Identification of Hybrid Indica Paddy Rice Grain Varieties Based on Hyperspectral Imaging and Deep Learning

Area of Science:

• Agricultural Science
• Data Science
• Spectroscopy

Background:

• Accurate classification of paddy rice varieties is crucial for quality control, impacting food security and market value.
• Hyperspectral imaging coupled with machine learning offers potential for precise rice variety classification.
• Challenges include managing high-dimensional spectral data and ensuring model interpretability.

Purpose of the Study:

• To develop and evaluate an optimized CNN-Transformer model for hybrid indica paddy rice grain variety classification.
• To address challenges in spectral data dimensionality and variability.
• To enhance model interpretability in rice variety identification.

Main Methods:

• Utilized a CNN-Transformer model integrated with Standard Normal Variate (SNV) preprocessing.
• Employed Competitive Adaptive Reweighted Sampling (CARS) for optimal feature wavelength selection.
• Conducted interpretability analysis to understand model decision-making.

Main Results:

• The CNN-Transformer model achieved a classification accuracy of 95.33% and an F1-score of 95.40%.
• The model demonstrated superior learning from key wavelength features compared to baseline models.
• Identified critical spectral bands for classification: 400-440 nm, 580-700 nm, and 880-960 nm.

Conclusions:

• Hyperspectral imaging combined with the proposed machine learning approach provides a powerful and interpretable method for automated rice quality control.
• The study highlights the effectiveness of the CNN-Transformer model for precise hybrid indica paddy rice classification.
• Key spectral regions were identified, aiding future research and application in agricultural practices.