基于YOLOv5s-CTC的红外图像光伏组件缺陷检测

Infrared Image PV Module Defect Detection Based on YOLOv5s-CTC

针对无人机航摄光伏组件红外图像分辨率、对比度低且背景复杂,光伏组件小目标缺陷较难检测的问题,基于YOLOv5s提出了一种针对光伏组件缺陷的目标检测模型YOLOv5s-CTC。通过在Backbone中嵌入注意力机制,提高网络对于图像通道和空间位置特征的感知能力。在原网络已有3个检测尺度的基础上增加小目标检测层,增加了网络对微小缺陷目标的检测能力。将原网络Neck部分替换为CsCFPN,使网络更好地融合多尺度的特征信息。实验表明,相较于原网络模型YOLOv5s,所提改进算法在参数量和计算量略有提升的情况下,mAP_(50)为75.9%,mAP_( 50-95)为37.9%,分别提升了2.7%、2.5%,本改进算法对光伏组件缺陷检测具有一定的有效性。

In view of the problems of low resolution,low contrast and complex background of PV module infrared images captured by UAVs,and the difficulty of detecting defects in small targets of PV modules,this paper proposes an object detection model YOLOv5s-CTC based on YOLOv5s.By embedding the attention mechanism in Backbone,the network is able to improve the perception of image channels and spatial location features.By adding a small target detection layer on top of the existing three detection scales of the original network,the network is able to increase the detection capability of the network for small defects.By replacing the Neck of the original network with the CsCFPN,the network is able to better integrate the feature information of multi-scales.Experiments show that,compared with the original network model YOLOv5s,the mAP_(50) of the improved algorithm proposed in this paper is 75.9%and the mAP_(50-95 )is 37.9%,which is an improvement of 2.7%and 2.5%,respectively,although the number of parameters and the amount of computation have a slight increase.The results also prove the effectiveness of this paper s improved algorithm for the detection of defects in photovoltaic modules.