Fruit Classification using Convolutional Neural Network via Adjust Parameter and Data Enhancement
Liuchen Wu, Hui Zhang, Ruibo Chen, Junfei Yi
Abstract
Fruit is one of the most popular products in the market. Automatic and accurate classification of fruit can bring great convenience to fruit sellers. However, there are great similarities between some apple varieties and pears and peaches, and these kinds of fruit are generally popular, which has increased the difficulty of this task. Aiming at this problem, this paper proposes a method of fruit automatic recognition and classification based on convolutional neural network. First, we obtained two color fruit image data set (public data set and self-made data set). The public data sets is composed of fruit images with simple background, while the fruit images in the self-made data set are taken in a complex environment. Then, on the basis of convolutional neural network, we conducted several research experiments through parameter adjustment, and achieved the highest average classification accuracy of 99.8% on the public data set. In the self-made data set, the classification accuracy is 90.2%. Finally, we improved the classification accuracy of the self-made data set from the original 90.2% to 98.9% by adopting appropriate data enhancement techniques.