CN-122016845-A - Power equipment metal corrosion detection method and device based on multichannel multi-mode imaging

CN122016845ACN 122016845 ACN122016845 ACN 122016845ACN-122016845-A

Abstract

The invention discloses a method and a device for detecting metal corrosion of power equipment based on multi-channel multi-mode imaging, and relates to the field of metal corrosion detection. The method comprises the steps of obtaining multi-mode images synchronously collected by the multi-channel imaging equipment, dividing a local detection area of a key component based on an equipment structure model, standardizing each channel image to obtain a multi-spectrum image, extracting initial characteristics through an independent encoder, obtaining fusion characteristics through cross-mode attention fusion module learning mode semantic association, combining environment illumination information and pre-stored statistics, carrying out residual fusion after normalizing the characteristics through an environment self-adaptive correction module, and generating final unified characteristics, thereby judging corrosion conditions. The invention can overcome the complex illumination influence of strong light, weak light, color temperature change and the like, ensures stable distribution of fusion characteristics, realizes identification of tiny and early corrosion characteristics, meets the requirement of power grid operation and maintenance on defect detection accuracy, and greatly improves the reliability and stability of outdoor scene detection.

Inventors

DENG SHANQUAN
ZHU JUNWEI
ZHANG XINGSEN
BIAN MEIHUA
CHEN HENG
HE YUYIN
Ji Shuolei

Assignees

广西电网有限责任公司电力科学研究院

Dates

Publication Date: 20260512
Application Date: 20251217

Claims (10)

1. The method for detecting the metal corrosion of the power equipment based on the multi-channel multi-mode imaging is characterized by comprising the following steps of: s1, acquiring multi-mode image data of the surface of power equipment to be detected by multi-channel imaging equipment, wherein the multi-mode image data comprises image data of a visible spectrum and image data of a short-wave infrared spectrum; S2, based on a predefined power equipment structure model, segmenting a local detection area comprising at least one key component from the multi-mode image data; S3, carrying out standardized preprocessing on the image data of each channel in the local detection area to obtain spectral image data corresponding to each channel; S4, respectively extracting the characteristics of each spectrum image data by using a plurality of independent encoders to obtain initial characteristic representations corresponding to each mode; S5, inputting the initial characteristic representation of each mode into a cross-mode attention fusion module, and learning semantic association among modes through a multi-head self-attention mechanism to generate and fuse mode invariant characteristics of the cross-mode to obtain fusion characteristics corrected by cross-mode information; S6, inputting the fusion features into an environment self-adaptive feature correction module, carrying out normalized calibration on the features based on the acquired environment illumination information and prestored statistics, and generating final unified feature representation through residual fusion; And S7, judging whether metal corrosion exists on the surface of the key component in the local detection area according to the unified characteristic representation.
2. The method for detecting metal corrosion of power equipment based on multi-channel multi-mode imaging according to claim 1, wherein the image data of the visible spectrum comprises three independent channels which are respectively arranged in blue, green and red wave band regions, and the image data of the short wave infrared spectrum comprises a plurality of independent channels which are uniformly distributed in different wavelength ranges.
3. The method for detecting metal corrosion of power equipment based on multi-channel and multi-mode imaging according to claim 1, wherein the preprocessing further comprises ambient illumination compensation, specifically: and carrying out self-adaptive contrast enhancement or brightness equalization on the image data of each channel according to the ambient illumination intensity recorded during acquisition.
4. The method for detecting metal corrosion of an electrical device based on multi-channel and multi-modality imaging according to claim 1, wherein the cross-modality attention fusion module performs the following operations: s51, splicing initial characteristic representations of all modes to form a unified characteristic sequence; S52, mapping the characteristic sequences into a query matrix, a key matrix and a value matrix respectively; s53, calculating the attention weight of each position in the characteristic sequence to all positions through a multi-head self-attention mechanism, and carrying out weighted summation on the value matrix to obtain attention output; S54, linearly projecting the attention output to generate a mode invariant feature; S55, carrying out weighted fusion on the mode invariant feature and the corresponding initial feature representation in a residual mode, and outputting corrected fusion features.
5. The method for detecting metal corrosion of an electrical device based on multi-channel and multi-modality imaging according to claim 1, wherein the environment adaptive feature correction module performs the following operations: S61, obtaining characteristic statistics matched with the current environment illumination condition, wherein the statistics comprise mean values and variances, and different statistics sets are pre-stored for different illumination levels; s62, carrying out centering and variance normalization processing on the input fusion features by utilizing the matched statistics; and S63, carrying out residual weighted fusion on the processed characteristics and the original fusion characteristics, and outputting a final unified characteristic representation, wherein the fusion weight of the residual weighted fusion process is adjusted according to the ambient lighting condition.
6. The method for detecting metal corrosion of electric equipment based on multi-channel and multi-mode imaging according to claim 5, wherein the statistic sets of different illumination levels are obtained by simulating or actually collecting samples under different weather and different time periods in a training stage and grouping and calculating.
7. The method for detecting metal corrosion of an electrical device based on multi-channel and multi-modality imaging according to claim 1, wherein said determining whether metal corrosion exists on the surface of the critical component in the local detection area according to the unified feature representation comprises: s71, inputting the unified feature representation into a classification network, and acquiring a corrosion probability map of each pixel point or region in the local detection region; S72, combining the prior knowledge of the space structure of the key component, and carrying out space constraint aggregation on the corrosion probability map to obtain an aggregation result; s73, judging the corrosion state based on the aggregation result, and outputting the range information of the corrosion area.
8. The method for detecting metal corrosion of electric equipment based on multi-channel multi-mode imaging according to claim 1 is characterized in that the independent encoders adopt a mixed architecture of shared bottom layer parameters and independent high layer parameters, wherein the bottom layer parameters are used for extracting basic edges and texture features, and the high layer parameters are used for extracting semantic features of spectral features of corresponding channels.
9. The method for detecting metal corrosion of electrical equipment based on multi-channel and multi-mode imaging according to claim 1, wherein the local detection area is segmented, and different spatial context ranges are adopted for different types of key components of the electrical equipment.
10. A multi-channel multi-modality imaging-based power equipment metal corrosion detection device, characterized in that it employs the method of any one of claims 1 to 9, comprising: The image data acquisition module is used for acquiring multi-mode image data of the surface of the power equipment to be detected by the multi-channel imaging equipment synchronously, wherein the multi-mode image data comprises image data of a visible spectrum and image data of a short wave infrared spectrum; An image segmentation module for segmenting a local detection region comprising at least one key component from the multi-modal image data based on a predefined power device structural model; the image processing module is used for carrying out standardized preprocessing on the image data of each channel in the local detection area to obtain spectrum image data corresponding to each channel; the initial feature extraction module is used for respectively extracting the features of each spectrum image data by utilizing a plurality of independent encoders to obtain initial feature representation corresponding to each mode; The cross-modal attention fusion module is used for learning semantic association among the modes through a multi-head self-attention mechanism by the initial feature representation of each mode, generating and fusing the cross-modal mode invariant features, and obtaining fusion features corrected by the cross-modal information; the environment self-adaptive feature correction module is used for carrying out normalized calibration on the features based on the collected environment illumination information and prestored statistics, and generating final unified feature representation through residual fusion; and the corrosion judging module is used for judging whether metal corrosion exists on the surface of the key component in the local detection area according to the unified characteristic representation.

