基于并联卷积神经网络的水果品种识别

为了解决传统的水果图像识别算法在特征提取上的缺陷,以及传统卷积神经网络识别率低的问题,设计了一种基于并联卷积神经网络来提取水果特征的识别方法,利用ELU激活函数替代ReLU激活函数,利用最大类间距损失函数结合传统SoftmaxWithLoss损失函数来提高对相似品种的识别准确率。选取Fruit-360数据集中的8个品种,利用边界均衡生成对抗网络(BEGAN)结合传统的数据增强方法生成大量高质量的数据集,并用其进行训练。结果表明,该模型对8个品种的平均识别准确率达98.85%,具有良好的识别效果。

Fruit variety recognition based on parallel convolutional neural network

In order to solve the defects of traditional fruit image recognition algorithms in feature extraction and the low recognition accuracy of traditional convolutional neural networks, a parallel convolutional neural network was proposed to extract fruit features. ELU activation function was introduced instead of ReLU activation function in the proposed model. Besides, a combination of maximum class spacing loss function and the traditional SoftmaxWithLoss loss function was designed to improve the recognition accuracy of similar varieties. The data of 8 fruit varieties in Fruit-360 data set was selected in the present study, and enhanced by the boundary equilibrium generative adversarial network (BEGAN) combined with the traditional data augmentation to generate a large number of high-quality data for model training. It was shown that the average recognition accuracy of 8 fruit varieties reached 98.85% and exhibited good recognition effect.