Urban Forest Tree Classification Using UAV-Based High-Resolution Imagery

Unmanned aerial vehicle (UAV) remote sensing has a high potential for vegetation monitoring in complex urban landscapes. Acquiring information about tree species composition is needed for urban forest management but the field survey of these areas is time-consuming and costly. The goal of this research was to explore the ability of UAV-based RGB imagery for species classification using RGB-based vegetation indices and linear discriminant analysis. Five distinct species including two conifers and three broadleaves were selected in the study area, and the LDA algorithm was applied on raw bands, vegetation indices, and band ratios. The results show a higher accuracy for classification of conifer trees (especially Cupressus arizonica with user’s accuracy of 0.85) rather than broadleaf species. The highest model accuracy was obtained mainly based on the red band, and the overall accuracy for LDA classification was 0.69.