Enhanced Discrimination of Thawed and Nonfrozen Chicken Thighs Using Convex Hull Peeling in Visible Spectral Imaging

Abstract

Hyperspectral imaging (HSI) combines spectral and spatial data, producing complex 3D datasets that require efficient data reduction methods for improved computational efficiency and prediction accuracy. This study introduces convex hull peeling to enhance the discrimination of thawed and nonfrozen chicken thighs. By removing pixels with noise-dominated spectra and targeting deeper data layers, this method improved model robustness and reduced training time from 426 to 5 s. Essential spectral pixels (ESPs), located on the convex hull in principal component space, effectively preserved critical data, achieving 81% classification accuracy, comparable with using the full dataset. Sensitivity and specificity were 74% and 89%, respectively, demonstrating improved specificity with a slight trade-off in sensitivity. Piece-based accuracy reached 100%, highlighting the potential of this approach for noninvasive food quality assessment. This study underscores the efficiency and adaptability of ESPs and convex hull peeling for complex datasets.