摘要
为了解决无人机巡检光伏组件的效率和识别准确率低的问题,提出了一种基于超分辨率和双池化融合的光伏组件缺陷检测方法。首先,使用生成对抗网络(GAN)对光伏组件图像数据进行扩展,建立可用于光伏电站缺陷目标检测的图像数据集;然后,构建图像超分辨网络,减小图像数据集的噪声和提高局部区域的纹理特征。最后,将单次多边框检测(SSD)的主干网络替换为双池化方式融合的特征提取网络(VGG19_MP),在不提高网络参数的情况下,学习更深层次的纹理结构。结果表明基于超分辨率网络和双池化融合的光伏组件缺陷检测算法精确率达到了98.21%,平均检测时间为0.066 s,相较于对比的检测算法提高了0.9%~9.1%,平均检测时间提高了0.01~0.07 s,为光伏组件缺陷的精确识别提供了更有效的检测方法。
In order to solve the problem of low efficiency and recognition accuracy in unmanned aerial vehicle inspection of photovoltaic modules,the paper propose a defect detection method for the photovoltaic modules based on super-resolution and dual-pooling fusion.Firstly,data augmentation technology(DAT)is used to expand the image data of the photovoltaic modules,and establish an image dataset that can be used for defect target detection in photovoltaic power plants.Then an image super-resolution network is constructed to reduce the noise in the image dataset and improve the texture features of local regions.Finally,the backbone network of the object detection framework is replaced with a feature extraction network fused in a dual pooling manner(VGG19-MP),learning deeper texture structures without increasing network parameters.The results shows that the accuracy of the proposed method is 98.21%,and an average detection time is 0.066 seconds.Compared with the comparative detection algorithms,the accuracy improves by 0.9~9.1%,and the average detection time increases by 0.01~0.07 seconds,providing a more effective detection method for the precise identification of photovoltaic module defects.
作者
艾上美
周剑峰
张必朝
张涛
王红斌
AI Shangmei;ZHOU Jianfeng;ZHANG Bichao;ZHANG Tao;WANG Hongbin(Chuxiong Power Supply Bureau of Yunnan Power Grid Co.,Ltd.,Chuxiong 675000,China;School of Information Engineering and Automation,Kunming University of Science and Technology,Kunming 650000,China)
出处
《智慧电力》
北大核心
2023年第12期53-58,共6页
Smart Power
基金
国家自然科学基金资助项目(61966021)。
关键词
超分辨率网络
双池化融合
无人机巡检
数据增强
目标检测框架
super-resolution network
dual-pooling fusion
drone inspection
data enhancement
target detection framework
