Discriminant analysis of compound salvia miltiorrhizas based on laser-induced breakdown spectroscopy combined with deep learning

Ensuring authenticity of herbal products like compound salvia miltiorrhiza tablets is critical yet challenging due to variability in raw materials and manufacturing. Here, we present a novel, non-destructive approach combining laser-induced breakdown spectroscopy (LIBS) with deep learning for manufacturer identification. LIBS spectra from five brands revealed distinct elemental fingerprints, but overlapping patterns in PCA highlighted classification complexity. A convolutional neural network (CNN) with normalization (Nor) preprocessing achieved 99.76% accuracy on an external test set, outperforming many traditional machine learning models. SHapley Additive exPlanations (SHAP) analysis confirmed that model decisions relied on chemically meaningful spectral features (220–390 nm) linked to key elements, enhancing interpretability. This LIBS-CNN framework enables rapid, reagent-free, and accurate source tracing, offering a powerful tool for quality control, anti-counterfeiting, and batch consistency verification in herbal and food products.