基于ECA和BIFPN的低照度环境下的行人目标检测算法  

Pedestrian target detection algorithm in low illumination environment based on ECA and BIFPN

作　　者：相敏月 涂振宇[1] 孙逸飞 方强 马飞 XIANG Minyue;TU Zhenyu;SUN Yifei;FANG Qiang;MA Fei(School of Information Engineering,Nanchang Institute of Technology,Nanchang 330000,China)

机构地区：[1]南昌工程学院信息工程学院,南昌330000

出　　处：《智能计算机与应用》2023年第9期189-193,共5页Intelligent Computer and Applications

基　　金：江西省科技厅重点研发计划(20151BBE50077);江西省水利厅科技项目(KT201639)。

摘　　要：针对在低照度环境下多尺度行人目标检测准确率低的问题,本文提出了一种基于改进YOLOv5s的行人目标检测模型BE-YOLOv5s。首先,在YOLOv5s的主干网络中融入ECA通道注意力机制,突出目标特征同时抑制低照度环境的干扰;其次,引入加权双向特征金字塔BIFPN,增强特征融合,提升行人检测精度;最后,采用可见光图像和红外图像这两组数据进行对比研究。实验结果表明,改进后的BE-YOLOv5s模型在两种数据集上的平均精度均值mAP均有所提升,同时保持了原算法的高实时性。Aiming at the low accuracy of multi-scale pedestrian target detection in a low illumination environment,a pedestrian target detection model based on improved YOLOv5s is proposed.Firstly,the convolutional attention module ECA attention mechanism is integrated into YOLOv5s backbone network to highlight target characteristics and suppress interference from a low illumination environment.Then replace the original feature pyramid network,PANet,and introduce weighted bi-directional feature pyramid BIFPN to enhance feature fusion and improve pedestrian detection accuracy.Finally,the visible image and infrared image are used for comparative study.The experimental results show that the average precision mean mAP of the improved BE-YOLOv5s algorithm model on both datasets is improved,and the high real-time performance of the original algorithm is maintained.

关 键 词：行人检测 注意力机制 加权双向特征金字塔 低照度环境 YOLOv5s 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]