基于ResNeXt和迁移学习的花卉种类识别研究  

Flower Recognition Based on ResNeXt and Transfer Learning

作　　者：吴媛媛 梁礼明 彭仁杰 尹江 WU Yuanyuan;LIANG Liming;PENG Renjie;YIN Jiang(School of Electrical Engineering and Automation,Jiangxi University of Science and Technology,Ganzhou 341000,China)

机构地区：[1]江西理工大学电气工程与自动化学院,江西赣州341000

出　　处：《种业导刊》2022年第1期42-48,共7页Journal of Seed Industry Guide

基　　金：国家自然科学基金项目(51365017,61463018);江西省自然科学基金面上项目(20192BAB205084);江西省教育厅科学技术研究重点项目(GJJ170491)。

摘　　要：针对传统花卉识别方法分类不准确和泛化能力低等问题,提出一种基于Res Ne Xt50网络和迁移学习的花卉种类识别模型DB-Res Ne Xt50。首先,使用Mixup数据增强方法进行数据扩充;其次,利用预训练Res Ne Xt50模型迁移学习,省去网络训练基础特征步骤;最后,在主干网络全连接层前加入含有15层Bottleneck的稠密块,提高特征整合能力。结果表明,DB-Res Ne Xt50网络模型对花卉数据集识别准确率为97.99%。与传统深度学习模型相比,该模型在识别率上有很大提升,总体性能优于现有算法性能。In view of the problems of inaccurate classification and low generalization ability of traditional flower recognition model DB-ResNeXt50,a flower species recognition algorithm based on ResNeXt50 network and transfer learning was proposed.Firstly,the Mixup data enhancement method was used for data expansion;Secondly,by using the pre training ResNeXt50 model for transfer learning,the basic feature steps of network training were omitted;Finally,a dense block with 15 layers of Bottleneck was added in front of the full connection layer of the backbone network to improve the ability of feature integration.The results showed that the classification accuracy of the DB-ResNeXt50 network on the flower data set was 97.99%.Compared with the traditional deep learning model,the recognition rate of this model was greatly improved,and the overall performance was better than that of the existing algorithms.

关 键 词：ResNeXt50 迁移学习 花卉识别 稠密块 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]