首页|基于YOLOv4模型的工件快速识别方法改进研究

基于YOLOv4模型的工件快速识别方法改进研究

扫码查看
原文链接
  • 万方数据
  • 维普
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.
Research on Improvement of Rapid Workpiece Recognition Method Based on YOLOv4 Model
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 opti-mize the backbone feature extraction network, reducing network parameters and improving detection speed. Second-ly, 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 adapta-bility 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.

intelligent manufacturingworkpiece recognitionYOLOv4 modellightweight networkGhostNet

左皓楠、胡桂川、蒲小霞、侯文赛、邓春燕

展开 >

重庆科技学院 机械与动力工程学院,重庆 401331

重庆科技学院 安全工程学院,重庆 401331

智能制造 工件识别 YOLOv4模型 轻量化网络 GhostNet

重庆市科技局人工智能技术创新重大主题专项重点研发项目技术创新与应用示范专项产业类重点研发项目重庆科技学院研究生科技创新计划&&&&

CSTC2017RGZN-ZDYFX0026CSTC2018JSZX-CYZDX0061YKJCX2120309YKJCX2120713YKJCX2120727

2024

10.19406/j.issn.1673-1980.2024.02.009

重庆科技学院学报(自然科学版)
重庆科技学院

重庆科技学院学报(自然科学版)

影响因子：0.329
ISSN：1673-1980
年,卷(期)：2024.26(2)
  • 7