Abstract:
This study investigates the application of Convolutional Neural Networks (CNNs) for food identification, categorization, and nutritional prediction, responding to the increasing need for automated dietary evaluation tools in healthcare and consumer sectors. The study employs extensive datasets, including Food-101 for food classification and a tailored nutrition dataset for estimating nutritional values, to establish a comprehensive system for food image analysis. The study utilizes four advanced CNN architectures—InceptionV3, VGG, ResNet, and EfficientNet—for feature extraction and model training, each selected for its distinctive capacity to grasp intricate visual patterns and nuances seen in food photos. The experimental framework entails a thorough assessment of different architectures utilizing standard performance criteria. In food classification, accuracy, precision, and recall are calculated to evaluate the models' effectiveness in identifying and categorizing a variety of food products. The study evaluates nutrition prediction performance using regression-based metrics, including mean absolute error (MAE) and root mean squared error (RMSE). Comprehensive testing and cross-validation reveal that, although all models exhibit effective performance, ResNet regularly surpasses its competitors in categorization and nutrition prediction tasks. In addition to the quantitative findings, the study offers insights into the practical ramifications of implementing deep learning models in real-world dietary monitoring systems. CNNs' capacity to autonomously extract pertinent elements from intricate food images presents a promising opportunity for creating intelligent, userfriendly applications that aid users in monitoring their nutritional intake and making informed dietary decisions. The findings highlight the capacity of deep learning to reconcile old manual dietary assessments with contemporary data-driven methodologies, thereby enhancing public health outcomes.
