Wood Recognition by Visible/Near Infrared Spectroscopy Based on Multivariate Empirical Mode Decomposition

A fast and accurate wood identification method based on visible/near infrared spectroscopy was explored. Taking 8 kinds of imported wood as the research object, multivariate empirical mode decomposition（MEMD） and maximum mutual information coefficient（MIC） were used to decompose, screen and reconstruct the collected spectral data, and then the continuous projection method（SPA） was used to extract the feature bands, and combined with XGBoost classifier for classification and recognition. In order to further verify the feasibility of the proposed method, the wood recognition method was compared with empirical mode decomposition（EMD） algorithm, traditional support vector machine（SVM）, K-nearest neighbor classification algorithm（KNN） and BP Neural network（Back Propagation Neural Network） classifier. The results showed that MEMD method was better than EMD method for visible/near infrared spectrum denoising. Compared with 90% of MEMD-SPA-SVM, 88% of MEMD-SPA-KNN and 89.2% of MEMD-SPA-BP, the average recognition accuracy of MEMD-SPA-XGBoost reached 96.5%. It can be seen that this method has a good application prospect in wood identification method.