Lipid fingerprinting integrated with deep learning for rapid authentication of 50 seed-derived medicinal herbs: Toward an automated identification platform

Seed-derived medicinal herbs (SDMHs) are rich in lipid components widely applied in food fields, contributing to flavor modulation and lipid metabolism regulation. However, their highly similar morphological features often lead to adulteration and misidentification. In this study, 608 batches from 50 SDMH species were analyzed using high-performance liquid chromatography coupled with evaporative light-scattering detection, which effectively separated fatty acids, phospholipids, and triacylglycerols within 60 min and enabled the construction of a comprehensive lipid fingerprint database. Seven machine learning and one deep learning models were compared, among which a one-dimensional convolutional neural network demonstrated superior performance, achieving 100 % accuracy on the training set and 98.4 % on the test set. Based on this model, an interactive identification platform was developed, integrating herb images, major lipid components, chromatographic fingerprints, and botanical information. The platform enables rapid and accurate authentication of SDMHs and makes it user-friendly to non-specialist users. This study presents a scalable lipid profiling strategy and an intelligent tool for SDMH quality control and authentication.