二次微分法於空載全波形光達之高斯波形擬合與地物分類

Second Derivative for Gaussian Waveform Fitting and Land Cover Classification with Airborne Full-waveform LiDAR

盧佑樺(Yu-Hua Lu) ； 蔡富安(Fu-an Tsai) ； 賴哲儇(Jhe-Syuan Lai)

《航測及遙測學刊》 21卷1期 (2016/08) Pp. 13-30

https://doi.org/10.6574/JPRS.2016.21(1).2

摘要

全波形光達(Full Waveform LiDAR)可完整記錄每條雷射光束在不同時間回傳的反射能量，若能藉由這些記錄下來的波形衍生出額外資訊，可望增進地形三維重建及地物分類的成效。波形擬合(fitting)與特徵萃取(feature extraction)是處理和分析全波形光達資料的重要過程。本研究利用高斯函數擬合空載全波形光達資料之波形，並以二次微分法搜尋迭代計算之初始值。基於波形擬合成果，可以衍生出振幅(amplitude)、波寬(width)、背向散射截面(backscatter cross-section)等波形參數，且配合強度(intensity)、正規化高程(normalized height)等傳統光達特徵與同步搭載之中像幅正射影像，作為地物分類之依據。另外，本研究亦比較簡易貝氏(naïve Bayesian)與隨機森林(random forests)等兩種分類器之成效。研究結果顯示，運用二次微分法決定高斯函數迭代計算之初始值，並搭配隨機森林分類器，能提供較佳的擬合及分類成果，且波形參數有助於植物類別的辨識。

關鍵字

全波形光達 ； 波形擬合 ； 地物分類 ； 二次微分法 ； 隨機森林分類器

並列摘要

Full-waveform LiDAR is an emerging active data acquisition tool in many applications. In addition to accurate positions of point clouds, full-waveform LiDAR provides complete wave signals of returned laser ray, which can be subsequently used to derive important characteristics of the targets. The waveform parameter extraction and analysis are two important operations for full-waveform LiDAR applications. In this research, the Gaussian modeling function with second derivative method was utilized for waveform fitting. Features extracted from the waveform, including width, amplitude, and backscatter cross-section, in conjunction with traditional LiDAR features, such as normalized height and intensity, and greenness index from image were used as primary attributes for land cover classification. Two classifiers were used and compared in this study, including Naïve Bayes and Random Forests. Experimental results demonstrate that using the second derivative method can produce higher fitting success rate and better classification results. The land cover classification results indicate that full-waveform features are helpful for distinguishing different vegetation targets. In addition, the decision-tree-based Random Forests classifier is more suitable for land cover classification of LiDAR data used in this study.

並列關鍵字

full-waveform LiDAR, waveform fitting ； land cover classification ； second derivative ； random forests