Research on Fruit Category Classification Based on Convolution Neural Network and Data Augmentation

Since fruit category classification plays an important role in fruit production, processing, transportation and sales, research on intelligent methods of fruit classification have a very important practical significance. Based-image machine learning methods of fruit classification are depend heavily on feature design, and its performance is not high. To resolve the defects of traditional intelligent methods, this paper proposes the fruit classification method based on convolution neural network (CNN). Through optimizing the network structure of AlexNet, we obtain a improved network model called IANet. IANet is a 10-layer network which consists of five convolutional layers, four fully connection layers and one output layer. Furthermore, based on Fruits-360 dataset, we use data augmentation methods to generate more train images. Data augmentation improves better classification performance. Finally, we compare the classification performance of IANet with other convolutional neural network models, AlexNet, 13-LayerCNN and HoreaCNN. The experimental results show that the overall accuracy of the IANet without data augmentation achieves 98.06%. While the overall accuracy of the IANet with data enhancement is 98.60%. The performance of fruit classification is better than that of state-of-the-art CNNs model.