A rapid identification study of Atractylodis Macrocephalae Rhizoma and its hot processed products based on composite feature reconstruction combined with OOA-BP algorithm.

Atractylodis Macrocephalae Rhizoma (AMR) is classified as a medicinal and edible homologous food in China. Based on the fusion of multi-source information, this study comprehensively reveals the changing patterns of appearance, texture and composition of AMR during thermal processing. The processing treatments gradually altered the hue (H), saturation (S), and value (V) of the samples, significantly increased the textural heterogeneity of charred samples, and affected the retention or transformation of key components like water and atractylenolide. Additionally, the near-infrared (NIR) spectra showed notable differences in intensity and response patterns between 1550 and 1950 nm. Using principal component analysis (PCA) and partial least squares discriminant analysis (PLS-DA), a separation trend among samples was noted, identifying 8 key variables that differentiate the varieties, including S-value, contrast, atractylenolide II, the 1906.8-1949.2 nm range and etc. After reconstructing composite features using multivariate statistical analysis, the Osprey Optimization Algorithm (OOA) optimized the Back Propagation (BP) model's hyperparameters, resulting in the OOA-BP classification model. This model demonstrated a 100 % accuracy rate in identifying AMR and its thermally processed products across both training and testing datasets. The study provides a scientific foundation and introduces a new framework for medicinal food quality control, improving functionality and enabling quick identification of processed AMR products.