Food Recognition System Using Classification Methods on UEC

EasyChair Preprint no. 10006

Abstract

The development of an autonomous food recognition system has a number of intriguing applications, including food waste management, marketing, calorie estimation, and daily nutrition tracking. Despite the importance of this topic, there aren't many papers on it. Furthermore, the literature comparison only considered best-shot performance, rather than the most popular approach of averaging across multiple trials. This experiment achieves the most recent accuracy of 90.02% on the UEC Food-100 database, outperforming the current best-shot performance experiment. This paper surveys the most popular deep learning methods used for food classification, presents publicly accessible food databases, publishes benchmark results for the food classification experiment averaged over 5-trials, and surveys publicly accessible food databases. The best results were obtained using the ensemble technique, which averaged the predictions of the ResNeXt and Dense Net models. The UEC Food-100 database and other food databases are used for all tests because it is one of the most popular datasets and because it presents a challenge due to the presence of multi-food photographs that must be clipped before processing.

Keyphrases: Artificial Neural Networks, image classification, machine learning

@Booklet{EasyChair:10006,
  author = {R Sakthi Vel and A Arul Selvan Gnanamonickam},
  title = {Food Recognition System Using Classification Methods on UEC},
  howpublished = {EasyChair Preprint no. 10006},

  year = {EasyChair, 2023}}