基于双阶段轻量YOLO的红外行人伪影检测算法北大核心CSCD

由于镜面回波效应,红外图像采集过程中行人不可避免出现反射倒影(本文简称“伪影”)区域,此时对后续行人检测会造成一定程度影响,对上述情况,本文提出一种基于轻量YOLO(You Only Look Once)的双阶段网络检测框架,先检测“行人-伪影”联合区域再精准定位伪影位置。首先,针对YOLOv5s轻量检测算法进行改进,使用LSM(Light Sample Module)双分支结构替换原下采样部分,并嵌入注意力机制来提高模型的特征整合能力,实现红外图像的背景过滤和联合区域提取。其次,对联合区域进行无失真矩形填充保持原始特征,设计轻量级行人伪影定位网络LS YOLO(Light Structur YOLO)检测联合区域获得最终的伪影位置坐标。实验结果表明,本文算法能够满足实时检测要求,在数据集中,相比其他算法获得更好的检测效果,行人伪影的检测正确率达到9545%。

Forged shadow detection algorithm based on two stagelightweight YOLO in infrared pedestrian

Due to the effect of mirror reflection echo,there are inevitably thermal radiation reflection areas (hereinafter referred to as "forged shadow") in the process of pedestrian infrared image acquisition,which affect the pedestrian detection to a certain extent.In this paper,a two stage network detection framework based on lightweight YOLO is proposed to detect the "pedestrian and forged shadow" joint area and accurately locate the shadow location.Firstly,the YOLOv5s lightweight detection algorithm is improved by replacing the original lower sampling part with LSM (Light Sample Module) double branch structure,and the attention mechanism is embedded to improve the feature integration ability of the model to achieve the background filtering and joint region extraction of infrared images.Secondly,the joint area is filled with distortion free rectangle to maintain the original features,and a lightweight shadow of pedestrian location network LS YOLO (Light Structur YOLO) is designed to detect the joint area to obtain the final position coordinates.The experimental results show that the proposed algorithm can meet the real time detection requirements.In the data set,compared with other algorithms,the detection accuracy of forged shadow reaches 95.45%.