基于深度学习的多视角螺钉缺失智能检测算法  

Intelligent screw absence detection algorithm based on deep learning with multiple perspectives

作　　者：于畅 伍星[1] 邓秋菊 YU Chang;WU Xing;DENG Qiuju(College of Computer Science,Chongqing University,Chongqing 400044,China;School of Big Dada and Computer Science Engineering,Chongqing College of Mobile Communication,Chongqing 401520,China)

机构地区：[1]重庆大学计算机学院,重庆400044 [2]重庆移通学院大数据与计算机科学学院,重庆401520

出　　处：《山东大学学报（工学版）》2023年第4期104-112,共9页Journal of Shandong University（Engineering Science）

基　　金：国家自然科学基金资助项目(6167211);重庆市教育委员会科学技术研究资助项目(KJQN202102401)。

摘　　要：为有效实现工业生产线螺钉缺失问题的智能检测,利用深度学习技术,提出并设计一种螺钉检测算法。该算法包括3个部分:基于目标检测算法实现螺钉自动检测;基于关键点检测的螺钉匹配算法消除零件位置变化影响;构建多视角检测结果融合算法降低零件相互遮挡影响。该算法已应用于多种型号的洗衣机内桶螺钉检测中,试验结果表明其正确率高达99.7%以上。与传统的人工检测方式相比,该算法具有更高的准确率和自动化程度,可以有效减少漏检和误检问题,为工业生产提供新的解决方案。To effectively achieve intelligent detection of missing screws in industrial production lines,a screw detection algorithm was proposed and designed using deep learning technology.The algorithm consisted of three parts:automatic screw detection based on an object detection algorithm,a screw matching algorithm based on key point detection to eliminate the influence of part position changes,and a multi-view detection result fusion algorithm to reduce the impact of part occlusion.This algorithm was applied to the detection of screws in various models of washing machine inner barrels,and experimental results showed that its accuracy rate was over 99.7%.Compared with traditional manual detection methods,this algorithm had higher accuracy and automation,and could effectively reduce missed and false detections,providing a new solution for industrial production.

关 键 词：深度学习 机器视觉 目标检测 关键点检测 多视角融合 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]