Transfer Learning Applied to Algerian Gastronomy Images Recognition

Abstract

Automatic food recognition has become a relevant necessity in the computer vision (CV) systems industry, i.e., the mobile applications industry for intelligent food delivery or smart agriculture, food ripeness rates, etc., and a topical scientific subject for ongoing scientific research, and a highly topical scientific subject for which scientific research is ongoing. to advance and improve the algorithms used in the field, as well as the performance of this type of embedded system, which on the one hand is in direct contact with the real environment, and on the other is tolerant of a certain acceptable error rate given people's state of health and dietary desires. Among these systems is machine learning (ML), which is part of the systems generated by artificial intelligence (AI), and among its sub-disciplines is deep learning (DL), which is a widely used approach in the field of artificial neural networks (ANN). In this context, we will explore these concepts and take part in the development of an automated classification system for images of Algerian gastronomy, based on deep learning using the idea of transfer learning from the best-performing classical models in the field. Image classification, such as InceptionV3, resNet, etc. Experiments will be carried out on an existing dataset developed by the Ministry containing 6 food categories, and we have increased this dataset to 9 categories