摘要
通过深度学习网络模型来实现玉米叶片病害识别分类已成为主流,但深度学习的模型需要拥有较多的数据集,然而实际情况是人工获得的样本种类和数据是有限的,且小样本数据集下的模型容易出现过度拟合从而丧失泛化能力。基于此背景,本文提出了一种迁移学习下的小样本玉米叶片病害识别方法,一方面从迁移学习角度出发,解决了因少样本而导致模型泛化差的问题;另一方面从深度学习的方向出发,采用并训练AlexNet、ResNet50和MobileNetV2模型,并对比三种模型在基于迁移学习下的病害识别准确率。研究结果表明,迁移学习有助于提高小样本泛化能力,MobileNetV2模型更适合小样本玉米叶片病害的识别。
It is the main method to recognize maize leaf disease by deep learning network model,but the deep learning model needs more data sets,but in fact,the sample types and data obtained artificially are limited,moreover,the model with small sample data set is prone to over-fitting and loss of generalization ability.Based on this background,this paper proposes a method of identifying maize leaf disease with small samples by migration learning.On the one hand,it solves the problem of poor generalization caused by small samples from the perspective of migration learning,On the other hand,Alexnet,resnet 50 and mobilenet V2 models were adopted and trained from the direction of deep learning to compare the accuracy of disease recognition based on the three models.The results showed that the mobility learning could improve the generalization ability of small samples,and the mobilenet V2 model was more suitable for the identification of leaf diseases in small samples.
作者
李婷婷
王晴晴
唐琦
张浩
惠向晖
LI Ting-ting;WANG Qing-qing;TANG Qi;ZHANG Hao;HUI Xiang-hui(Henan Agricultural University,College of Information and Mangement Science,Zhengzhou,Henan 450046,China)
出处
《新一代信息技术》
2023年第24期1-5,共5页
New Generation of Information Technology
基金
河南省教育科学规划2023年度一般课题(No.2023YB0038)
河南农业大学本科教育教学改革研究与实践项目(No.2023XJGLX045)
河南省教育厅-本科高校2023年度产教融合研究项目(教办高[2024]13号-重点项目8)
关键词
小样本
病害识别
迁移学习
深度学习
small sample size
disease identification
transfer learning
deep learning
