Raman spectral analysis of Chinese yam for geographical origin discrimination using an attention-enhanced deep...

Dioscorea spp. (Chinese yam) is a medicinal and edible plant whose chemical composition and spectral characteristics are strongly influenced by geographical origin. Developing a rapid and non-destructive Raman spectroscopic strategy is important for characterizing origin-related spectral variations in yam samples. In this study, an attention-enhanced deep learning approach was developed for the analysis of complex Raman spectra of Chinese yam. By integrating multi-scale feature extraction with adaptive feature enhancement, the proposed method effectively mitigates peak overlap, improves the representation of weak origin-discriminative spectral patterns, and suppresses background interference. A total of 960 Raman spectra were collected from 192 yam slice samples across six major production regions. The proposed method achieved a classification accuracy of 98.08%, outperforming conventional machine learning models and standard 1D-CNNs. Attention visualization further revealed that the model focused on chemically meaningful Raman fingerprint regions, indicating improved interpretability in spectral feature extraction. These results demonstrate that Raman spectroscopy combined with attention-enhanced deep learning provides an effective and non-destructive strategy for geographical origin discrimination of Dioscorea spp. and spectral analysis of complex medicinal and edible plant matrices.