ANALYSIS OF THE PERFORMANCE OF TEXTURE FEATURES IN TREE SPECIES CLASSIFICATION OCCURRENCE-BASED EXTRACTION APPROACH

The texture feature of images plays an important role in tree species classification. The selection of important texture features in tree species identification can be supported via performance testing by considering the same texture features produced by different bands and different texture features produced by the same band in tree species classification. In this study, we used the 8-band WorldView-2 image as the data source and extracted three types of texture features, i.e., data range, mean, and variance, by applying occurrence measures to the obtained data, then used the maximum likelihood classifier for image classification. The results show that the classification accuracies of the texture features extracted from the red edge, near-infrared 1, near-infrared 2 bands (55.0768%, 53.7395%, and 51.1020%, respectively) were higher than those of other bands (ranging from 39.5740% to 43.9079%); the texture feature classification accuracy of the mean (77.9470%) was higher than those of the data range (45.3938%) and variance (45.3938%); when the eight bands combined with all mean and all data range, the classification accuracy of tree species reached 89.7845%. The results demonstrate that the effective combination of spectral bands and important texture features can improve the tree species classification results.