A Two-Phase Classification of Urban Vegetation Using Airborne LiDAR Data and Aerial Photography

Abstract

The nDSM is commonly used in land cover classification with different heights. However, it is difficult to classify vegetation accurately from other land covers, such as buildings, with only the height information. Meanwhile, traditional remotely sensed imageries have difficulties discriminating different urban vegetation components such as trees and shrubs. Therefore, it is essential to combine both remote sensing imagery and height information obtained from Light Detection and Ranging (LiDAR) data for classification of detailed vegetation components. In this paper, a two-phase classification method is proposed to fuse the airborne LiDAR data and aerial photography imagery to obtain detailed urban vegetation classification map. The first step is to distinguish vegetation from buildings, bare ground, and shade. In this step, two different fusion approaches and two classification methods were used, and the result with the highest accuracy for vegetation classification was selected as the input of the second step. The second step is to output the classification map of vegetation class into vector polygons and utilizes them to separate the vegetation LiDAR points from the nonground points. Then tree, shrub, and lawn points can be easily classified from the vegetation points due to their different heights. The proposed method yielded a classification result with an overall accuracy of 83.39% and a kappa coefficient of 0.79. Moreover, the producer accuracies of vegetation class (tree, shrub, and lawn) are 95.20%, 61.66%, and 79.35%, respectively.