EzzeRecipe: Ingredient detection and Recipe Recommendation Mobile Application Using CNN

Abstract

The most fascinating fact of this “EzzeRecipe” Application is it implements the Convolutional Neural Network (CNN) method. This application is aimed to make cooking easy and interesting for beginners and build basic cooking skills. This mobile application recommends mouthwatering recipes by detecting ingredient images taken using mobile camera or gallery. CNN used for ingredient detection. The CNN model is based on custom ingredient dataset. The dataset includes 10 types of vegetable and other ingredient images captured with mobile phone. After labelling, preprocessing dataset was Segmented into train (70%) validation (10%) and test (20%) dataset. Then dataset was trained to build CNN model and converted into TesorflowLite model. Finally integrated to flutter Application. With 3300 images dataset, the model produces an excellent accuracy of 95.91%. Also tried EfficientNet on my custom dataset and got 60% accuracy. Flutter application is able to Fetch Recipe from database and display on mobile screen.