打印

作者

Xiaofeng Wang;Yi Wang;Chaowei Zhou;Lichang Yin;Xiaoming Feng

来源

URBAN FORESTRY & URBAN GREENING,Vol.58,Issue1,Article 126958

语言

英文

关键字

Aerial photogrammetry;Multiple features;Random forest classification;Single tree segmentation;Tree height;UAV

作者单位

School of Land Engineering, Chang’an University, Xi’an, 710064, China;The Key Laboratory of Shaanxi Land Consolidation Project, Chang’an University, Xi’an, 710064, China;Key Laboratory of Land Surface Pattern and Simulation, Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing, 100101, China;State Key Laboratory of Urban and Regional Ecology, Research Center for Eco-Environmental Sciences, Chinese Academy of Sciences, Beijing, 100085, China;School of Land Engineering, Chang’an University, Xi’an, 710064, China;The Key Laboratory of Shaanxi Land Consolidation Project, Chang’an University, Xi’an, 710064, China;Key Laboratory of Land Surface Pattern and Simulation, Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing, 100101, China;State Key Laboratory of Urban and Regional Ecology, Research Center for Eco-Environmental Sciences, Chinese Academy of Sciences, Beijing, 100085, China

摘要

Fine monitoring of tree species is essential to supporting the urban forest management. Data acquired from unmanned aerial vehicles (UAVs) not only have very high spatiotemporal resolution, but also contain the vertical structure of trees which is important in the fine recognition of vegetation types. However, the research of combining multi-dimensional features in classification is still very limited. In our study, we extracted the spectral information, vegetation morphological parameters, texture information, and vegetation indexes based on UAV ultrahigh resolution images to build an object-oriented-based random forest (RF) classifier at the single tree scale. Establishing 6 classification scenarios that combines multiple data sources, multi-dimensional features, and multiple classification algorithms, our results show that: (1) UAV images can effectively detect surface fragments. The accuracy of RF classification based on UAV multiple features was high at 91.3 %, which was 20.5 % higher than the results by using high-resolution Baidu maps; (2) for mapping the tree species of urban forest, tree morphological characteristics, texture information, and vegetation indexes improved the classification accuracy by 2.9 %, 1.9 %, and 7.1 %, respectively, resulting in meaningful improvement of classification effects; and (3) the accuracy of RF classification based on UAV data was much higher than the maximum likelihood classification (MLC) results. Compared with the latter, the former can effectively avoid salt and pepper noise. The workflow of information extraction and urban forest classification based on UAV images in this paper yields high performance, which has important significance as a reference for future relevant research.