基于双阶段混合模型的密封电池高精度空间定位  

High-precision spatial positioning of sealed batteries based on a hybrid two-stage model

作　　者：葛亚明 葛旭 GE Yaming;GE Xu(Education Center of Experiments and Innovations,Harbin Institute of Technology(Shenzhen),Shenzhen 518055,China;School of Mechanical Engineering and Automation,Harbin Institute of Technology(Shenzhen),Shenzhen 518055,China)

机构地区：[1]哈尔滨工业大学(深圳)实验与创新实践教育中心,广东深圳518055 [2]哈尔滨工业大学(深圳)机电工程与自动化学院,广东深圳518055

出　　处：《兵器装备工程学报》2023年第11期293-298,307,共7页Journal of Ordnance Equipment Engineering

基　　金：广东省重点领域研发计划资助项目(2020B0909030001)。

摘　　要：为高效对密封电池进行高精度空间定位与轮廓焊接,提出一种基于双阶段混合模型的高精度密封电池轮廓定位方法,并设计相关实验。该方法将图像分割网络与传统图像处理方法相结合,构成双阶段混合模型,从而实现端到端的密封电池空间定位与机械臂焊接,并具有较高的定位精度。在自己制作的电池数据集上进行模型训练,并在六轴机械臂上进行焊接实验。结果表明:相比于UNet网络,本文中提出的CA-UNet具有更快的训练和推理速度,以及相近的分割精度,使得双阶段混合模型可以很好地实现密封电池轮廓定位。This paper proposes a high-precision sealed battery contour positioning method based on a two-stage hybrid model and designs related experiments in order to efficiently carry out high-precision spatial positioning and contour welding of sealed batteries.The method combines the image segmentation network with traditional image processing methods to form a two-stage hybrid model,allowing for end-to-end spatial positioning of the sealed battery and welding of the robotic arm with high positioning accuracy.This paper performs model training on the self-created battery data set and welding experiments on the six-axis robotic arm.The results demonstrate that:when compared to the UNet network,the CA-UNet proposed in this paper has faster training and inference speeds,as well as comparable segmentation accuracy,indicating that the two-stage hybrid model can accurately realize the contour location of sealed batteries.

关 键 词：深度学习 图像分割 高精度边缘定位 

分 类 号：TP29[自动化与计算机技术—检测技术与自动化装置]