Aerial Lidar Data Classification Using Weighted Support Vector Machines

This paper presents our research on classifying scattered 3D aerial Lidar height data into ground, vegetable (trees) and man-made object (buildings) using Support Vector Machine algorithm. To this end, the most basic theory of SVM is first outlined and with concern to the fact that features are differed in their contribution to classification, Weighted Support Vector Machine (W-SVM) technique is proposed. Second, four features consist of height, height variation, plane fitting error and Lidar return intensity are identified for classification purposes. In this step, features are normalized respectively and their weight that indicates feature's contribution to certain class or multi-class as a whole are calculated and specified. Third, Based on W-SVM technique, one 1AAA1 solution to multi-class classification is proposed by integration "one against one" and "one against all" solution together. Finally, the classification results of LIDAR data with presented technique clearly demonstrate higher classification accuracy and valuable conclusions are given as well.