Food Recognition and Nutrition Analysis by Resnet-50

Abstract

Diabetes is a serious problem that is of major concern today. Maintaining and monitoring the diet by keeping in check the nutritional values of food intake can help planning a healthy diet. Numerous Computer Vision techniques are encompassed for automatic Food Image Recognition followed by nutritional values estimation. Advancements in convolutional neural networks grounded on Deep Learning methods have resulted in greater accuracy in the image recognition. This study proposes a fine-tuned model trained on Resnet-50 achieving an accuracy of 81.59% for identifying food images on the Food-101 dataset entailing 101 food categories with 1000 images in each category. With an additional MLP for nutritional values estimation through a CSV file maintained beforehand. The fusion of Convolutional Neural Networks (CNN) and Multi-Layer Perceptron (MLP) approaches for analyzing nutritional values in food images involves combining the distinct strengths of these architectures to enhance the overall system's performance, particularly on complex image datasets like Food 101.