基于改进YOLOv5的管纱与纱锭智能识别方法  

Intelligent recognition method of yarns and spindles based on improved YOLOv5

作　　者：冯灵 张玉宁 程静 王鹏 杨雅宁 FENG Ling;ZHANG Yuning;CHENG Jing;WANG Peng;YANG Yaning(College of Physics and Electronic Information Engineering,Ningxia Normal University,Guyuan,Ningxia 756000,China)

机构地区：[1]宁夏师范大学物理与电子信息工程学院,宁夏固原756000

出　　处：《毛纺科技》2025年第4期101-110,共10页Wool Textile Journal

基　　金：宁夏自然科学基金项目(2022AAC03318)。

摘　　要：针对基于机器视觉的智能落纱机落纱时存在纱管与纱锭识别准确率低、识别速度慢、动态适应性不足的问题,提出一种改进YOLOv5的管纱与纱锭识别方法。该方法首先在颈部网络(Neck)部分引入轻量化GSConv卷积,以更少的参数量生成更多的特征图,在降低算法计算量的同时提高识别准确率;其次,采用EIoU作为定位损失函数,使模型专注于高质量锚框的回归,提高网络的收敛速度;最后采用自适应激活函数MetaAconC替代YoLov5中的SiLU激活函数,通过引入额外的参数提高模型的泛化性。实验结果表明:与标准的YOLOv5相比方法,采用改进YOLOv5方法识别管纱和纱锭时,其精确度、召回率、平均精度均值分别提升了0.8%、0.9%和0.1%,前传耗时降低了1.3 ms,可以有效提升智能落纱机的性能。To address the issues of low accuracy,slow speed,and inadequate dynamic adaptability in yarns and spindles recognition during doffing in machine vision-based doffing machines,an improved YOLOv5-based recognition method was proposed.Firstly,the lightweight GSConv convolution in the Neck part of the network was introduced,which generated more feature maps with fewer parameters,thus improving recognition accuracy while reducing the computational load.Secondly,the EIoU was used as the localization loss function,which enabled the model to focus on the regression of high-quality anchor boxes,thereby improving the network′s convergence speed.Finally,the MetaAconC adaptive activation function was used to replace the SiLU activation function in YOLOv5,enhancing the model′s generalization ability by introducing additional parameters.Experimental results show that,compared to the standard YOLOv5,the improved YOLOv5 method increases the precision,recall,and mean average precision of yarns and spindles recognition by 0.8%,0.9%and 0.1%,respectively,and reduces the forward propagation time by 1.3 ms,effectively enhancing the performance of intelligent doffing machines.

关 键 词：机器视觉 深度学习 管纱识别 纱锭识别 YOLOv5 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]