Similarity Classification of Indonesian Traditional Snack Using YOLOv8 Algorithm

The main objective of this research is to develop a model that can classify the similarities between traditional Indonesian snacks. Images of traditional Indonesian snacks were selected based on their similarity in shape, texture, and colour. The image collection is carried out by going directly to shops that sell traditional snacks. In this study, we modified the Y olov8 architecture used for classification tasks. This modification involves two main changes: the replacement of the backbone from EfficientNet to ResNet-18 and the replacement of the activation function from ReLu to Swish in the ResNet-18 architecture. The model was trained over 95 epochs. The dataset used in this study comprises 1989 images of traditional Indonesian snacks, spanning 16 classes and divided into 70% training data, 20% validation data, and 10% test data. Experimental results demonstrate the success of the modified YOLOv8 algorithm in classifying similar snacks, achieving an accuracy of 94,18%, precision of 94,36%, F1-score of 94,23%, and loss of 0.003.