch, a combination of 24 two-dimensional convolutional layers and pooling layers is used to extract spatial features, where the convolution kernel size is 3*3; In the spectral branch, 18 one-dimensional convolutional layers and pooling layers are combined to realize spectral feature extraction, and the convolution kernel size is 1*1.
Citation Information
Patent Citations
Equipment detection system based on multispectral images
CN112381784A
Fiber image highlight detection and elimination method based on hyperspectral imaging
CN113744154A
Insulator surface defect detection method and system, storage medium and equipment
CN118608459A
Hyperspectral Characteristic Band Selection Method and Active Hyperspectral Imaging Device Using the Same
US20230371818A1
Method and apparatus for calculating algae coverage of insulator surface, and storage medium
WO2023284104A1
