Classification of typical tree species in a karst area of Guizhou Province based on principal component analysis and support vector machine

• Published in: Spectroscopy letters 2021-04, Vol.54 (4), p.305-315
• Main Authors: Wang, Zhi-Jie, Liu, Shu-Jun, Yang, Qing-Qing, Peng, Hai-Lan
• Format: Article
• Language: English
• Subjects: Classification of tree species; hyperspectral; Kernel functions; Model accuracy; Nature reserves; Polynomials; principal component analysis; Principal components analysis; Radial basis function; Rainforests; Species classification; support vector machine; Support vector machines
• ISSN: 0038-7010; 1532-2289
• DOI: 10.1080/00387010.2021.1931790

The accurate identification of forest tree species is the basis for the utilization and protection of forest resources. In this study, we collected hyperspectral data on 3287 leaves from seven typical tree species in the Caohai Nature Reserve and Huaxi District, Guizhou Province, China. The Savitzky-Golay smoothing, baseline, normalize, standardized normal variable, and moving average methods were used to preprocess the original spectral reference after removing the abnormal values, and then the dimension was reduced by principal component analysis. Finally, the reduced data were classified by support vector machine with the linear, polynomial, radial basis function, and Sigmoid kernel functions. The results show the following: (1) the principal component analysis + support vector machine method is feasible for tree species identification, however, the recognition results of different preprocessing methods, different combinations of principal components, and different support vector machine classification methods are quite different; (2) the best combination is 20 principal components + the normalize + linear support vector machine model, achieving a classification overall accuracy of 98.97%.