Rapid fat quantification in Torreya grandis kernels using portable near-infrared spectroscopy: effects of shelling...

Fat is a dominant nutritional constituent of nuts, significantly influencing their appearance, sensory properties, storage, and processing behavior. As a key quality attribute, rapid and accurate quantification of fat is essential. This study aimed to develop portable near-infrared spectroscopy (NIRS) models for the rapid determination of fat content in Torreya grandis (T. grandis) kernels. Three physical states of the kernels were evaluated: in-shell kernels, de-shelled kernels, and kernel granules. Spectra were acquired using a Smart Eye 1700 portable NIR spectrometer, while reference fat contents were determined using the traditional Soxhlet extraction method. Principal component analysis (PCA) was employed to explore spectral structures, and outliers were identified and removed using Mahalanobis distance combined with concentration residual analysis. Following outlier removal, the dataset was partitioned into calibration and prediction sets via the SPXY (sample set partitioning based on joint x-y distances) algorithm. Partial least squares regression (PLSR) models were then established utilizing various preprocessing strategies.The results demonstrated that the kernel granules provided the optimal prediction performance. The best-performing model, designated SG-SNV-PLSR-K, achieved a determination coefficient of the calibration set ( $R_c^2$ ) of 0.92 and a determination coefficient of the prediction set ( $R_p^2$ ) of 0.89. Compared to the optimal de-shelled kernel model (MSC-PLSR-DS), the granule-based model improved calibration and prediction accuracy by 3.99% and 12.65%, respectively. Furthermore, compared to the in-shell model (SNV-PLSR-IS), the accuracy improved by 16.32% and 20.36%, respectively.These findings indicate that portable NIRS, combined with chemometric analysis, offers an effective and high-throughput approach for fat screening in T. grandis kernels. The developed workflow not only facilitates rapid quality assessment for this specific species but also provides a transferable methodology for fat analysis in other nut varieties.