Bangladeshi Food Detection Using Yolov5

Abstract

Bangladesh is a country of having diverse cuisine. Many people from different parts of the world visit Bangladesh. Bangladeshi cuisine has is very popular to people of many countries. But many people are unfamiliar with the names of different Bangladeshi dishes. So this is a Machine Learning Approach to make the Bangladeshi dishes visible around the world. I have created a dataset classifying 21 Bangladeshi dishes using machine learning. After collecting images for processing the dataset was created. After creating the dataset, the Yolov5 model was trained on the dataset. The model showed good performance having mAP_0.5 score of 97.5% and mAp_0.5:0.95 score of 92.1%. The model was able to detect all the given classes correctly. This dataset can be useful in various purposes in future. It can be used in calorie estimation, dietary assessment and other purposes. There are future scopes for this project. The model can be used to build an Android Application to detect foods from images and videos. The application can be useful for children to get familiar with different dishes and also for tourists who don’t recognise Bangladeshi foods properly.