基于深度学习的目标分拣教学实验系统  

Target Sorting Teaching Experiment System Based on Deep Learning

作　　者：张银胜 单慧琳[1,2] 何秉坤 郭子民[1] ZHANG Yinsheng;SHAN Hinlin;HE Bingkun;GUO Zimin(College of Electronic and Information Engineering,Wuxi University,Wuxi 214105,China;College of Electronic and Information Engineering,Nanjing University of Information Science and Technology,Nanjing 210044,China)

机构地区：[1]无锡学院电子信息工程学院,江苏无锡214105 [2]南京信息工程大学电子与信息工程学院,江苏南京210044

出　　处：《软件导刊》2023年第11期85-90,共6页Software Guide

基　　金：国家自然科学基金项目(62071240);江苏省一流本科课程课题(2021YLKC005);无锡学院2021年教学改革研究课题(JGZD202109,JGYB202108)。

摘　　要：针对机器学习课程要求,结合电子信息类目标检测实际工程问题,提出一种基于深度学习的目标分拣教学实验系统。通过Jetson Nano识别、分类摄像头拍摄的目标,利用控制芯片对目标当前位置进行三维坐标转换,并控制机械臂进行抓取、分拣。同时,改进YOLOv3目标检测算法,利用MobileNetV2代替特征提取网络以减少特征尺度,将调整后的Focal loss损失函数替换YOLOv3的损失函数,并将锚点框聚类算法替换为K-Means++。实验表明,改进后算法的平均准确率提升23.4%,平均检测时间减少9.35 ms,检测精度和速度得到显著提升,可满足深度学习、机器视觉等先进技术的实践教学要求,有助于提升学生的创新与实践能力。Aiming at the requirements of machine learning courses and combining with practical engineering problems of electronic information object detection,a deep learning based object sorting teaching experimental system is proposed.Identify and classify the targets captured by the Jetson Nano camera,use a control chip to convert the current position of the targets into three-dimensional coordinates,and control the robotic arm to grasp and sort them.At the same time,YOLOv3 target detection algorithm is improved,MobileNetV2 is used to replace the feature extraction network to reduce the feature scale,the adjusted Focal loss loss function replaces YOLOv3 loss function,and the anchor box clustering algorithm is replaced by K-Means++.The experiment shows that the average accuracy of the improved algorithm is increased by 23.4%,the average detection time is reduced by 9.35 ms,and the detection accuracy and speed are significantly improved.It can meet the practical teaching requirements of advanced technologies such as deep learning and machine vision,and help to enhance students′innovation and practical abilities.

关 键 词：实验系统 深度学习 目标分拣 机械臂 Jetson Nano 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]