ObjectiveBased on the classification of airborne hyperspectral imagery, the method of combining different scale texture features with object-oriented classification was less applied to the classification of tree species, and the related research was mainly for single tree species identification without considering multiple tree species, therefore, the ability to identify tree species in complex forests needs further study. However, the study in such research was scanty, so this study intended to explore the application of different scale texture features and object-oriented classification techniques in the fine classification of tree species.

MethodWe used airborne hyperspectral data with object-oriented classification to classify tree species. According to the type of land cover in the study area, we used stratified classification method to distinguish non-forest land, other forest land and forested land, and finely classify tree species in forest land. Feature variables were extracted from airborne hyperspectral images, including independent component analysis (ICA) transformation images and spatial texture features, analyzed the spectral reflectance and the suitable texture scale of each tree species. The tree species were classified according to different scale texture features, and compared tree species classification results of different features by support vector machine (SVM).

ResultConcerning the object-oriented tree species classification combined with texture features, the overall accuracy of single-scale texture features was 87.11%, and the Kappa coefficient was 0.846. Combined with different scale texture features, the overall accuracy was 89.13%, and the Kappa coefficient was 0.87. Compared with the accuracy based on spectral features, the classification accuracy was improved by 4.03% and 6.05%, respectively. It showed that in object-oriented classification, the addition of texture features had a significant effect on improving the accuracy of tree species classification. The classification accuracy of tree species combined with different scale texture features was higher than single scale texture features, especially in the classification of other broadleaved tree species and Masson pine (Pinus massoniana), the producer accuracy was higher than single texture scale by 5.48% and 6.12%, respectively.

ConclusionDifferent scale texture feature is more advantageous than single-scale texture feature, which improves the contribution of texture features in tree species classification. The spectrum of integrated airborne hyperspectral imagery and object-oriented classification of different scale texture features make the tree species identification more precise and accurate. It is effective for tree species classification in complex forest, and meets the application requirements in fine tree species identification based on airborne hyperspectral images.