基于深度学习的纱管识别方法研究  

Research on Yarn Bobbin Detection Method Based on Deep Learning

作　　者：王青[1] 吕绪山 党帅 姜越夫 梁高翔 赵恬恬 薛博文 WANG Qing;LYU Xushan;DANG Shuai;JIANG Yuefu;LIANG Gaoxiang;ZHAO Tiantian;XUE Bowen(College of Mechanical and Electrical Engineering,Xi’an Polytechnic University,Xi’an 710048,China)

机构地区：[1]西安工程大学机电工程学院,陕西西安710048

出　　处：《机械与电子》2023年第12期20-26,共7页Machinery & Electronics

基　　金：陕西省重点研发计划项目(2022GY 307)。

摘　　要：为提高自动络筒机的工作效率,视觉检测准确率的提升尤为重要,在YOLOv5的基础上提出了一种改进的纱管识别方法。将网络原有的SiLU激活函数替换为表现力更好的Mish激活函数。将CIoU定位损失函数替换为考虑了真实框和预测框之间方向匹配性的SIoU损失函数,使网络更快速地收敛。将原有的C3模块替换为嵌入了CA注意力机制的CCA模块,使网络在提取特征上具有更好的表现力。制作纱管数据集,并对数据集进行数据增强使模型具有更好的鲁棒性和泛化能力。通过试验得出,所提的改进YOLOv5网络在识别准确率上达到了97.30%,召回率达到了98.17%,mAP_0.5达到了98.58%,改进后的网络相较于原网络,在识别性能上有显著提升。To improve the efficiency of automatic winding machines,it is particularly important to improve the accuracy of visual inspection.This paper proposes an improved yarn bobbin detection method based on YOLOv5.The original SiLU activation function of the network is replaced by the Mish activation function with better expressive power.The CIoU localization loss function is replaced with the SIoU loss function,which takes into account the directional matching between the true and predicted Bounding Box,allows the network to converge more quickly.The original C3 module is replaced with a CCA module embedded with a CA attention mechanism,which gives the network better performance in extracting features.Besides,the yarn bobbin dataset is produced and data augmentation is applied for better robustness and generalization of the model.Through the experiment,it was concluded that the improved YOLOv5 network proposed in this paper achieved 97.30%in recognition precision;98.17%in recall;and 98.58%in mAP_0.5.The improved network has a significant improvement in recognition performance compared with the original network.

关 键 词：纱管识别 注意力机制 深度学习 YOLOv5 

分 类 号：TP317.4[自动化与计算机技术—计算机软件与理论]