改进YOLOV5的密集行人检测算法研究
Research on Improved YOLOV5 Algorithm for Dense Pedestrian Detection
周龙刚 1魏本昌 1魏鸿奥 1张路 1刘洋1
作者信息
- 1. 湖北汽车工业学院 电气与信息工程学院,湖北 十堰 442002
- 折叠
摘要
针对密集行人漏检以及检测精度低等问题,提出一种改进YOLOV5的特征融合算法FPCA-YOLOV5.首先,通过添加空间池化金字塔结构SPPFCSPC与CA注意力相结合,使模型具有更强的表达能力.其次,在网络中增加PP模块,检测层由3层变为4层,从而对小目标的检测更加准确.最后,设计了一种全新的下采样机制CAConv,使网络在处理特征图时更加关注于重要通道.实验结果表明,在公共数据集WiderPerson上,改进的YOLOV5模型相较于原始模型的召回率提升了3.4%,平均精度提升了2.3%.整体性能相较于原始模型明显提升,证明了FPCA-YOLOV5算法在目标检测中的有效性.
Abstract
A feature fusion algorithm FPCA-YOLOV5 with improved YOLOV5 is proposed to address the issues of missed detection of dense pedestrians and low detection accuracy.Firstly,by combining the spatial pooling pyramid structure SPPFCSPC with CA attention,the model has stronger expressive power.Secondly,adding PP modules to the network and changing the detection layer from three to four layers can achieve more accurate detection of small targets.Finally,a novel downsampling mechanism,CAConv,was designed to enable the network to focus more on important channels when processing feature maps.The experimental results show that on the public dataset WiderPerson,the improved YOLOV5 model has increased recall by 3.4%and average accuracy by 2.3%compared to the original model.The overall perfor-mance is significantly improved compared to the original model,demonstrating the effectiveness of the FPCA-YOLOV5 algorithm in object de-tection.
关键词
YOLOV5/行人检测/特征融合/注意力机制/小目标检测Key words
YOLOV5/pedestrian detection/feature fusion/attention mechanism/small object detection引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据