空載全波形光達之波形分析、特徵萃取及分類

Abstract

全波形光達為光達系統最新的發展趨勢，傳統多重回波光達僅提供三維離散點及反 射強度，而全波形光達提供完整的回波訊號。比較兩者，多重回波光達資料提供幾何特 徵及單點反射強度，而全波形光達提供完整的幾何及回波特特徵，因此全波形光達有助 於地表重建及地物判識。本計畫以三年為期進行全波形光達之波形分析、特徵萃取及地 物分類，全程研究重點為利用回波訊號進行資訊萃取並提升地形及地物的分辨能力。第 一年目標為建立一維波形分析的演算法，研究重點是以不同數學模式回波訊號進行分 解，以取得點位座標及波形參數，其中波形參數包含振輻、波寬、形狀因子等。第二年 目標為建立三維波形特徵萃取的演算模式及地形分類，研究重點是整合時間序列之多重 完整回波，考量波形之間空間的關聯性，完成波形特徵萃取，其中萃取的特徵包含紋理 特徵、高程變化特徵、法向量角度變化特徵等，最後進行地面及非地面之分類。第三年 目標為整合全波形光達及同步獲取之光學影像進行地物分類，將波形分析及特徵萃取的 成果做為分類的圖層，以隨機樹分類器(Random Forest Classifier)整合光達及光譜圖層進 行分類，研究重點為分析全波形特徵對分類的幫助，並分析各特徵的效益。

Full-waveform (FWF) lidar is an advanced technology in the development of lidar system. Traditionally, multi-echo (ME) lidar only provides 3-D point clouds and intensity; but FWF lidar provides the entire returned signals. ME lidar only provides geometric property but FWF provides geometric and waveform properties. Hence, FWF lidar is able to improve the surface reconstruction as well as landcover classification. This three-year project plans to establish the procedure of waveform analysis, feature extraction and classification for airborne full waveform lidar data. The core technique will be used to improve the capability of data interpretation of FWF lidar. The works in the first year will deal with 1-D signal analysis of waveform. This study will compare different mathematic models in waveform decomposition. The works in the second year will include 3-D waveform feature extraction, analyzing the spatial relationship between waveform by combining the sequential waves. The extracted features are texture, height difference, normal angle difference and others. The author will select the ground points using the extracted features and geometric properties. The works in the third year will focus on the combination of FWF lidar and multispectral image for landcover classification. The waveform features and multispectral information are combined in Random Forest Classifier (RFC). RFC is used to separate different types of land covers and also to evaluate the importance of features.