计算机工程与应用2024，Vol.60Issue(15)：336-343,8.DOI:10.3778/j.issn.1002-8331.2312-0089
基于改进YOLOv7-tiny的光伏电池缺陷检测算法
Solar Cell Defect Detection Algorithm Based on Improved YOLOv7-tiny
摘要
Abstract
To address the issue of unstable solar energy conversion efficiency in photovoltaic cells and improve the quality of photovoltaic cells,this article proposes a photovoltaic cell defect detection algorithm,PSD-YOLO,based on the improved YOLOv7-tiny with the introduction of the lightweight convolution module PSDConv.Initially,the DW(Depthwise)con-volution in GSConv is replaced with Partial convolution,reducing memory access and improving detection speed.Addi-tionally,the decoupled fully connected attention(DFC)mechanism from GhostNetv2 is incorporated,enhancing the capa-bility of lightweight algorithms to detect complex defect types in photovoltaic cells while maintaining deployability.In the loss function section,CIoU is replaced with EIoU,accelerating convergence and improving regression accuracy.Experi-mental results demonstrate that the PSD-YOLO model reduces parameter and computational complexity by 18.3%and 16.7%,respectively,compared to the YOLOv7-tiny model.With a model size of only 4.9 million,it achieves a 5.3 per-centage points improvement in mAP@0.5,attaining higher detection performance while achieving a smaller model size.关键词
YOLOv7-tiny/光伏电池/缺陷检测/注意力机制/损失函数Key words
YOLOv7-tiny/solar cell/defect detection/attention mechanism/loss function分类
信息技术与安全科学引用本文复制引用
徐威,李为相,方志,孙圆,陈闯..基于改进YOLOv7-tiny的光伏电池缺陷检测算法[J].计算机工程与应用,2024,60(15):336-343,8.基金项目
国家自然科学基金(62303217) （62303217）
江苏省高等学校基础科学(自然科学)研究项目(23KJB510006) （自然科学）
浙江大学工业控制技术全国重点实验室开放课题(ICT2023B34) （ICT2023B34）
寒地建筑综合节能教育部重点实验室开放基金(JLJZHDKF022023009). （JLJZHDKF022023009）
