摘要
无人机自动化巡检是解决大型分布式光伏系统运维需求的有效方案。无人机航拍产生大量光伏板图像数据,需要算法实现更高的识别精度和更快的识别速度,为此提出一种改进的SSD算法,用于检测光伏组件缺陷。新算法在原有SSD算法中嵌入注意力机制,并使用迁移学习策略提高检测速度和准确率,能够对光伏组件普遍存在的玻璃破碎、受光面发黄、灰尘等进行自动识别和分类。通过与Faster-RCNN、YOLO3、VGG16-SSD算法对比,实验结果表明,改进SSD算法在识别准确率、召回率和检测速度方面表现良好,能有效提升光伏组件缺陷识别的效率。
Automated inspection of UAV is an effective solution to meet the operation and maintenance requirements of large-scale distributed photovoltaic system. Unmanned aerial vehicle produces a large number of PV panel image data, which requires a algorithm to achieve higher recognition accuracy and faster recognition speed. Therefore, an improved SSD algorithm is proposed to detect PV module defects.The new algorithm embeds the attention mechanism into the original SSD algorithm, and uses the transfer learning strategy to improve the detection speed and accuracy. It can automatically identify and classify the glass breakage, yellowing of the light receiving surface, dust, etc., which are common in photovoltaic modules. Compared with Faster-RCNN, YOLO3 and VGG16-SSD algorithms, the experimental results show that the improved SSD algorithm performs well in recognition accuracy, recall rate and detection speed, and can effectively improve the efficiency of photovoltaic module defect recognition.
作者
钟泳松
徐凌桦
周克
ZHONG Yongsong;XU Linghua;ZHOU Ke(The Electrical Engineering College,Guizhou University,Guiyang 550025,China)
出处
《微处理机》
2022年第1期22-25,共4页
Microprocessors
基金
国家自然科学基金(61861007)
贵州省工业攻关项目(黔科合支撑[2019]2152)2020年
贵州大学混合式课程建设项目“计算机控制技术”(2020030)。
关键词
迁移学习
SSD算法
深度学习
注意力机制
光伏板检测
Transfer learning
SSD algorithm
Deep learning
Attention mechanism
Photovoltaic panel detection