Precise identification of individual tree species in urban areas with high canopy density by multi-sensor UAV data in two seasons

Unmanned aerial vehicle (UAV) platforms with multi-sensors can provide data with high resolution for tree species classification. However, few studies have explored the fusion of multi-sensor UAV data in two seasons for urban tree species classification with high canopy density. Therefore, in this paper, UAV hyperspectral, LiDAR data, and visible images in two seasons were used to explore their performance in urban tree species classification with the following steps: (1) extraction of spectral, structural and texture features; (2) classification of 14 urban tree species using the random forest classifier based on different feature combinations; (3) classification of individual tree species in high canopy density urban areas using the proposed hybrid pixel- and object-based classification method; (4) investigation of the seasonal effects on urban tree species classification with multi-source UAV data; and (5) accuracy assessment. By comparative experiments, the results demonstrate that the fusion of spectral and three-dimensional spatial information, the proposed hybrid method, and the integration of two-season datasets significantly improve the overall accuracy of urban tree species classification. The advantages and imitations of multi-sensor UAV data fusion are also discussed. This study is expected to provide a new method for precise census and refined management of urban forest resources.