ProFoodNet: Enhancing Protein Prediction from Food Images Using Advanced Machine and Deep Learning Techniques

In this study, we present ProFoodNet, an advanced protein prediction system from food images leveraging both machine learning and a seven-layer deep Convolutional Neural Network (CNN). ProFoodNet aims to address the challenge of accurately estimating protein content in food items, which is crucial for managing dietary intake in individuals with protein-related health conditions. Our approach utilizes the image database IDPE containing 990 images of various food products. First, we employ gradient-based edge detection operators (Prewitt, Sobel, and Kirsch) to extract image features. Two prediction models are then trained and tested using these features: a deep CNN and a linear regression model using a Support Vector Machine (SVM). The deep CNN model outperforms the SVM-based model by achieving the lowest average prediction error (±1.94), according to experimental results. Our findings highlight the potential of advanced machine and deep learning techniques in improving the accuracy of protein prediction from food images, facilitating dietary management and personalized nutrition advice.