| 基于视觉引导的工业棒材上料系统研究 |
| |
| 引用本文： | 王西志,管声启,张理博,刘 通,郝振虎.基于视觉引导的工业棒材上料系统研究[J].机械与电子,2023,41(5):19-23. |
| |
| 作者姓名： | 王西志 管声启 张理博 刘 通 郝振虎 |
| |
| 作者单位： | 西安工程大学机电工程学院,陕西 西安 710048 |
| |
| 摘 要： | 为了提高生产效率，设计一种基于视觉引导的工业棒材上料系统。首先，为了实现视觉引导进行工业棒材上料，设计了工业棒材上料总体方案，并对上料机械结构模型进行选型设计。然后，为了实现棒材的自动识别和位姿检测，提出了一种基于改进YOLOv5的旋转目标识别与定位算法。该方法在YOLOv5主干特征网络上，添加高效ECA通道注意力机制模块，利用其避免降维，并通过适当跨通道交互策略提高特征提取能力；为了增强不同尺度的特征信息融合，将原特征增强网络替换成BiFPN加权双向特征金字塔网络，进行自上而下和自下而上的多尺度特征融合，提高棒材识别准确率并获取平面位置信息；在此基础上，采用双目视觉进行立体匹配获取棒材的深度位置信息，最终实现棒材立体位姿检测。对所提上料系统进行实验验证，棒材识别的平均精度为99.4%,抓取棒材成功率达到90%及以上。
|
| 关 键 词： | 上料系统 深度学习 位姿检测 两指机械手 |
Research on Industrial Bar Feeding System Based on Visual Guidance |
| |
| WANG Xizhi,GUAN Shengqi,ZHANG Libo,LIU Tong,HAO Zhenhu.Research on Industrial Bar Feeding System Based on Visual Guidance[J].Machinery & Electronics,2023,41(5):19-23. |
| |
| Authors: | WANG Xizhi GUAN Shengqi ZHANG Libo LIU Tong HAO Zhenhu |
| |
| Affiliation: | ( School of Mechanical and Electronic Engineering , Xi ’an Polytechnic University , Xi ’an 710048 , China ) |
| |
| Abstract: | In order to improve the production efficiency , this paper designs an industrial bar feeding system based on visual guidance.First of all , in order to realize the visual guidance of industrial bar feeding , the overall scheme of industrial bar feeding is designed , and the selection and design of the feeding mechanical structure model are designed.Then , in order to realize automatic bar recognition and pose detection , a rotating target recognition and location algorithm based on improved YOLOv5 is proposed.In this method , an efficient ECA channel attention mechanism module is added to the YOLOv5 backbone feature network , and the feature extraction ability is improved by using its avoidance reduction and appropriate cross-channel interaction strategy.In order to enhance the feature information fusion of different scales , the original feature enhancement network was replaced with BiFPN weighted bidirectional feature pyramid network , and the top-down and bottom-up multi-scale feature fusion was carried out to improve the bar recognition accuracy and obtain the plane position information.On this basis , the depth and position information of the bar is obtained by stereo matching with binocular vision , and finally the stereo pose detection of the bar is realized.Through the experimental verification of the feeding system in this paper , the average accuracy of bar recognition is 99.4% , and the success rate of grasping bar is 90% or above. |
| |
| Keywords: | feeding system deep learning pose detection two finger manipulat |
|
| 点击此处可从《机械与电子》浏览原始摘要信息 |
|
点击此处可从《机械与电子》下载全文 |
|
