Study on discrimination of Glycyrrhizae Radix et Rhizoma (Licorice) varieties and origin traceability based on composite feature reconstruction combined with HLOA-CNN algorithm.

Glycyrrhizae Radix et Rhizoma (Licorice) is a functional food, requiring accurate varieties and origin identification for authenticity. This study hypothesizes that integrating multi-source phenotypic and chemical characteristics with an optimized deep learning model allows for precisely identifying Licorice samples by variety and origin. It collected such samples, extracted phenotypic features (RGB color, visible spectrum, Tamura texture), measured intrinsic components, and built a multi-source feature dataset. Through principal component analysis (PCA) and partial least squares discriminant analysis (PLS-DA), a separation trend was observed among the samples, and 14 key variables distinguishing different varieties were identified (Roughness, 534.7-562.6 nm, etc.) and 8 key variables distinguishing Glycyrrhiza uralensis Fisch (GUF) from different origins (Blue, glycyrrhizic acid, etc.) (VIP >1). The Horned Lizard Optimization Algorithm (HLOA) optimized CNN hyperparameters after reconstructing composite features with multivariate statistical analysis, creating the HLOA-CNN classification model. The model achieved 100 % accuracy in identified Licorice varieties and origins in training and testing datasets. Study results validated multi-source feature fusion and intelligent optimization for authenticating functional foods and provided new quality control and traceability methods.