Gentiana rigescens Franch. ex Hemsl. (G. rigescens) is a well-known medicinal plant in China, which has been affected by environmental changes and human activities, and the quality of G. rigescens varies from different origins, which increases consumers' concerns. Therefore, this study employs Fourier Transform Infrared Spectroscopy (FTIR) combined with machine learning methods to construct a rapid, accurate, and comprehensive approach for the precise classification and identification of origin and plant part of G. rigescens. The results indicate that mean diurnal range, temperature seasonality, and elevation are significant factors contributing to the differences in the growth environments of G. rigescens. When using FTIR spectroscopy to identify the origin and parts of G. rigescens, the application of appropriate preprocessing (second derivative, SD) and feature extraction (Intersection) could effectively enhance the performance of the model. Transforming the FTIR spectral data preprocessed by the SD into two-dimensional correlation (2DCOS) images enabled the Residual Convolutional Neural Network (ResNet) to accurately identify the origin and parts of G. rigescens. The classification accuracy rates for the training set, testing set, and external validation set all reached 100.00%, with no risk of overfitting. The prediction results of the Maximum Entropy (MaxEnt) model indicated that the moderately to highly suitable habitats for this species were mainly distributed in the Yunnan-Guizhou Plateau and the Hengduan Mountains area in northwestern Yunnan. Conditions with small temperature variations and sufficient precipitation were conducive to the accumulation of the active ingredients in G. rigescens. This study can provide a reference for the scientific cultivation, quality evaluation, and resource conservation of G. rigescens.
扫码关注我们
求助内容:
应助结果提醒方式:
