U-Mamba-Spectra: A novel generative and explainable framework for camel milk adulteration detection using...

Area of Science:

• Food Science
• Analytical Chemistry
• Spectroscopy

Background:

• Camel milk's high value and limited supply make it susceptible to adulteration.
• Authenticity verification is crucial for consumer safety and market integrity.
• Existing detection methods may lack interpretability or comprehensive accuracy.

Purpose of the Study:

• To develop an interpretable framework (U-Mamba-Spectra) for detecting camel milk adulteration.
• To enable both qualitative and quantitative analysis of adulteration using near-infrared (NIR) spectroscopy.
• To enhance model robustness, especially under limited sample conditions.

Main Methods:

• Utilized near-infrared (NIR) spectroscopy (780-1830 nm) on 460 camel milk samples with varying adulteration levels (0-100%).
• Developed a U-Mamba classifier integrating U-Net and Mamba modules for spectral dependency analysis.
• Employed SHAP analysis for identifying key spectral bands and a two-stage MCR-ALS-RF model for quantitative prediction.
• Implemented a hybrid generative model (CVAE-CWGAN-GP-SAM) for spectral data augmentation.

Main Results:

• The U-Mamba classifier outperformed conventional machine learning and deep learning models in classification accuracy.
• SHAP analysis identified spectral bands linked to water, protein, and fat content.
• The MCR-ALS-RF model achieved high accuracy (recoveries 96.8-99.5%, RSD < 5%) and low detection limits (0.025-0.060 g/100 g).
• Spectral augmentation improved classification accuracy from 95.21% to 98.60%.

Conclusions:

• U-Mamba-Spectra provides an effective and interpretable solution for camel milk adulteration detection.
• The framework demonstrates strong performance in both qualitative and quantitative analysis.
• The developed spectral augmentation technique enhances model robustness with limited data.