A Non-Invasive Method to Classify the Sweetness Levels of Apples

Abstract

Since the ability of convolution Neural Networks (CNN) on image classification has been discovered by research workers. Many studies revealed using CNN on image classification can make a breakthrough continuously. The CNN can be used for many different researches. However, there is less study using CNN to classify or recognize the sweetness of the fruit. Therefore, this paper will use CNN model to develop a non-invasive classifier for the sweetness levels of the apples that are sweet, normal and not sweet. The researchers of this paper collected a total of 130 apples to generate the dataset. There are 130 average sweetness data of the apples and 1506 apples appearance images in the dataset. Also, five different parameter settings were used to train AlexNet. The results of training were compared and validated between each setting. The best experience result of training and validation accuracy comes to 99.86% and 81.00% separately in this paper. As a consequence, the result revealed CNN has ability on fruit sweetness classification.