Related Experiment Video
Updated: Jul 16, 2026
13:19
Deep Neural Networks for Image-Based Dietary Assessment
Published on: March 13, 2021
Image-Based Prediction of Food Weight and Nutritional Composition in Bowl-Served Meals Using Semantic Segmentation
Xu Ji1, Yiran Feng1,2,3, Haolin Lu1
1Department of Mechanical Engineering and Automation, Dalian Polytechnic University, Dalian 116034, China.
Nutrients
|July 15, 2026
Summary
This study introduces an advanced image-based method for predicting nutritional content in bowl-based meals, overcoming challenges like food occlusion and adhesion. The approach enhances accuracy in dietary assessment and intelligent nutrition management.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Nutritional Science
Background:
- Image-based dietary assessment offers an intuitive method for nutritional monitoring.
- Challenges in bowl-based meals include food adhesion, stacking, and occlusion, impacting accuracy.
- Existing methods struggle with precise volume, weight, and nutritional prediction in complex meals.
Purpose of the Study:
- To develop an accurate nutrition prediction method for bowl-based foods using image analysis.
- To address challenges of food adhesion, stacking, and occlusion in multi-category meals.
- To improve the stability of volume and nutritional composition prediction for dietary assessment.
Main Methods:
- Integrated semantic segmentation (DBP-FDSNet), multi-view 3D reconstruction, and occlusion compensation.
- Improved DBP-FDSNet with detail enhancement and boundary-assisted supervision for segmentation quality.
- Employed a layered compensation strategy for occluded staple foods and volume error reduction.
Main Results:
- DBP-FDSNet achieved a mean Intersection over Union (mIoU) of 80.51% and Boundary F1 Score (bF1) of 85.73%.
- Whole-bowl Mean Absolute Percentage Error (MAPE) for Calories, Fat, Carbohydrate, Protein, and mass were 13.23%, 18.51%, 14.18%, 13.35%, and 10.85%, respectively.
- Demonstrated improved stability in category-level volume and nutritional prediction.
Conclusions:
- The proposed method offers a feasible solution for image-based dietary assessment in complex meal scenarios.
- Successfully enhances the accuracy of nutritional prediction for bowl-based foods.
- Provides a foundation for intelligent nutrition management systems.
