Glycemic Index Classification in Food Images: A Deep Transfer Learning Approach

This paper presents an innovative approach to categorize food images based on their glycemic index (GI) value using deep transfer learning. By employing pre-trained models, we aim to enable individuals to make healthier dietary choices. The proposed research explores transfer learning for its adaptability, reduced data needs, and enhanced generalization. VGG16, DenseNet, and ResNet50 models are used for transfer leering and classification of food images into low, medium, and high GI values. Interpretability techniques are also incorporated to provide visual explanations of the model’s predictions. Experimental outcomes in terms of standard metrics and curves validate the effectiveness of the transfer learning methods in classifying GI of foods. Aligning with rising health-consciousness, the proposed work contributes to computational nutrition by integrating machine learning (ML) with nutritional knowledge for developing solutions that promote healthier lifestyles.