基于深度学习的大米加工新鲜度分类方法  

A deep-learning-based method for classifying rice processing freshness

作　　者：訾薇宇 舒忠平[1] ZI Wei-yu;SHU Zhong-ping(Shangluo Vocational&Technical College,Shangluo 726000,China)

机构地区：[1]商洛职业技术学院,陕西商洛726000

出　　处：《粮食与饲料工业》2024年第5期71-75,共5页Cereal & Feed Industry

基　　金：商洛职业技术学院重大科研项目“绿色经济助推商洛康养之都建设的路径探究”(JYKT202401)。

摘　　要：为提高大米加工新鲜度分类的精度和速度,提出一种基于深度学习的分类方法。方法以VGG19网络为基础分类网络,通过在该网络基础上引入SE注意力机制加强对重要通道特征的关注,并采用PReLU函数替换ReLU函数作激活函数,同时将网络的最后一层池化层替换为全局混合池化,并删除前两层全连接层,对VGG19网络进行了改进。最后,以大米新鲜度为研究对象,采用改进VGG19网络进行新鲜度分类,实现了大米新鲜度分类。仿真结果表明,改进VGG19网络实现了精确、快速地大米新鲜度分类,平均准确率、精确率、召回率和F 1值分别达到97.81%、97.63%、97.89%、97.56%,且具有较快的检测速度,测试时间为275 s,提高了大米加工新鲜度分类的精度和速度。To improve the accuracy and speed of rice processing freshness classification,a classification method based on deep learning was proposed in this paper.Based on VGG-19 architecture,the method introduced SE(squeeze-and-excitation)attention mechanism to follow more closely the features of critical channels and substituted ReLU function with PReLU function for activation purpose.Meanwhile,VGG-19 network was materially modified by replacing its bottom pooling layer with global mixed pooling and deleting the first two fully connected layers.Then,with rice freshness as the research object,the modified VGG-19 network was implemented to classify the rice by its freshness and was proven effective.Simulation results indicate the modified VGG-19 could accurately and quickly classify the rice by freshness.Its average accuracy,precision,recall ratio,and F 1 value were 97.81%,97.63%,97.89%,and 97.56%,respectively.It was testified as fast in rice detection,as the test took only 275 s.The method proposed hereby did improve both the accuracy and speed of freshness-based rice processing classification.

关 键 词：大米加工 新鲜度分类 深度学习 VGG19网络 大米新鲜度 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] TS212[自动化与计算机技术—计算机科学与技术]