Comparison and Assessment of Different Classification Methods Based on Object Based Analysis Using LiDAR Data and Optical Imagery in Urban Area

In recent years urban classification becomes very important caused by urban growth and high rate of urbanization. Classification and recognition of urban classes in different information layers for supplementation and updating of urban database is considered by researchers and managers. The goal of this paper is comparison and evaluation of different urban classification methods base on object based analysis by using LIDAR data and optical imagery. This paper includes three main phases. First step of workflow is co registration and preprocessing of LIDAR data and high resolution imagery to prepare multi source data for urban classification. Second step followed by hierarchal multi resolution segmentation at different scales to exhibit different features which are consist of building, roads, vegetation area and vehicles. Segmentation contains three main levels. Selection of hierarchal segmentation parameters is a try and error task and segmentation validation is done by visual assessment. After object production convenient features should be introduced to the classification algorithms. Finally thresholding, nearest neighbor and fuzzy nearest neighbor classification at each level of hierarchy was performed. Last step is result assessment and interpretation. By result evaluation, nearest neighbor classification with 0.99 over all accuracy was nominated as best classifier in first level. In second level of hierarchy nearest neighbor classification with 0.985 shows the highest overall accuracy. In third level fuzzy nearest neighbor classification and thresholding show 0.841 over all accuracy.