网络出版:2025-10-31,
纸质出版:2025-10-31
移动端阅览
崔建伟, 王月明. 基于改进YOLOv8的光伏组件缺陷检测研究[J]. 太阳能学报, 2025,(10):189-196.
崔建伟, 王月明. 基于改进YOLOv8的光伏组件缺陷检测研究[J]. 2025, (10): 189-196.
崔建伟, 王月明. 基于改进YOLOv8的光伏组件缺陷检测研究[J]. 太阳能学报, 2025,(10):189-196. DOI: doi:10.19912/j.0254-0096.tynxb.2024-0913.
崔建伟, 王月明. 基于改进YOLOv8的光伏组件缺陷检测研究[J]. 2025, (10): 189-196. DOI: doi:10.19912/j.0254-0096.tynxb.2024-0913.
针对航拍高分辨率图像下光伏组件的鸟粪与遮挡缺陷以及二极管损坏形成光斑缺陷的检测问题
提出改进深度学习模型YOLOv8n结合切片辅助超推理的检测算法。以2667张可见光与热成像图像进行切图后标注鸟粪、遮挡、光斑3种缺陷构建光伏组件数据集。首先
在YOLOv8n的主干网络上
通过添加3层CBS模块
构建小目标检测层
加强小目标特征信息的传递。其次
在Backbone部分添加全局注意力机制
采用通道与空间注意力的框架
使模型能更好地捕捉全局特征信息。基于上述两处对网络结构的改进设计其消融实验
实验结果表明改进模型相较于基础模型的mAP50值和mAP50∶95分别提升4.3%和1.5%;设计改进模型与其他目标检测模型的对比实验
实验结果表明改进模型的准确率优于其他检测模型。最后
结合切片辅助超推理进行先切片后检测
设计添加切片辅助超推理的整体模型对比实验
实验结果表明改进后的YOLOv8-PG+SAHI模型对高分辨率光伏组件图像下小目标缺陷的检测能力最优
准确率可达88.73%。上述实验表明改进模型更适合航拍高分辨率图像下光伏组件的小目标检测。
Aiming at the problem of detecting bird droppings and shading defects as well as light spot defects formed by diode damage in photovoltaic modules under aerial high-resolution images
a detection algorithm combining the improves deep learning model YOLOv8n with slice-assisted super-reasoning is proposed. The PV module dataset is constructed by slicing 2667 visible and thermal imaging images and labelling them with three kinds of defects: bird droppings
shading and light spots. Firstly
on the backbone network of YOLOv8n
a small target detection layer is constructed by adding a 3-layer CBS module to enhance the transmission of small target feature information. Secondly
a global attention mechanism is added to the Backbone part
adopting the framework of channel and spatial attention
so that the model can better capture the global feature information. Based on the above two improvements to the network structure to design its ablation experiments
the experimental results show that the improved model improves the mAP50 value and mAP50:95 by 4.3% and 1.5%
respectively
compared with the base model; to design the comparison experiments between the improved model and the other target detection models
and the experimental results show that the improved model's accuracy is better than that of the other detection models. Finally
combined with slice-assisted hyper-reasoning for slicing before detection
the overall model comparison experiments are designed with the addition of slice-assisted hyper-reasoning
and the experimental results show that the improved YOLOv8-PG+SAHI model is optimal for detecting small target defects in high-resolution photovoltaic module images
and the accuracy rate can reach 88.73%. The above experiments show that the improved model is more suitable for small target detection of PV modules under aerial high-resolution images.
廖智兴, 刘全兵, 黄斯珉, 等. 基于表面改性的光伏组件冷凝除尘特性研究[J]. 太阳能学报, 2022, 43(11): 33-40.
王艳, 申宗旺, 赵洪山, 等. 基于LS-DCGAN的GCSE-DenseNet光伏组件缺陷识别方法[J]. 太阳能学报, 2023, 44(10): 165-172.
HANAFY W A, PINA A, SALEM S A.Machine learning approach for photovoltaic panels cleanliness detection[C]//2019 15th International Computer Engineering Conference (ICENCO). Cairo, Egypt, 2019: 72-77.
陈亚芳, 廖飞, 黄新宇, 等. 多尺度YOLOv5的太阳能电池缺陷检测[J]. 光学精密工程, 2023, 31(12): 1804-1815.
彭自然, 张颖清, 肖伸平. 基于YOLOv5的太阳电池表面缺陷检测[J]. 太阳能学报, 2024, 45(6): 368-375.
YE L, WANG X Y, AN A M, et al.Deeplab-YOLO: a method for detecting hot-spot defects in infrared image PV panels by combining segmentation and detection[J]. Journal of real-time image processing, 2024, 21(2): 52.
王道累, 姚从荣, 李超, 等. 光伏组件热斑效应智能化检测的综述及展望[J]. 太阳能学报, 2024, 45(5): 527-536.
陈辉, 张傲, 孙帅, 等. 基于改进DeepLabv3+的含噪声热红外图像光伏热斑检测方法[J]. 太阳能学报, 2023, 44(11): 23-30.
JI S J, LING Q H, HAN F.An improved algorithm for small object detection based on YOLO v4 and multi-scale contextual information[J]. Computers and electrical engineering, 2023, 105: 108490.
WANG W Z, WU B, LYU J N, et al.Regular and small target detection[C]//MultiMedia Modeling: 25th International Conference. Thessaloniki, Greece, 2019.
王巍, 赵宽, 杨耀权, 等. 基于CSwin的航拍光伏组件红外图像热斑检测方法[J]. 太阳能学报, 2023, 44(10): 142-147.
樊嵘, 马小陆. 面向带钢表面小目标缺陷检测的改进YOLOv7算法[J]. 合肥工业大学学报(自然科学版), 2024, 47(3): 303-308, 316.
马超伟, 张浩, 马新明, 等. 基于改进YOLOv8的轻量化小麦病害检测方法[J]. 农业工程学报, 2024, 40(5): 187-195.
赵陆阳, 刘长良, 刘卫亮, 等. 基于CBAM-DSC-UNet模型的时空风速预测算法[J]. 太阳能学报, 2024, 45(10): 497-505.
李宝奇, 黄海宁, 刘纪元, 等. 基于改进SSD的合成孔径声纳图像感兴趣小目标检测方法[J]. 电子学报, 2024, 52(3): 762-771.
刘传洋, 吴一全, 刘景景. 基于视觉的输电线路金具锈蚀缺陷检测方法研究进展[J]. 仪器仪表学报, 2024, 45(3): 286-305.
丁习文, 程宏昌, 袁渊, 等. 基于改进SSD算法的紫外像增强器视场瑕疵检测方法[J]. 兵工学报, 2024, 45(12): 4350-4363.
王金鹏, 何萌, 甄乾广, 等. 基于COF-YOLO v8n的油茶果静、动态检测计数[J]. 农业机械学报, 2024, 55(4): 193-203.
王安静, 袁巨龙, 朱勇建, 等. 基于改进YOLOv8s的鼓形滚子表面缺陷检测算法[J]. 浙江大学学报(工学版), 2024, 58(2): 370-380, 387.
0
浏览量
0
下载量
0
CSCD
关联资源
相关文章
相关作者
相关机构
京公网安备11010802024621