Airborne Light Detection and Ranging (LiDAR) is a sensor that can generate terrain elevation and intensity data of very large areas with high precision and dense resolution. The intensity and elevation are co-registered, which eliminates the need for tedious registration after the fact. This combined LiDAR data can be used to classify different topographic features. Given the enormous amount of such data generated all over the world, total automation in these classification processes in a batch process is highly desired if not a critical need.;In this dissertation, a novel LiDAR-based automated standing waterbody extraction (LASWE) algorithm is presented. The special characteristics of water bodies that helped with the development of LASWE are: a) the essentially flat surface of water bodies and b) the surface elevation of water bodies is lower than that of its surrounding areas. In addition, LiDAR intensity return from the water surface has special characteristics which are: a) the specular reflection from smooth water surface and b) the spectral reflectance of water is low compared with vegetation and other topographic features.;The LASWE algorithm employs a novel histogram analysis method for segmentation of flat areas and then an SVM classifier to eliminate false detections. An intensity-based classifier was also used to remove other false detections that were not eliminated by the SVM classifier. An iterative pixel-based maximum likelihood classification (MLC) technique was employed to fine-tune water surface detection at the land edge of water bodies. Three LiDAR datasets from different geographical locations were split into training and testing sets for validating our LASWE algorithm. The classification accuracy of the algorithm was measured by calculating the overall accuracy and the standard Cohen's kappa coefficient. The LASWE algorithm was found to give classification accuracy greater than 97.92%.;Speed of computation is of essence in all classification problems. The multiresolution LASWE (MLASWE) algorithm presented in this dissertation is an upgraded version of LASWE, which was designed to detect water surface significantly faster than the LASWE algorithm without compromising the classification accuracy. The MLASWE algorithm detects water surface with a coarse resolution first and then with a intermediate resolution and finally, with a fine resolution in addition to using efficient buffering techniques.;In summary, the MLASWE algorithm presented in this dissertation is a robust, raster-based method that detects water surfaces at high speed, in a fully automated mode and with high accuracy.
展开▼
