基于YOLOv4模型的工件快速识别方法改进研究  

Research on Improvement of Rapid Workpiece Recognition Method Based on YOLOv4 Model

作　　者：左皓楠 胡桂川[1] 蒲小霞 侯文赛 邓春燕 ZUO Haonan;HU Guichuan;PU Xiaoxia;HOU Wensai;DENG Chunyan(School of Mechanical and Power Engineering,Chongqing University of Science and Technology,Chongqing 401331,China;School of Safety Engineering,Chongqing University of Science and Technology,Chongqing 401331,China)

机构地区：[1]重庆科技学院机械与动力工程学院,重庆401331 [2]重庆科技学院安全工程学院,重庆401331

出　　处：《重庆科技学院学报（自然科学版）》2024年第2期56-61,共6页Journal of Chongqing University of Science and Technology：Natural Sciences Edition

基　　金：重庆市科学技术局人工智能技术创新重大主题专项重点研发项目“基于全生命周期数据的超高水头冲击式发电机组智能预警与诊断系统”(CSTC2017RGZN-ZDYFX0026);技术创新与应用示范专项产业类重点研发项目“压气机整体叶盘高性能表面机器人磨削系统技术研发及应用示范”(CSTC2018JSZX-CYZDX0061);重庆科技学院研究生科技创新计划项目“智能制造中工件信息获取的智能算法研究”(YKJCX2120309);“基于协同深度学习的水电机组导轴承故障检测及性能劣化预测模型研究”(YKJCX2120713);“基于轻量化网络模型的水电机组转轮缺陷检测模型研究”(YKJCX2120727)。

摘　　要：为强化智能制造场景中机器人对工件的快速识别性能,从改进YOLOv4模型入手作了方法优化研究。首先,利用Ghost模块优化主干提取网络,以减少网络参数,提高网络检测速度;其次,引入DRConv卷积优化特征提取网络,以弥补对主干网络进行优化所造成的精度损失;最后,引入GAM注意力模块,以强化在光线不足条件下的适应性。通过对YOLOv4模型的改进,在保证较高识别精度和检测速度的同时,使模型规模得以简化,使工件快速识别网络趋于轻量化。To enhance the rapid recognition of workpieces by robots in intelligent manufacturing scenarios,a method optimization study was conducted by improving the YOLOv4 model.Firstly,the Ghost module was utilized to optimize the backbone feature extraction network,reducing network parameters and improving detection speed.Secondly,the DRConv convolution was introduced to optimize the feature extraction network,compensating for accuracy loss after optimizing the backbone network.Finally,the GAM attention module was introduced to enhance adaptability under low-light conditions.Through the improvement of the YOLOv4 model,the model scale is simplified and rapid recognition networks for workpieces tend to be lightweight while ensuring high recognition accuracy and detection speed.

关 键 词：智能制造 工件识别 YOLOv4模型 轻量化网络 GhostNet 

分 类 号：TH165[机械工程—机械制造及自动化]