Fig. 1: Schematics of this work: The proposed self-supervised learning-based Raman spectral preprocessing (RSPSSL) scheme and the experimental design for its comprehensive performance evaluation.

a The general pipeline of the RSPSSL scheme includes spectral training datasets, an RSGAN model, and an RSBPCNN model. b Experimental design for the spectral preprocessing performance evaluation through multiple biomedical applications, including cancer diagnosis, paraquat concentration quantification, and Raman hyperspectral image preprocessing