基于RegNet网络的马铃薯病害识别研究
Research on potato disease identification based on RegNet network
石放 1王莹 1王新法 1马玉琨1
作者信息
摘要
为克服传统网络模型结构固化、对马铃薯病害识别率低的问题,以PlantVillage数据集中的五类马铃薯病害为研究对象,对图像进行随机放大缩小、水平翻转、垂直翻转等操作进行数据增强.使用一种基于网络设计空间思想设计出具有高度灵活性的RegNet网络模型,利用PoLy损失函数对RegNet进行改进,并加入注意力机制,对数据增强后的马铃薯病害图片进行预测,再与传统网络模型AlexNet和GoogLeNet进行对比.试验结果表明:改进后的RegNetX在马铃薯识别方面具有良好的性能,最高准确率可达99.8％,模型准确率超过AlexNet与GoogLeNet,可为马铃薯病害识别作参考.
Abstract
In order to overcome the problems of solidified structure of traditional network model and low recognition rate of potato diseases,five types of potato diseases in PlantVillage dataset are taken as the research object,and the images are randomly zoomed in and out,horizontally flipped,vertically flipped and so on for data enhancement.Then a RegNet network model with a high degree of flexibility is designed using a network-based design space idea,and the PoLy loss function is used to improve RegNet and the attention mechanism is added to predict the potato disease images after data enhancement,and the traditional network models are compared with AlexNet and GoogLeNet.The experimental results show that the improved RegNetX has good performance in potato recognition,the highest accuracy can reach 99.8％,and the model accuracy is higher than AlexNet and GoogLeNet,which can be used as a reference for potato disease recognition.
关键词
马铃薯/农作物病害/RegNet网络/图像识别/网络设计空间Key words
potato/crop diseases/RegNet network/image recognition/network design space引用本文复制引用
基金项目
河南省科技攻关计划(212102110234)
河南省科技攻关计划(222102320080)
河南省高等学校重点科研项目(22A210013)
河南省博士后科研启动基金(202102090)
河南省新乡市科技重大专项(21ZD003)
出版年
2024
国家科技期刊平台
NSTL
万方数据