A Knowledge-based Approach to Urban Feature Classification Using Aerial Imagery with Lidar Data

While the spatial resolution of remotely sensed data has improved, multispectral imagery is still not sufficient for urban classification. Problems include the difficulty in discriminating between trees and grass, the misclassification of buildings due to diverse roof compositions and shadow effects, and the misclassification of cars on roads. Recently lidar (light detection and ranging) data have been integrated with remotely sensed data to obtain better classification results. In this study, we first conducted maximum likelihood classification (MLC) experiments, a traditional pixel-based classification method, to identify features suitable for urban classification using lidar data and aerial imagery. The addition of lidar height data improved the overall accuracy by up to 28 and 18 percent, respectively, compared to cases with only red-green-blue (RGB) and multispectral imagery. To further improve classification, we propose a knowledge-based classification system (KBCS) that includes a three-level height, "asphalt road vegetation, and non-vegetation" (A-V-N) classification rule-based scheme and knowledge-based correction (KBC). The proposed KBCS improved overall accuracy by 12 and 7 percent compared to maximum likelihood and object-based classification, respectively.