基于卷积神经网络的花朵品种的识别  

Identification of flower varieties based on convolution neural network

作　　者：杨静亚 李景霞 王振宇 程海[1] YANG Jing-Ya;LI Jing-Xia;WANG Zhen-Yu;CHENG Hai(School of Electronic Engineering,Heilongjiang University,Harbin 150080,China)

机构地区：[1]黑龙江大学电子工程学院

出　　处：《黑龙江大学工程学报》2019年第4期90-96,共7页Journal of Engineering of Heilongjiang University

基　　金：国家自然科学基金资助项目(61471158,61571181)

摘　　要：基于BP算法的卷积神经网络应用于图像识别领域,它有自动学习特征,比传统的图像识别方法的准确率更高。介绍了基于卷积神经网络的花朵品种的识别,构建CNN神经网络模型,运用BP算法优化参数,激活函数采用稀疏性较好的Relu调整输出,在牛津大学102种花卉的数据集的基础上,增加了5种,准确率为83.01%,测试随机采取5种花卉进行识别分类,准确率最高为85%。Convolution neural network based on BP algorithm has been used in image recognition field.It can learn features automatically by itself and has higher accuracy than traditional image recognition methods.This paper mainly introduces the flower variety recognition based on convolution neural network,constructs the CNN neural network model,optimizes the parameters by using BP algorithm,and adjusts the output with the sparsity Relu for activation function.On the basis of the data set of 102 species of flowers at Oxford University,5 species have been added,with an accuracy rate of 83.01%,the test randomly used 5 kinds of flowers to identify and classify,the highest accuracy of 85%.

关 键 词：图像识别 卷积神经网络 花朵识别 深度学习 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]