摘要
针对苹果叶面病害识别问题,文章在对现有的VGG网络模型研究的基础上,提出了一种在VGG-16网络模型基础上进行改进的方法,即VGG-16-BN模型.首先对西北农林科技大学制作的数据集进行处理,然后将max pooling层替换为sort-pool2d池化层,使模型更快地收敛,最后在每一卷积层后增设BN层,加快训练速度,防止过拟合.实验结果表明,新的网络模型在数据集中的准确率相较于未增设BN层有了一定提高,其准确率达到96.67%.
In response to the problem of apple leaf disease identification,one improved method based on the VGG-16 network model,namely the VGG-16-BN model,is proposed based on the existing VGG network model research.Firstly,the paper processes the dataset produced by Northwest A&F University,and then re⁃places the max pooling layer with the sort pool2d pooling layer to make the model converge faster.Finally,BN layer is added after each roll of accumulation layer to speed up training and prevent overfitting.The experimen⁃tal results show that the accuracy of the new network model in the dataset has been improved to some extent compared to the absence of a BN layer,with an accuracy of 96.67%.
作者
赵慧勐
刘向举
金彬峰
ZHAO Hui-meng;LIU Xiang-ju;JIN Bin-feng(College of Computer Science and Engineering,Anhui University of Science and Technology,Huainan 232001,China)
出处
《白城师范学院学报》
2023年第5期59-64,共6页
Journal of Baicheng Normal University
基金
国家自然科学基金项目(61572034)
安徽省科技重大专项(18030901025).
关键词
病害识别
苹果
叶片
VGG-16
图像分类
最大池化层
批标准化
disease identification
apple
blade
VGG-16
image classification
maximum pooling layer
batch standardization
