Advances in deep learning for intelligent non-destructive spectral analysis: From preprocessing and data augmentation to interpretability, fusion, and transfer learning