Establishment and optimization of a prediction model for the chemical constituent of Leonurus japonicus Houtt. based on multispectral analysis and machine learning

As the demand for Leonurus japonicus Houtt. continues to rise and its cultivation scale expands, there is an urgent need to professionalize and refine its cultivation management. To address this, the introduction of rapid, non-destructive monitoring techniques has become crucial for guiding scientific cultivation. This study aimed to develop a method combining multispectral imaging and machine learning to estimate the chemical constituent content of L. japonicus leaves. Spectral preprocessing methods (including MSC, SG, SNV, FD, and SD) and feature screening techniques (GA, UVE, SPA) were integrated with PLSR, MLR, and KNN algorithms to achieve non-destructive prediction of seven key constituents in L. japonicus leaves. Results showed that optimal prediction models varied by constituent: PLSR with raw spectra performed best for moisture (R²v = 0.8861), KNN combined with SD preprocessing and UVE feature screening excelled for total alkaloid (R²v = 0.8068), and PLSR with MSC preprocessing and UVE screening was superior for total flavonoid (R²v = 0.8422). MLR with SD preprocessing yielded the highest accuracy for rutin (R²v = 0.7646), while PLSR with SD preprocessing achieved optimal prediction for isoquercetin (R²v = 0.7991). However, models for stachydrine (R²v = 0.6941) and leonurine (R²v = 0.5581) exhibited lower performance. All models demonstrated robust predictive capability (RPD > 2) and satisfactory external validation ( p > 0.05). This research validates the effectiveness of multispectral technology for predicting chemical constituent and supports precise planting management of L. japonicus . • Integrating multispectral technology and machine learning. • Spectral preprocessing and feature screening improve model performance. • Adopting internal and external validation to ensure model stability. • Providing technical support for the planting and harvesting of L. japonicus .