融合空间深度信息的光伏板缺陷检测

Fusion of Spatial Depth Information for Defect Detection in Photovoltaic Panels

光伏板的转换效率通常受到缺陷的限制,这些缺陷会降低其性能和寿命,高精度的光伏板缺陷检测算法对于确保其性能和可靠性有重要作用。文章针对工业场景下光伏板缺陷检测技术存在的特征提取能力不足、检测精度低的问题,提出了一种融合空间深度信息的光伏板缺陷检测模型(FSDNet)。FSDNet使用YOLOv5作为基础模型,设计空间深度信息融合模块使得特征图空间信息和深度信息有效融合,增强模型全局语义信息的编码能力。试验结果表明,FSDNet较YOLOv5s基础模型的平均检测精度提升6.00%,达到86.70%,单张图像平均检测速度达到209.49FPS。

The conversion efficiency of photovoltaic panels is usually limited by defects,which can reduce their performance and lifespan.High precision photovoltaic panel defect detection algorithms play an important role in ensuring their performance and reliability.The article proposes a photovoltaic panel defect detection model(FSDNet)that integrates spatial depth information to address the issues of insufficient feature extraction capability and low detection accuracy in photovoltaic panel defect detection technology in industrial scenarios.FSDNet uses YOLOv5 as the basic model and designs a spatial depth information fusion module to effectively fuse feature map spatial information and depth information,enhancing the encoding ability of the model's global semantic information.The experimental results indicate that,The average detection accuracy of FSDNet compared to the YOLOv5 s basic model has increased by 6.00%,reaching 86.70%,and the average detection speed of a single image has reached 209.49 FPS.