Geographical origin discrimination of goji berry using complementary Raman and NIR spectroscopy

The geographical origin traceability of goji berry is essential for ensuring product quality and consumer protection. However, single-modality spectroscopy often fails to capture subtle compositional differences among producing regions. To address this issue, a Multimodal Collaborative Attention Network (MCAN) was developed to integrate Raman and near-infrared (NIR) spectral data by exploiting their complementary information. The model incorporates collaborative attention and attention-based feature fusion to enable effective cross-modal interaction and adaptive integration of Raman and NIR features. Experimental results show that MCAN achieves an accuracy of 99.07%, outperforming single-modality and conventional fusion models. Furthermore, cross-dataset validation on an independent Chinese yam dataset suggested the transferability of the proposed framework to another plant-origin classification task, achieving an accuracy of 98.25%. Feature visualization confirms improved inter-class separability through multimodal fusion. This study provides a robust and interpretable approach for non-destructive geographical origin traceability of agricultural products.