Impact of sample preparation form on geographical origin identification of Lonicera japonica Flos by Vis-NIR spectroscopy: a form-aware analytical strategy.

Vis-NIR spectroscopy provides a rapid and non-destructive approach for geographical origin identification of Lonicera japonica Flos, but classification performance may be influenced by sample preparation form and validation strategy. To investigate the influence of sample preparation form, four sample forms-dried flowers, pressed dried-flower pellets, powder, and pressed powder pellets-were evaluated. Form-related spectral variation was first characterized using PCA and PERMANOVA. For geographical-origin classification, preprocessing strategies were evaluated separately for each form, and separate PLS-DA models were developed using Replicate 1. The resulting models were evaluated using Replicate 2, a separate spectral acquisition of the same biological batches, to assess repeated-acquisition performance. Strict Leave-One-Batch-Out (LOBO) validation was additionally performed at the biological-batch level, and RBF-SVM and random forest were used as supplementary classifiers. Sample form significantly affected spectral structure (pseudo-F = 123.7448, p = 0.0001). In Replicate 2, powder showed the highest PLS-DA performance (Accuracy = 0.9375, Macro-F1 = 0.9437, MCC = 0.9239), whereas dried flowers showed the lowest performance. However, strict LOBO validation yielded substantially lower accuracies (0.0000-0.1250), indicating that spectrum-level cross-validation based on technical replicate spectra had overestimated batch-level generalization. The limited dataset, with only eight biological batches and five of six origins represented by a single batch, restricted rigorous independent-batch validation. These findings indicate that sample preparation form substantially affects Vis-NIR spectral characteristics and repeated-acquisition classification behavior, while reliable geographical-origin generalization requires substantially more independent biological batches per origin.