A Comprehensive Machine Learning Framework for Detection of Red Sumac (Rhus coriaria) Adulteration with Barberry Using Hyperspectral Imaging

This research proposes a comprehensive machine learning framework for detecting red sumac adulteration with barberry using hyperspectral imaging. Various features including statistical features, Discrete Cosine Transform (DCT) features, and shape features were extracted and analyzed for classification performance. Results demonstrate that among traditional classifiers, Random Forest achieved the highest accuracy (۹۱.۱۱%) when utilizing all features, with statistical features having the most significant impact on classification. Among deep learning approaches, LSTM networks outperformed all other methods with accuracy exceeding ۹۶%, demonstrating the importance of sequential spectral information in adulteration detection. Further analysis revealed that even with aggressive spectral down sampling (۳۰×), the LSTM model maintained high accuracy (۹۳.۹%), suggesting potential for implementation in cost-effective systems. These findings advance food authentication technology by offering both high-performance detection methods and practical implementation strategies for identifying red sumac adulteration, contributing to food safety and quality assurance efforts.

کلیدواژه ها:

Hyperspectral Imaging ، Sumac adulteration ، Machine Learning ، Feature Extraction ، Deep Learning