摘要
针对二轮车驾乘人员头盔佩戴问题,提出一种基于YOLOv3-tiny的轻量化头盔检测模型。将原始模型主干网络进行轻量化处理,减少检测模型的参数量,在网络中添加U型特征二次融合模块,引入关于边框距离的DIoU损失函数,用于提高检测模型的特征提取能力和识别精度。在测试集上的实验表明,改进后的模型相比原YOLOv3-tiny模型表现出更高的查全率和mAP及F1指标,且在保持较小参数量的同时,具有优于深度网络YOLOv3的检测性能。
Aimed at the helmet wearing problem of two-wheeled vehicle drivers and passengers,a lightweighted helmet detection model based on YOLOv3-tiny is proposed.The original model backbone network was light-weighted to reduce the amount of parameters of the detection model,and a U-shaped feature secondary fusion module was added to the network.The DIoU loss function about the distance of bounding boxes was introduced to improve the feature extraction ability of the detection model and recognition accuracy.Experiments on the test set show that the improved model exhibits higher recall rate and mAP and F1-score than the original YOLOv3-tiny model,and it has better detection performance than the deep network YOLOv3 while maintaining a small amount of parameters.
作者
杨国亮
李世聪
邹俊峰
龚家仁
Yang Guoliang;Li Shicong;Zou Junfeng;Gong Jiaren(School of Electrical Engineering and Automation,Jiangxi University of Science and Technology,Ganzhou 341000,Jiangxi,China)
出处
《计算机应用与软件》
北大核心
2024年第5期147-152,共6页
Computer Applications and Software
基金
国家自然科学基金项目(51365017)
江西省教育厅科技计划项目(GJJ190450)。
关键词
头盔检测
轻量化网络
特征融合
边框损失函数
Helmet detection
Lightweighted network
Feature fusion
Bounding box loss function
