Classification of Indonesian Traditional Snacks Based on Image Using Convolutional Neural Network (CNN) Algorithm

Abstract

Some people consider traditional snacks are out of date. Many of the traditional snacks were abandoned by the community and began to switch to more modern foods so that people sometimes do not recognize the traditional cakes in circulation. This study to develop model recognition traditional Indonesian snacks. As technology development, image recognition using the Convolutional Neural Network (CNN) method as classification method using the pre-trained MobilenetV2 model as the basic model can be used. From total dataset of 1545 images of traditional cakes consisting of 8 categories, they are divided into 80% train data and 20% test data. After going through the training and testing process, the accuracy results are 98.9% for train data and 90.5% for test data. Model testing performed on the new test data resulted in an accuracy of 92.5% where the model managed to classify 74 images from 80 images of traditional cakes according to their categories which were presented in the form of confusion matrix. Several experiments were also carried out to find the parameters that produce the model with the best accuracy, namely the effect of the number of epochs, the effect of the dataset distribution scenario, and the effect of the size of the learning rate.