Medicinal and edible homologs (MEHs) offer significant preventive and therapeutic benefits for various diseases and health functions. However, the widespread application of MEHs faces significant challenges, particularly in quality control and rapid identification. In this study, we present a novel approach that combines surface-enhanced Raman spectroscopy (SERS) based on spectral set, referred as "SERSome", with deep learning to develop an identification model for analyzing MEHs. The platform uses silver nanoparticles prepared via reduction with NaBH4, activation with sodium borohydride, and aggregation with calcium ions. The method avoids the use of additional protective agents during the reaction process, thereby reducing interference from the protective agents and other materials. Additionally, the method also overcomes the fluorescence interference from MEHs. SERSomes provide comprehensive molecular fingerprint recognition, significantly enhancing the system's detection accuracy. The introduction of Ca2+ as aggregation agent promotes the aggregation of silver nanoparticle, significantly enhancing the electromagnetic field of the nanoparticle system. The study achieves up to 98 % accuracy in identifying specific MEHs. The integration of SERSomes with deep learning offers a promising methodology for the rapid and modern detection of MEHs, advancing the MEHs industry and contributing to the protection of human health.