基于改进AlexNet与CUDA的大豆快速三分类方法  

Soybean Rapid Tri-classification Method Based on Improved AlexNet and CUDA

作　　者：林伟 仲伟波[1] 袁毓 齐国庆 李浩东 LIN Wei;ZHONG Weibo;YUAN Yu;QI Guoqing;LI Haodong(School of Electronic Information,Jiangsu University of Science and Technology,Zhenjiang 212100)

机构地区：[1]江苏科技大学电子信息学院,镇江212100

出　　处：《计算机与数字工程》2023年第12期2997-3003,共7页Computer & Digital Engineering

摘　　要：为了能够精确、快速实现大豆籽粒分类,提出了基于改进AlexNet与CUDA的大豆籽粒快速三分类方法。以大豆籽粒多分类为目标,构建大豆籽粒图像库;根据快速分类的任务要求,对传统AlexNet模型进行改进并采用统一计算设备架构C++实现;以NVIDIA Jetson TX2为核心构建大豆籽粒快速分类系统。训练集及验证集分类准确率分别可达98%和94%;对于在线采集的大豆图像进行分类测试准确率约为93%,且一粒大豆籽粒分类耗时约6ms,能够满足快速分类的应用需求。In order to realize soybean seed classification accurately and quickly,a fast tri-classification soybean seed classifi-cation method based on improved AlexNet and CUDA is proposed.Aiming at multi-classification of soybean seeds,a soybean seed image library is constructed.According to the task requirements of fast classification,AlexNet is improved and implemented by CU-DA C++.A rapid soybean seed classification system is constructed based on NVIDIA Jetson TX2.The classification accuracy of training set and validation set can reach 98%and 94%respectively.The classification accuracy of soybean images collected online is about 93%,and it takes about 6ms to classify a soybean seed,which can meet the application requirements of rapid classifica-tion.

关 键 词：图像处理 机器视觉 大豆籽粒分类 AlexNet改进模型 统一计算设备架构 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]