CLASSIFICATION OF LIDAR DATA BASED ON MULTI-CLASS SVM F. Samadzadegan a , B. Bigdeli a , P. Ramzi a, * a Department of Geomatics Engineering, Faculty of Engineering, University of Tehran, Kargar-shomali Avenue, Tehran, Iran - (samadz,bigdeli,pramzi)@ut.ac.ir Commission VI, WG VI/4 KEY WORDS: LIDAR data, Classification, Classifier Fusion, Multi-class SVM, Genetic Algorithm, Parameter Optimization ABSTRACT: LIght Detection And Ranging (LIDAR) is a powerful remote sensing technology in the acquisition of the terrain surface information for object classification and extraction. Major benefits of this technique are its high level of automation during data capturing and its spatial resolution. Because of high complexities and difficulties in urban areas, many researchers focus on the using of LIDAR data in such area. Consequently, one of the challenging issues about LIDAR data is classification of these data in urban area for identification of different objects such as building, road and tree. Several urban classification methods have been proposed for classification of LIDAR data. Support Vector Machines (SVM), one of the new techniques for pattern classification; have been widely used in many application areas such as remote sensing. SVM is a binary classification method but in some researches like remote sensing or pattern recognition, we need more than two classes. One solution for this difficulty is to split the problem into a set of binary classification before combining them. Multi-class SVM is one solution for solving mentioned problem. The one- against-one and the one-against-all are the two most popular strategies for Multi-class SVM. One problem that faces the user of an SVM is how to choose a kernel and the specific parameters for that kernel. Applications of an SVM therefore require a search for the optimum settings for a particular problem. This paper proposes a classification technique, which we call the Genetic Algorithm Multi-Class SVM (GASVM), that uses genetic algorithm as a method for kernel‘s parameter optimization for one of the Multi-class SVM classifiers. We have used genetic algorithm for optimizing γ and C parameters of RBF kernel in Multi-class SVM. The classification‘s results of LIDAR data by use of this presented technique clearly demonstrate higher classification accuracy. * Corresponding author. This is useful to know for communication with the appropriate person in cases with more than one author. 1. INTRODUCTION Remotely sensed data has been widely used to land cover classification and object extraction (Wehr, Lohr, 1999; Haitao, 2008). Light Detection And Ranging (LIDAR) is one of the recent remote sensing technologies that is widely used for Digital Terrain Model (DTM) data collection and also for other studies including 3D extraction, urban management, atmospheric studies, and so on (Clode, 2004; Alharthy, Bethel, 2003). Comparing to other remote sensing data sources, LIDAR has its advantages such as acquisition of very dense data in a short period of time. LIDAR data contains plenty of scene information, from which most ground features such as roads, buildings and trees are discernible. More recently, advancements in LIDAR enabled the acquisition of dense point clouds. Major benefits of this technique are its high level of automation during data capturing and its spatial resolution. With point densities of up to several points per square meter, LIDAR data has become a valuable additional source for the reconstruction of different urban objects (Wehr, Lohr, 1999). Classification of LIDAR data into objects such as building, tree and road in complex area is a challenging research topic in pattern recognition and remote sensing studies (Bartels, Wei, 2006; Brzank, Heipke, 2007). Several urban classification methods have been proposed for classification of LIDAR data (Kraus, Pfeifer, 1998; Zhang, 2003). The Support Vector Machine (SVM) has emerged in recent years as a popular approach to the classification of data. (SVM) were first suggested by Vapnik (1995) and have recently been used in a range of problems including pattern recognition (Pontil and Verri, 1998), bioinformatics (Yu, Ostrouchov, Geist, & Samatova, 1999), and text categorization (Joachims, 2000). SVM by itself is a binary classification but in some researches like remote sensing or pattern recognition, we usually have more than two classes. Multi-class SVM is the solution for this problem which is has been utilized in some researches (Wetson, Watkins, 1998; Naotosi, 2007). When using SVM, one problem is confronted: how to set the best kernel parameters. Proper parameters setting can improve the SVM classification accuracy. A GA-based regularization parameter can also be optimized using GAs in (Frohlich and Chapelle,2003). The parameters that should be optimized include penalty parameter C and the kernel function parameters such as the γ for the radial basis function (RBF) kernel. Huang and Wang used Genetic algorithm as a method for parameter optimization of Support Vector Machine (Huang, Wang, 2006). The objective of this research is to simultaneously optimize the parameters and feature subset without degrading the SVM