NutriFoodNet: A High-Accuracy Convolutional Neural Network for Automated Food Image Recognition and Nutrient Estimation

Abstract

To detect food items in images using Convolutional Neural Networks (CNNs) plays a crucial role in promoting healthier dietary decisions and addressing global nutrition issues. With the rise of online food delivery systems, precisely discerning food items within images and gauging their nutritional components stands as a pivotal undertaking to ensure that people are consuming a balanced diet. Due to its capacity to identify and reliably classify images, CNN is a successful approach for image recognition. By using CNN for food recognition, it is possible to automate the process of nutrient estimation and provide users with more information about their food choices. This might have a substantial effect on public health by encouraging a healthy diet and reducing the incidence of malnutrition in all its forms. An efficient food image recognition method is developed using a convolutional neural network named NutriFoodNet. Popular pre-trained models like ResNet-18, ResNet50 and Inception V3 were at the center of our attention. A model called NutrifoodNet is developed by modifying the Inception V3 model by using the well-known Food101 dataset, which includes 101,000 picture samples of 101 food varieties. To gauge the model's efficacy, it's imperative to consider metrics such as precision, classification accuracy, F1 score, and recall as fundamental benchmarks. A comparative study was also conducted using up-to-date benchmarks. The results indicated that NutriFoodNet achieved a classification accuracy of 97.3%, outperforming other leading-edge models. An Algorithm is proposed to find the calorie information from different nutrients and comparison with the existing models is also done.