为提高汽车工艺涂胶质量及机械臂作业效率,针对基于深度学习的双目视觉车顶焊缝涂胶机械臂系统,提出了一种SEmYOLOv5 算法,在主干网络上增加SE(squeeze and excitation)注意力机制,同时在颈部网络上增加一组采样模块,提高焊缝的识别能力.对提取到的图像进行图像处理,使得更好的提取车顶焊缝的特征信息从而得到特征点坐标,采用B样条曲线法对机械臂进行轨迹规划.改进后的算法相较原YOLOv5 算法的mAP值提升了6.76%,针对该系统进行实验并验证了提出的基于深度学习的双目视觉车顶焊缝涂胶机械臂系统的有效性.
Research on Glue Manipulator for Roof Weld Based on Deep Learn
In order to improve the coating quality of automobile technology and the working efficiency of the robot arm,a SEmYOLOv5 algorithm is proposed for a binocular vision robot system for gluing roof welds based on deep learning,in which SE(squeeze and excitation)attention mechanism is added to the backbone network,and a group of sampling modules are added to the neck network to improve the weld recognition ability.Image processing is carried out on the extracted image,so that the feature information of the roof weld can be better extracted and the coordinates of the feature points can be obtained;Finally,the trajectory planning of the manipulator is carried out by using B-spline curve method.Compared with the o-riginal YOLOv5 algorithm,the mAP value of the improved algorithm is increased by 6.76%.Experiments are carried out on this system and the effectiveness of the proposed system is verified.
roof weldSEmYOLOv5hand-eye calibrationmechanical arm gluingtrajectory planning
NETL
NSTL
万方数据
