Related Experiment Videos
Rice Safety Risk Data Generation Model Based on Gated Feature Transformation and Adaptive Diffusion.
Huawei Jiang1, Ruomeng Hu1, Wanbao Sheng1
1College of Information Science and Engineering, Henan University of Technology, Zhengzhou, China.
Journal of Food Science
|June 4, 2026
Summary
This study introduces RADiff, a novel data generation model that creates high-fidelity rice contaminant data, effectively addressing data scarcity in food safety assessments. The model significantly improves risk evaluation by providing reliable data for analysis.
Area of Science:
- Food Science and Technology
- Data Science and Artificial Intelligence
- Environmental Science
Background:
- Food safety risk assessment for rice is crucial but hindered by data scarcity due to costly laboratory testing.
- Effective evaluation of chemical contaminants like heavy metals and mycotoxins in rice requires comprehensive monitoring data.
Purpose of the Study:
- To develop a high-fidelity tabular data generation model (RADiff) for rice risk assessment under small-sample conditions.
- To address limitations in traditional data generation methods by incorporating adaptive mechanisms and advanced network architectures.
Main Methods:
- Proposed RADiff model utilizing an adaptive layer normalization (AdaLN) mechanism for parameter adjustment.
- Designed a residual gated cross transformation (RGCT) network to model interdependencies between heavy metals (Pb, Cd, iAs) and mycotoxins (AFB1, BaP).
- Optimized the reverse diffusion process with skip sampling and an x0-prediction paradigm for enhanced efficiency and quality.
Main Results:
- Generated data using RADiff demonstrated high consistency with real-world empirical data in statistical distribution.
- RADiff significantly outperformed baseline models, including SMOTE, with lower Wasserstein distance (0.0029) and KL divergence (0.0061).
- Incorporating generated data improved performance in downstream food safety risk classification tasks.
Conclusions:
- The RADiff model effectively alleviates the problem of limited rice sample data for food safety evaluations.
- Generated data provides reliable support for downstream risk assessment tasks, enhancing the overall food safety framework.
- The study validates the utility of advanced AI models in addressing real-world challenges in food safety monitoring and analysis.