Transfer Learning for Classification of Fruit Ripeness Using VGG16

Abstract

Early diagnosis of maturity carried out by experts in laboratory tests is often not applicable for fast and inexpensive implementation. Using deep learning, an image of various fruits used as data input. Training deep learning models requires large, hard-to-come datasets to perform the task in order to achieve optimal results. In this study. There are 4 research objects, namely apples, oranges, mangoes, and tomatoes used totaling around 9000 training data. Data were trained using 200 epoch iterations using the transfer learning method with the VGG16 models. At the top layer of both models, the same MLP is applied with several parameters, data is converted from RGB to L * a * b with the aim of being a color descriptor on the fruit. Trained using CNN VGG16 with the transfer learning method. The Dropout 0.5 shows the best performance of experiment with 4 scenario that used different technique and show result the best performance with an average score of accuracy rate from scenario 4 is 92%.