基于改进YOLOv5的工业棒料识别研究  

Research on industrial bar material recognition based on improved YOLOv5

作　　者：胡宏升 叶树林 张仁冬 HU Hongsheng;YE Shulin;ZHANG Rendong(School of Mechatronic Engineering and Automation,Foshan University,Foshan 528225,China)

机构地区：[1]佛山大学机电工程与自动化学院,广东佛山528225

出　　处：《佛山科学技术学院学报(自然科学版)》2025年第2期43-48,共6页Journal of Foshan University(Natural Science Edition)

摘　　要：为提高制造业中棒料自动上料的效率,提出了一种改进的Gs-YOLOv5s模型。将YOLOv5s的骨干网络替换为MobileNetV3的block模块,减少了模型参数并提高了检测速度。在颈部网络中嵌入GSConv模块,丰富输出特征图并增强模型对棒料特征的捕捉能力,同时减少了参数量。采用MPDIoU损失函数,通过引入距离惩罚和面积惩罚,使得模型在复杂背景下的棒料检测更加准确。实验结果表明,Gs-YOLOv5s模型的精度仅下降0.2%,但检测速度达到81帧/s,提高了45.3%,参数量减少了57.6%。该改进模型在保持精度的同时,实现了显著的轻量化和检测速度提升,适合在实际生产中部署使用。This paper presents an improved Gs-YOLOv5s model to improve the efficiency of automatic bar feeding in manufacturing industry.By replacing the backbone network of YOLOv5s with the block module of MobileNetV3,the model parameters are reduced and the detection speed is increased.The GSConv module is embedded in the neck network to enrich the output feature map and enhance the model’s ability to capture the bar feature,while reducing the number of parameters.Finally,the MPDIoU loss function is used to make the model more accurate in the complex background by introducing distance penalty and area penalty.The experimental results show that the accuracy of Gs-YOLOv5s model is only reduced by 0.2%,but the detection speed reaches 81 frames/s,which is increased by 45.3%,and the number of parameters is reduced by 57.6%.The improved model achieves significant lightweight and detection speed improvement while maintaining accuracy and is suitable for deployment in actual production.

关 键 词：棒料识别 目标检测 YOLOv5s MobileNetV3 GSConv MPDIoU 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TH16[自动化与计算机技术—计算机科学与技术]