吉林农业大学学报2023，Vol.45Issue(6)：791-802,12.DOI:10.13327/j.jjlau.2023.20259
基于改进ConvNeXt网络的人参分级模型
Ginseng Grading Model Based on Improved ConvNeXt Network
摘要
Abstract
In order to solve the problem of small feature variability among ginseng grading classes and heavy reliance on professionals,a ginseng data set containing 5 116 images with three classes in different contexts was established,and a ginseng grading model based on an improved ConvNeXt framework was proposed.Firstly,the channel shuffle module was embedded in the backbone network after down-sampling to fully fuse the channel features to improve the grading accuracy;secondly,the ConvNeXt Block module was optimized and improved,and the Re-parameterization module was fused with it to improve the characterization ability of the model,enrich the feature space of the con-volutional blocks,and further improve the model's accuracy;finally,the original activation function GELU was replaced by the PReLU activation function to increase the nonlinear variability of the neu-ral network model,which improved the model's accuracy and efficiency.The experimental results showed that the accuracy,precision,recall and specificity of ginseng classification reach 94.44％,91.58％,91.04％,and 95.82％,and the loss rate has been reduced to 0.24％,which proves the valid-ity of this study and can provide reference for ginseng quality classification.关键词
ConvNeXt/人参分级/通道混洗/结构重参数化/激活函数/深度学习Key words
ConvNeXt/ginseng grading/channel shuffle/structural reparameterization/activa-tion function/deep learning分类
农业科技引用本文复制引用
翟梦婷,张丽娟,朴欣茹,李伟,李东明..基于改进ConvNeXt网络的人参分级模型[J].吉林农业大学学报,2023,45(6):791-802,12.基金项目
国家自然科学基金青年科学基金项目(61801439),吉林省重点研发项目(20210204050YY),吉林省教育厅项目(JJKH20210747KJ),吉林省环保厅项目(202107),吉林省中青年领军团队及创新人才支持计划项目(20200301037RQ) （61801439）
