Author : Stewart He
Publisher :
ISBN 13 : 9781369616828
Total Pages : pages
Book Rating : 4.6/5 (168 download)
Book Synopsis Features for Learning and Visualizations of Forest LiDAR Data by : Stewart He
Download or read book Features for Learning and Visualizations of Forest LiDAR Data written by Stewart He and published by . This book was released on 2016 with total page pages. Available in PDF, EPUB and Kindle. Book excerpt: Climate change is one of the biggest political, social, and scientific issues facing the global community today with the health of the world's forests playing a critical role. In order to monitor the status of forests around the world environment scientists often rely on two specialized technologies: LiDAR(LIght Detecting And Ranging) and hyperspectral imagery. LiDAR of various resolutions enable researchers to scan hundreds of square kilometers at a time creating massive and detailed point clouds. At the same time hyperspectral imagery captures how light is reflect off the same surfaces across several hundred light frequency bands. Visualizing and analysing these datasets has continuously challenged scientists in both forestry and computer science. In this dissertation, we continue this effort to understand and utilize this data. Visualizing point clouds of forest proves to be challenging due to the unique nature of the problem. Forests are porous, thus scanning a forest produces little to no surfaces or useful normals which many point cloud visualization techniques rely upon. Instead we present a method of rendering silhouettes around points providing very useful depth cues without relying on normals. We then implement this technique to run web browsers like Firefox of Chrome to reduce setup time and on Android devices using Google Cardboard to provide stereoscopic 3D. There is a wide range of useful properties that can be extracted from LiDAR and hyperspectral imagery. In this dissertation we focus on identifying species at tree level using shape features combined with hyperspectral features. We present two different sets of shape features. One uses a binning approach that we refer to as histogram shape descriptors. The other uses a set of complex basis functions called Zernike polynomials. We use these polynomials to generate a set of Zernike moments that are used as features. We compare how these shape descriptors work in a variety of tests meant to represent a range of species and forest types.
