Özet:

Spatial data produced with airborne Lidar(Light Detection and Ranging) systems are obtained with high accuracy, fast and low cost. However, manual processing of the data for object extraction is time consuming and labor intensive. Supervised/unsupervised classification methods can be used to make this process automatic. Classification of Lidar data as ground and non-ground data is called filtering. Filtering is very important in creating a Digital Elevation Model using Lidar data. In this study, the discrete-return Lidar test data obtained from the flight at 1200 meters altitude in Bergama district with the Riegl LMS-Q1560 Lidar system produced in 2014 under the chairmanship of the General Directorate of Mapping was used. The Lidar point cloud was grouped into clusters by analyzing it with the Self Organizing Maps (SOM), which is an unsupervised artificial neural network method. Feature classes were determined by comparing clusters with satellite images. The accuracy of the feature classes obtained by this method was calculated by examining all points of the classes which were visually determined. The minimum number of neurons of neural network was determined according to the error values. As a result of filtering the Lidar point cloud with SOM method, Type-1 error was found as 11.54%, Type-2 error was 19.43% and total error was 16.41%. In accordance with the results obtained, it was seen that SOM neural networks with the number of neurons determined could be used effectively in filtering the airborne Lidar data.

Anahtar Kelimeler: