Near-Infrared Spectroscopy-Based Chilled Fresh Lamb Quality Detection Using Machine Learning Algorithms

Abstract

Traditionally methods for assessing mutton quality rely on physical and chemical examination analyses that necessitate precise experimental environment conditions and specialized knowledge, often resulting in the compromise of the sample's structural integrity. To address these challenges, this study explores the application of near-infrared spectroscopy (NIR) as a non-destructive alternative for mutton quality evaluation, leveraging its operational simplicity, rapid analysis capabilities, and minimal requirement for technical expertise. Among various spectral data preprocessing techniques evaluated, multiple scattering correction (MSC) was found to significantly enhance model detection performance. Furthermore, principal component analysis (PCA) combined with the Mahalanobis Distance method was utilized for outlier identification. Finally, a mutton freshness detection model is constructed based on stacking ensemble learning, yielding an impressive accuracy rate of 0.976, outperforming other advanced approaches. In conclusion, our findings establish a robust theoretical framework for the rapid and non-destructive assessment of meat freshness, contributing to advancements in meat quality detection.