基于双通道多尺度注意力机制的光伏板裂缝检测方法

Photovoltaic Panel Crack Detection Method Based on Dual-Channel Multi-Scale Attention Mechanism

针对目前传统边缘检测方法提取出的图像边缘轮廓模糊、不连续等问题,提出一种基于双通道多尺度注意力机制的光伏板裂缝检测方法,实现对图像低级边缘、边界、目标轮廓的检测;首先构建了双通道主干网络,包含语义分支通道和空间细节分支通道;其次,基于多尺度原则构建了多尺度及注意力机制模块,对特征图像的高、宽、通道的维度变换,分配特征权重,在捕捉跨通道信息的同时,还能够捕捉方向感知和位置感知的信息;最后将空洞融合模块融合到语义分支通道中,提升网络提取特征信息的能力。实验结果表明,所提出的算法对光伏板图像边缘检测性能有提升,相较HED、RCF与FCN算法,F_(1)值提升了2.83%、0.37%与1.54%,获得了较为清晰的裂缝图像。

Aiming at the problems of fuzzy and discontinuous image edge contours extracted by traditional edge detection methods,a photovoltaic panel crack detection method based on dual-channel multi-scale attention mechanism is proposed to detect the detections of low-level edges,boundaries,and target contours.Firstly,the dual-channel backbone network was constructed,including the se-mantic branch channel and spatial detail branch channel.Secondly,based on the multi-scale principle,the multi-scale and attention mechanism was built to transform the dimensions of the height,width,and channel for the feature image,allocate the feature weights,capture the cross-channel information,and capture the direction and position information.Finally,the hole fusion module was integrated into the semantic branch channel,enhancing the ability of network to extract the feature information.Experimental re-sults show that the proposed algorithm improves the edge detection performance of photovoltaic panel images.Compared to the holis-tically-nested edge detection(HED),richer convolutional features(RCF)and Fully convolutional network(FCN)algorithms,the F,value of the proposed algorithm was increased by 2.83%,0.37%and 1.54%,respectively,which obtained the clearer crack images.