Non-Destructive Authentication of Rice Varieties Using Hyperspectral Imaging and Machine Learning

This study develops a non-destructive method for authenticating four commercial rice varieties: Sargol, Domsiyah, AliKazemi, and Aria-Paria, by integrating hyperspectral imaging (HSI) with machine learning. Hyperspectral data in the visible–near infrared range (۴۰۰–۹۵۰ nm) were acquired from ۴,۳۰۵ individual rice grains. Effective preprocessing mitigated initial anisotropic spatial sampling by resizing images to achieve isotropic resolution, preventing grain loss during segmentation. Two analytical strategies were investigated: one relying on handcrafted features and another directly exploiting reduced spectral profiles as sequential data. Comparative evaluation showed that models trained on sequential spectral information consistently outperformed feature-based methods. Among the evaluated classifiers, a Support Vector Machine (SVM) achieved the highest classification accuracy of ۹۲.۶۲%, exceeding both classical machine learning models and deep learning approaches, including Long Short-Term Memory (LSTM) networks, one-dimensional Convolutional Neural Networks (۱D-CNNs), and a hybrid CNN-LSTM architecture. The proposed HSI–SVM framework demonstrates strong potential for accurate rice variety authentication and offers practical applicability in quality control and supply chain monitoring.

کلیدواژه ها:

Hyperspectral Imaging ، LSTM ، SVM ، ۱D-CNN ، Feature Extraction ، Rice Variety Identification