Large Scale Image Classification of Exotic Fruits in Indonesia Using Transfer Learning Method with MobileNet Model

Abstract

Exotic fruit is a fruit that is not widely known to the public. In Indonesia, there are many exotic fruits such as rambutan, passion fruit, mangosteen, longan, guava, and many more. Classification of exotic fruit images is needed because of the lack of knowledge from outsiders about exotic fruits in Indonesia. To his end, developing robust artificial intelligence using deep learning is necessary. CNN is the development of the Multilayer Perceptron (MLP) which is designed to process two-dimensional data and in the type of Deep Neural Network because of the high network depth and widely applied to image data. By utilizing the transfer learning method and a little fine-tuning, the efficient model like MobileNet expected to be better than without transfer learning in FruitNet model. Our contribution is applying efficient transfer learning MobileNet for Exotix Fruits in Indonesia which achieves 87% accuracy in average using more than 1000 images. The model performs better than previous model of FruitNet which only reaches 43% accuracy in average.