基于卷积神经网络的水果图像分类识别研究

Classification and Recognition of Common Fruit Images Based on Convolutional Neural Network

为了解决传统水果图像分类识别算法人工提取特征的缺陷,将卷积神经网络应用到水果图像识别上,基所创建的数据集,参照经典的卷积神经网络模型Le Net-5结构,提出更适合本数据集的卷积神经网络结构,首先对水果数据集进行分类标签,将苹果、梨、橙子、橘子、桃子分别标记为0、1、2、3、4,然后将图片分批次投入模型训练,该模型构建了一个输入层、两个卷积层、两个池化层、两个全连接层和一个输出层。卷积神经网络通过底层提取特征,再进一步更深层次提取特征,最后得到目标的分类。实验结果表明,所提出的卷积神经网络结构不仅在数据集上取得了较高的识别准确率,而且与传统的水果图像分类识别算法相比较,卷积神经网络避免了人工提取特征的繁琐过程。

In order to solve the defects of traditional fruit image classification and recognition algorithm:the convolution neural network is applied to the recognition of fruit image.Based on the data set created in this paper,the convolution neural network structure which is more suitable for this data set is proposed with reference to the LeNet-5 structure of the classic convolution neural network model.The set is labeled,and apples,pears,oranges,oranges, and peaches are marked as 0,1,2,3 and 4 respectively.Then the pictures are put into model training.The model constructs an input layer,two coiling layer,two pool layer,two full connection layer and one output layer.Convolution neural network extracts features from the bottom layer,and further extracts features from deeper levels,and finally gets the classification of targets.The experimental results show that the proposed convolution neural network structure not only achieves high recognition accuracy on the data set,but also compared with the traditional algorithm of fruit image classification and recognition,and the convolution neural network avoids the complicated process of artificial extraction of characteristics.