基于改进YOLOv5s的仓储货物检测算法研究  

Research on an Improved YOLOv5s-based Algorithm for Warehouse Goods Detection

作　　者：王影[1] 王晨[1] 贾永涛 刘麒[1] WANG Ying;WANG Chen;JIA Yongtao;LIU Qi(School of Information and Control Engineering,Jilin Institute of Chemical Technology,Jilin City 132022,China;Fertilizer Plant,China Petroleum Jilin Petrochemical Company,Jilin City 132000,China)

机构地区：[1]吉林化工学院信息与控制工程学院,吉林吉林132022 [2]中国石油吉林石化公司化肥厂,吉林吉林132000

出　　处：《吉林化工学院学报》2024年第1期51-58,共8页Journal of Jilin Institute of Chemical Technology

基　　金：吉林市科技成果项目(201750244);吉林市科技成果项目(20190502118);吉林化工学院科研项目(2018064);吉林化工学院重大科技项目(2016033);吉林化工学院重大科技项目(2018017)。

摘　　要：针对目前仓储货物分类速度慢、易出错、灵活性差等问题,提出了一种改进YOLOv5s的货物检测算法,对仓储货物进行预分类。首先,根据仓储货物的外形特征,将其分为包装箱与包装袋两大类,形成训练数据集;其次,将骨干网络更换为具有更小模型尺寸的MobileNetV3,加快推理;再次,添加SE注意力机制模块,旨在提高模型的检测精度;最后,结合α_CIoU损失函数,增强模型的灵活度。通过实验验证,改进后的算法相较于原始算法在精确率(Precision,P)、平均类别精度(mean Average precision,mAP)和帧率(Frames per second,FPS)三方面分别提升2.1%、0.5%和10.6%,能够高效地完成对仓储货物的预分类工作。A modified YOLOv5s detection algorithm had been proposed to address the issues of slow classification speed,error-proneness,and low flexibility in warehouse goods categorization.The algorithm aims to pre-classify warehouse goods.Firstly,based on the external characteristics of warehouse goods,they were divided into two main categories:packaging boxes and packaging bags,forming a training dataset.Secondly,the backbone network was replaced with MobileNetV3,a smaller-sized model,to accelerate inference.Additionally,an SE attention mechanism module was added to enhance the detection accuracy of the model.Finally,theα_CIoU loss function was incorporated to improve the flexibility of the model.Experimental results demonstrated that the improved algorithm achieves a 2.1%increase in precision(P),a 0.5%increase in mean Average Precision(mAP),and a 10.6%increase in Frames Per Second(FPS)compared to the original algorithm.It enables efficient pre-classification of warehouse goods.

关 键 词：YOLOv5s 仓储货物 检测算法 预分类 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] F252[自动化与计算机技术—计算机科学与技术]