基于卷积神经网络的脐橙品质分类研究  

Research on Quality Classification of Navel Orange Based on Convolutional Neural Network

作　　者：朱冬梅 王敏[1,2] ZHU Dongmei;WANG Min(School of Mathematics and Computer Science,Gannan Normal University,Ganzhou 341000,China;Key Laboratory of Jiangxi Province for Numerical Simulation and Emulation Technique,Ganzhou 341000,China)

机构地区：[1]赣南师范大学数学与计算机科学学院,江西赣州341000 [2]赣南师范大学江西省数值模拟与仿真技术重点实验室,江西赣州341000

出　　处：《赣南师范大学学报》2018年第6期25-28,共4页Journal of Gannan Normal University

基　　金：江西省科技支撑计划项目(20151BBE50076);江西省教育厅科技项目(GJJ151001,GJJ150984);江西省数值仿真与模拟技术重点实验室开放课题;赣南师范大学研究生创新基金项目.

摘　　要：为了解决脐橙样品差异大、分类难问题,文章利用卷积神经网络探讨脐橙品质分类问题.主要包含两方面的工作:首先构建脐橙品质分类数据集.其次,根据脐橙品质分类的高准确率和实时性需求,设计了包含2个卷积层、2个下采样层和1个全连接层的卷积神经网络模型.将脐橙图片批量归一化,以ReLU为激活函数,Maxpooling为下采样方法,并采用Softmax回归分类器训练并优化卷积神经网络.实验结果表明本文提出的基于卷积神经网络的脐橙分类准确率达到94.8%,与现有最好的传统分类方法相比,准确率高出5.21%.In order to solve the problem of large difference and difficulty in classification of navel orange samples,this paper employs convolutional neural network to explore the classification of navel orange quality.It mainly includes two aspects:firstly,to build the data set of navel orange quality classification;secondly,according to the high accuracy and real-time demand of the quality classification of navel orange,to design the convolution neural network model including two convolution layers,two lower sampling layers and one full connection layer.The images of navel oranges were normalized in batches,Re LU was taken as the activation function,Maxpooling was taken as the lower sampling method,and Softmax regression classifier was used to train and optimize the convolutional neural network.Experimental results show that the accuracy rate of navel orange classification proposed in this paper is 94.8%,which is 5.21%higher than the existing best traditional classification method.

关 键 词：脐橙品质分类 卷积神经网络 ReLU激活函数 Max-pooling下采样 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TP183[自动化与计算机技术—计算机科学与技术]