This study examines the impact of Far-Infrared (FIR) intensity and air temperature on rough rice properties, emphasizing the importance of optimizing these parameters to improve rice kernel quality. The research involved varying FIR power levels (0, 500, 100 Wm⁻²) and inlet air temperatures (30, 40, 50 °C), with results indicating that adjustments in temperature and FIR levels could reduce drying time and specific energy consumption (SEC), albeit potentially leading to increased percentage of cracked kernels (PCK) during rapid drying. Furthermore, the amylose content and viscosity of rice were found to be dramatically affected by incoming air temperature and FIR intensity. Higher temperatures and intensified FIR led to increased amylose levels resulting in greater peak and final viscosities. Moreover, dry heating affected the crystalline structure of the rice; elevated temperatures alongside FIR intensities contributed to enhanced resistance against shear forces leading to a firmer texture. To achieve optimal physicochemical properties during drying processes, it is recommended to optimize within a temperature range of 40–50 °C along with an FIR intensity of 100 Wm⁻². This balance enables control over critical factors such as amylose content, PCK, SEC—all essential for producing high-quality rice products aligned with consumer preferences.

This study introduces a hyperspectral imaging (HSI) system designed to automate rice kernel classification based on quality attributes, offering a non-destructive approach for seed quality assessment. The method leverages machine learning techniques, including principal component analysis (PCA), partial least squares discriminant analysis (PLS-DA), and support vector machines (SVM), to accurately differentiate rice varieties with similar visual appearances but distinct internal quality parameters, such as protein and amylose content. A key innovation of the approach lies in analysing spectral data from a single layer of rice seeds, enabling rapid assessment without the need for labour-intensive kernel separation or surface arrangement. We applied comprehensive pre-processing techniques, including first and second derivatives, and employed optimal wavelength selection through regression coefficient and inverse methods, which enhanced model prediction accuracy. The integration of PCA, PLS-DA, and SVM models achieved classification accuracies of up to 82.86% and 98.86%, respectively, with the combined PCA-SVM approach notably increasing accuracy while minimizing time complexity and overfitting. Our best-performing model demonstrated a classification accuracy of 100%, showcasing the potential of the proposed system for real-time, high-throughput rice quality evaluation. This framework provides a reliable solution for non-destructive seed quality assessment, with promising implications for broader applications in food industry.