Abstract：Based on support vector machine and Mahalanobis distance, the ability of mid-infrared spectrum analysis to identify imported rosewood, windmill wood, micro ebony, fuel rosewood and east African rosewood was explored. Five hundred group of test samples were collected and analyzed by the mid-infrared spectrometer, and the test data were preprocessed. Firstly, in order to ensure the validity of the samples, the abnormal spectra were diagnosed. Based on Wright’s test, two groups of abnormalities were found in rosewood and micro ebony, one group of abnormalities was found in windmill wood, fuel rosewood and east African rosewood respectively. In order to unify the sample size, five species of trees were excluded from the five sets of data, including the abnormal spectrum. Secondly, the research of tree species recognition in near-infrared spectroscopy was analyzed. The results showed that the first derivative processing of spectral data could improve the recognition accuracy. Therefore, the mid-infrared spectroscopy data were smoothed and first derivative processing. The eigenvalues of the spectral data were extracted by principal component analysis. The scatter plots of the first and second principal component scores of the test set showed that the clustering of the smoothed plus first derivative processed test set was smooth. Based on the scores of principal components, the recognition research was based on support vector machine and Mahalanobis distance. Considering the selection of the number of principal components in the recognition method would directly affect the accuracy of recognition, and usually, the selection of principal components only referred to the cumulative contribution rate. In order to make the selection of principal components more scientific, in the support vector machine identification method, the particle swarm optimization algorithm was used for parameter optimization, the relationship between the number of principal components (range [5, 30]) and the best discrimination accuracy under the 50-fold test was tested. The results showed that the optimal discriminating accuracy of the number of principal components in the range of [7, 11] of smoothing processing and smoothing plus first-order derivative processing was relatively high, and the optimal number of principal components was determined as 8 based on the corresponding discriminating accuracy. The first eight principal components were used as input variables, and the test set was tested based on support vector machine and Mahalanobis distance. The results showed that the correct recognition rates of the two recognition methods were higher, and the recognition rate of support vector machines was slightly higher than that of Mahalanobis distance. The recognition rate of smooth distance plus first-order derivative processing was better than that of smoothing processing. The correct recognition rate of support vector machine with smooth plus first-order derivative processing reached 98%, and the recognition effect was the best. Therefore, the mid-infrared spectrum can be used as an effective means to identify timber species.
Key words：Mid-infrared spectrum; Tree species identification; First derivative; Principal component analysis; Support vector machine; Mahalanobis distance
