Related Experiment Video
Updated: Sep 19, 2025
Deep Neural Networks for Image-Based Dietary Assessment
Published on: March 13, 2021
Structure from motion-convolutional neural network model (SfM-CNN) achieved accurate portable Chinese dietary
Peihua Ma1, Hsuan Chih Hong2, Xiaoxue Jia1
1Institute of Food Science and Technology, Chinese Academy of Agricultural Sciences, Key Laboratory of Agro-products Processing, Ministry of Agriculture and Rural Affairs, Beijing 100193, China; Department of Nutrition and Food Science, College of Agriculture and Natural Resources, University of Maryland, College Park, MD 20742, United States.
Abstract:
Accurately estimating the chemical composition of dietary intake is essential for health and nutrition management, especially in regions with complex culinary diversity like China. This study introduces a novel AI-driven solution using a Structure from Motion-Convolutional Neural Network (SfM-CNN) model to automate chemical composition analysis of Chinese food. By integrating advanced 3D reconstruction techniques with deep learning, specifically the Scale-Invariant Feature Transform (SIFT) algorithm, we achieved superior feature extraction and food volume estimation with less than 4 % error. Our model, trained on the newly developed ChineseDish-100 dataset, demonstrated an R2 of 0.949 for carbohydrate content estimation using the SIFT-ResNet50 architecture. The model's interpretability was enhanced through visualizations, facilitating parameter optimization and reliable chemical composition estimation. These results underscore the potential of AI-powered models in providing efficient, accurate, and culturally relevant dietary analysis tools, marking a significant advancement for nutritional science, food chemistry, and public health initiatives in culturally diverse regions.
