SYSTEM DEVELOPMENT OF FOOD INGREDIENT RECOGNITION FOR VEGETARIAN DIET

• Main Author: Fariha Putrisusari, Yasyfiana
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/69181
• Institution: Institut Teknologi Bandung

A vegetarian diet is a type of diet that eliminates meat, or seafood, or sometimes eggs and dairy products from the diet consumed There are several types of vegetarian diets namely vegan, lacto-vegetarian, ovo-vegetarian, and pesce- vegetarian. When traveling to a new place, it is necessary to know the language and the type of ingredients in order to identify the food you want to consume because the cut and presentation of the ingredients differ from region to region. Thus, it is quite difficult for people on a vegetarian diet to identify the ingredients contained in the food to be consumed. The food ingredient recognition system can be a solution to help people on a vegetarian diet to identify their food. Food ingredient recognition is done by utilizing the semantic segmentation model architecture, SegFormer. The SegFormer model architecture is used with the consideration that it has the best results in the benchmarking stage when compared to UNet, DeepLabv3, and SETR. There are two variations of the food ingredient recognition model created, namely SegFormer-All and SegFormer-Negative. SegFormer-All, which was trained using the entire food list in the FoodSeg103 dataset, was selected for use in the system because it had better evaluation results of vegetarian diet type classification than the SegFormer- Negative model. The average precision and recall values for the four vegetarian diet types in the SegFormer-All model were 0.733 and 0.962. The use of random cropping in the training data can improve the performance of the model. The SegFormer-All model successfully reduces interference between food ingredients and overcomes the high variation within a food ingredient class. The segmentation results of the ingredients recognition model are color highlighted on ingredients that cannot be consumed by certain diet types to make the food easier to identify.