改进YOLOv7-tiny的安全帽实时检测算法  

Improved YOLOv7-tiny Algorithm for Safety Helmet Real-time Detection

作　　者：赵敏 杨国亮 王吉祥 龚志鹏 ZHAO Min;YANG Guoliang;WANG Jixiang;GONG Zhipeng(School of Electrical Engineering and Automation,Jiangxi University of Science and Technology,Ganzhou 341411,China)

机构地区：[1]江西理工大学电气工程与自动化学院,江西赣州341411

出　　处：《无线电工程》2023年第8期1741-1749,共9页Radio Engineering

基　　金：江西省教育厅科技计划项目(GJJ190450);江西省教育厅科技项目(GJJ180484)。

摘　　要：针对当前施工现场人工监管作业人员安全帽佩戴费时费力且实时性较差等问题,提出了一种改进YOLOv7-tiny的安全帽实时检测算法。引入EPSANet Block金字塔拆分注意力模块,捕捉细节信息,使模型更加聚焦训练安全帽相关目标特征。设计参数量更少的Tiny-BiFPN结构作为原模型特征融合模块中的特征金字塔结构,增强模型多尺度特征融合,改善网络对于安全帽检测的漏检率。采用更为先进的定位损失函数SIoU Loss计算损失,添加所需回归的向量角度,提高模型训练过程中预测框的收敛速度及效率。此外,创建了一个多元化环境下的安全帽数据集,在此数据集上实验表明,改进后检测算法对于原YOLOv7-tiny的mAP值提高了2.89%,检测速度提高了4.8帧/秒,实现了更加实时、准确的安全帽检测需求。To address the time-consuming,laborious and poor real-time wearing process of safety helmets for manual supervision workers at construction sites,an improved YOLOv7-tiny algorithm for safety helmet real-time detection is proposed.First,the EPSANet Block pyramid split attention module is introduced to capture the detailed information and make the model more focused on helmet-related target features training.Secondly,the Tiny-BiFPN structure with less design parameters is used as the feature pyramid structure in the feature fusion module of the original model,which enhances the multi-scale feature fusion of the model and decreases the missed detection rate of the network for helmet detection.Finally,the more advanced positioning loss function,SIoU Loss,is used to calculate the loss,and the vector angle of the required regression is added to improve the convergence speed and efficiency of the prediction box during model training.In addition,a safety helmet dataset in a diversified environment is created,on which the experiments show that the mAP value of the original YOLOv7-tiny is increased by 2.89%by the improved detection algorithm,and the detection speed is increased by 4.8 frame/s.A more real-time and accurate safety helmet detection is realized.

关 键 词：安全帽检测 YOLOv7-tiny EPSANet Block Tiny-BiFPN结构 SIoU Loss 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]