Description

Power equipment metal corrosion detection method and device based on multichannel multi-mode imaging Technical Field The invention relates to the technical field of metal corrosion of power equipment, in particular to a method and a device for detecting metal corrosion of power equipment based on multichannel and multi-mode imaging. Background The metal parts of the power network are exposed to severe outdoor environments for a long time, and corrosion problems of the metal parts directly threaten the safety of the power network, and equipment failure and even serious accidents can be caused. Therefore, developing a metal corrosion detection technology which can realize early and accurate operation and does not influence the normal operation of the power grid has become an urgent need for ensuring the safe and stable operation of the power system. At present, the metal corrosion detection technology is mainly divided into contact type and non-contact type. The contact method (such as ultrasonic detection) needs power failure operation, and has low efficiency and narrow coverage. In the non-contact optical imaging technology, the single mode has the inherent defects that the visible light imaging is greatly influenced by illumination, overexposure is easy to occur under strong light, details are lost under weak light, the short wave infrared imaging is sensitive to early corrosion, but the specific wave band is easily interfered by water vapor, and the reliability is reduced in humid weather. Aiming at the defects of the current detection, a multi-mode fusion technology is proposed, and the detection capability is improved by fusing visible light and short-wave infrared image information. However, the existing general multi-mode fusion scheme still lacks adaptability to outdoor complex environments when applied to the specific scene of outdoor power equipment corrosion detection. The fusion process of the method can not be dynamically associated with the illumination conditions which are actually changeable during acquisition, so that the problem of unstable detection performance in different weather and different time periods is caused. Disclosure of Invention Aiming at the problem that in the prior art, the detection performance is unstable under different weather and different time periods due to the fact that the equipment corrosion image is not dynamically associated with the actual changeable illumination conditions during acquisition, the invention provides the multi-channel multi-mode imaging-based power equipment metal corrosion detection method and device, which can overcome the influence of complex illumination conditions such as strong light in sunny days, weak light in cloudy days, change of color temperature in the morning and evening time period and the like, ensure the data acquired under different weather and different time, ensure that the fusion characteristic distribution is stable, and greatly improve the reliability and stability of the detection result in a real outdoor scene. The specific technical scheme is as follows: In a first aspect, the invention provides a method for detecting metal corrosion of an electrical device based on multi-channel multi-mode imaging, comprising the following steps: s1, acquiring multi-mode image data of the surface of power equipment to be detected by multi-channel imaging equipment, wherein the multi-mode image data comprises image data of a visible spectrum and image data of a short-wave infrared spectrum; S2, based on a predefined power equipment structure model, segmenting a local detection area comprising at least one key component from the multi-mode image data; S3, carrying out standardized preprocessing on the image data of each channel in the local detection area to obtain spectral image data corresponding to each channel; S4, respectively extracting the characteristics of each spectrum image data by using a plurality of independent encoders to obtain initial characteristic representations corresponding to each mode; S5, inputting the initial characteristic representation of each mode into a cross-mode attention fusion module, and learning semantic association among modes through a multi-head self-attention mechanism to generate and fuse mode invariant characteristics of the cross-mode to obtain fusion characteristics corrected by cross-mode information; S6, inputting the fusion features into an environment self-adaptive feature correction module, carrying out normalized calibration on the features based on the acquired environment illumination information and prestored statistics, and generating final unified feature representation through residual fusion; And S7, judging whether metal corrosion exists on the surface of the key component in the local detection area according to the unified characteristic representation. Preferably, the image data of the visible spectrum comprises three independent channels which are respectively arranged in bl