Rapid origin discrimination of Meconopsis integrifolia coupled with machine learning techniques and multi-spectral data fusion strategies

To solve the origin traceability conundrum of an important industrial ornamental plants Meconopsis integrifolia and achieve a rapid and accurate origin identification, a discrimination model was devisedby integrating Near Infrared (NIR) and Mid-Infrared (MIR) spectroscopy techniques for both individual spectral analysis and multi-spectral data fusion strategies. In the context of single-spectral modalities, neither the optimized NIR nor MIR model was capable of accurately discriminating the origin. Using Data Fusion (DF) strategies based on the Machine Learning (ML) techniques, our study connected NIR and MIR spectra, and thus constructed Support Vector Machine (SVM) model. The results show that, in low-level fusion (LDF), SVM model manifests a recognition rate of 100.00 %, a prediction rate of 89.00 %, and an overall accuracy of 96.51 %. Medium-level fusion (MDF) scenario further improves the model’s prediction rate, and achieves a recognition rate of 99.80 %, a prediction rate of 93.00 %, and an overall accuracy of 97.62 %. In high-level fusion (HDF), the recognition rate peaked at 100.00 %, with the prediction rate of 94.00 %, the overall accuracy 98.10 %, and the external validation set’s accuracy rate 84.38 %. The results have distinctively demonstrated that data fusion can enhance model’s accuracy, and with the increase of fusion level, the model’s discriminative performance can be effectively improved. This study elucidates that the amalgamation of NIR and MIR spectroscopy techniques with DF strategies represents an effective, non-invasive, and practicable approach for the origin identification of M. integrifolia , furnishing invaluable technical support for flower industry and landscape engineering research and applications. • Models of NIR and MIR spectroscopy techniques for the origin traceability of M. integrifolia . • SVM Models based on DF strategies to connect NIR and MIR spectra by Machine learning techniques. • DF of SVM is an effective, non-destructive and practical method for the origin identification of M. integrifolia. • HDF of SVM outperforms MIR and NIR with 98.10 % overall accuracy in geographical authentication of M. integrifolia.