Preprocessing and Classification of SERS Spectra Using SA-CNN

Current analysis of Surface Enhanced Raman Spectroscopy (SERS) primarily focuses on characteristic peaks, determining the type of Raman spectra by observing the intensity and area information of these peaks. However, using only the information from characteristic peaks cannot fully represent the entire spectrum, especially when the characteristic peaks are not prominent, the feasibility of this method is relatively low. In this paper, we utilize deep learning and propose the SA-CNN algorithm. By denoising and baseline correction for preprocessing, followed by the design of a classification algorithm using convolutional neural network, we output the category of the spectrum. We validate the algorithm through experiments with three types of antibiotic Raman spectra, ultimately demonstrating that SA-CNN can effectively classify spectra.