基于CSwin的航拍光伏组件红外图像热斑检测方法  

A HOTSPOT DETECTION METHOD FOR INFRARED IMAGES OF AERIAL PHOTOVOLTAIC MODULES BASED ON CSWIN

作　　者：王巍 赵宽 杨耀权[1] 翟永杰[1] Wang Wei;Zhao Kuan;Yang Yaoquan;Zhai Yongjie(Department of Automation,North China Electric Power University,Baoding 071003,China)

机构地区：[1]华北电力大学自动化系,保定071003

出　　处：《太阳能学报》2023年第10期142-147,共6页Acta Energiae Solaris Sinica

基　　金：国家自然科学基金(U21A20486);河北省自然科学基金(F2021502008)。

摘　　要：针对航拍光伏组件红外图像低对比度、背景复杂和热斑小目标难检测的问题,提出一种融合Swin transformer的跨阶段局部网络(CSwin),降低参数量的同时可捕获图像全局位置和空间信息,并以CSwin为基础模块构建多尺度特征路径聚合网络(MPC)加强多尺度特征的信息交互以进一步提高小目标检测能力。对航拍光伏红外图像数据集进行定性和定量实验,证明该方法在航拍光伏红外图像热斑检测任务上的有效性。Aiming at the problems of low contrast,complex background and difficult detection of small hot spots in aerial photovoltaic infrared images,a cross-stage partial network incorporating Swin transformer(CSwin)is proposed to reduce the number of parameters and capture the global position and spatial information of the image.The multi-scale feature path aggregation network based on CSwin(MPC)is constructed to strengthen the information interaction of multi-scale features and further improve the small target detection ability.The qualitative and quantitative experiments are carried out on the aerial photovoltaic infrared image data set,and the effectiveness of the method in aerial photovoltaic infrared image hot spot detection in proved.

关 键 词：光伏组件 目标检测 深度学习 YOLOv5 多尺度特征融合 红外图像 热斑检测 Swin TRANSFORMER 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]