Lidar Waveform Classification Using Self-organizing Map

Most commercial LIDAR systems temporarily record the entire laser pulse echo signal, called full-waveform, as a function of time to extract the return pulses at data acquisition level in real-time; typically up to 4-5 returns. The new generation of airborne laser scanners, the full-waveform LiDAR systems, are not only able to digitize but can record the entire backscattered signal of each emitted pulse, which provides the possibility of further analyzing the waveform and, thus, obtaining additional information about the reflecting object and its geometric and physical characteristics. The LiDAR 3D point cloud and, subsequently, the derived DTM can be significantly improved by classifying the manmade objects and vegetation. Classification of the reflecting surfaces based on the statistical parameters of the backscattered waveform, however, is a non-trivial task. The suburban test area used in this study offers a great diversity of surface types, including roof, pavement, grass, and trees. In this paper, the feasibility of classifying the reflecting surface based on waveform data using an unsupervised classification method, Kohonen's Self-Organizing Map (SOM) is investigated. The correlation between the shape of the waveforms, using various statistical parameters, such as standard deviation, skewness, kurtosis and amplitude, with the properties of the reflecting surface was investigated. These four statistical parameters were used as input to the SOM classification to separate vegetation (trees and grass) and non-vegetation surfaces. The ranges calculated from the center of mass of the waveform was used to separate the non-vegetation surface into pavement and roof categories.