Adsorption Isotherm Analysis of SERS Spectroscopy via a Hybrid CNN-Transformer Approach

The analysis of adsorption isotherms serves as a critical tool for elucidating interfacial processes in surface-enhanced Raman scattering (SERS) studies; however, it has long been impeded by the subjective interpretations and inherent inefficiencies of conventional fitting methods. To overcome these limitations, a hybrid deep learning (DL) framework that combines one-dimensional convolutional neural networks (1D CNNs) with a transformer was designed for the identification and classification of adsorption isotherm models. Experimental validation was conducted utilizing SERS chips to adsorb crystal violet, malachite green, and methylene blue. The trained model attained classification accuracies surpassing 91% on simulated data and shown strong performance in modeling genuine experimental isotherms. This completely automated method markedly improves objectivity and efficiency in adsorption analysis. The results not only offer a practical instrument for SERS-based adsorption studies but also present a synthetic data-driven DL framework, providing a scalable resolution for small-sample learning difficulties in experimental chemistry.