Fingerprint spectrum construction for Chinese wolfberry (CW) origin tracing and key element prediction via multi-source data fusion and intelligent algorithms.

Chinese wolfberry (CW), an essential traditional functional food in China, faces challenges like market confusion about its origin and counterfeiting. A multimodal workflow integrating computer vision and inductively coupled plasma mass spectrometry (ICP-MS) was applied to 120 CW batches from four regions. Chromaticity, visible-light response, Gabor texture, and elemental data were combined to construct a fingerprint for origin discrimination and elemental prediction. Multivariate statistical analysis identified As, Zn, Fe, Energy, yellow-light response, CIE Z, and b⁎ as key variables with variable importance in projection (VIP) scores greater than 1. After VIP screening, convolutional neural network (CNN) achieved 100.0% training and held-out test accuracy, with a five-fold cross-validation balanced accuracy of 0.988 ± 0.018 and a test Macro-F1 of 1.000; processing time decreased by 31.5%. Temporal Convolutional Network-Gated Recurrent Unit (TCN-GRU) yielded test R2 values of 0.967-0.977 and RPDp values of 5.52-6.59 for As, Zn, and Fe, outperforming conventional regressors. These findings support multimodal CW evaluation, while broader validation will further improve model robustness.