不同地形因子對於高光譜影像輻射改正之影響探討

Abstract

高光譜影像含有較細緻且豐富的光譜資訊，不同物質擁有其獨特的光譜反射曲線，由於受到大氣及地形的影響，光譜影像需事先進行輻射校正，才能因應各種實際需求，但大多數研究仍致力於去除大氣效應，並提出改正模型，如常見的大氣改正模型：MODTRAN、FASCODE、FLAASH、QUAC等，而對於地形效應的影響予以簡化甚至忽略，即便有使用地形改正模型，通常會簡化參數，將入射平面視為蘭伯特平面，僅利用一個地形傾斜角進行簡單的地形改正，雖然也有研究嘗試將反射平面視作非蘭伯特平面，但卻鮮少同時考量地形資料的影響。過去有關地形效應輻射改正的研究大都使用DEM資料計算地形起伏變化，隨著影像空間解析度提升，已無法忽略地表的建物、植被，若能採用DSM資料，將更能呈現改正時的真實地貌，此外，在資料層面上無論是DEM或DSM資料皆已經過人為處理，屬於二手資料，並無法保有最原始的地形資訊，故將另外引入一手資料 — 光達點雲。 本研究針對台北市山區之高光譜影像，先透過QUAC方法進行大氣改正，所採用的地形改正方法包括 Cosine Correction、Minnaert、C Normalization、及Modified Minnaert等四種改正模型，另一方面引入點雲資料與不同空間解析度的DSM，最終分析不同地形資料對高光譜影像地形效應改正的影響。

The imaging spectrometers have the ability to acquire the hyperspectral images with rich and subtle spectral information of earth objects. However, the recorded spectral radiance in hyperspectral image is very subject to the variance of circumstance. In order to utilize this data in practical applications, an accurate radiometric correction of the hyperspectral image has to be performed in advance. The correction of radiometry includes two essential parts, one is atmospheric correction and the other one is topographic correction. Atmospheric correction technique removes spectral atmospheric transmission and scattered path radiance; topographic correction technique removes the effect of topography because of rugged surface areas. Many atmospheric correction models such as MODTRAN, FASCODE, FLAASH, and QUAC are maturing and have been applied in many fields. In contrast, there are many topographic correction models to deal with radiometric correction, however, most of them generally simplify parameters by using cosine correction, which regards the surface as an ideal diffusely reflecting plane (i.e. Lambertian plane). In reality, a simplified model is hard to correct the effect of topography accurately. Although there are a lot of non-Lambertian plane models were proposed, they seldom consider the influence of the types of topographic data. In the past, a small number of studies used DEM data as topographic data. With the increase of image space resolution, it can not ignore the surface of the building, vegetation. Therefore, using DSM data will be more appropriate choice to show the real landscape. In addition, both the DEM and DSM data are second-hand data, and can not maintain the primitive terrain information; hence this research introduces the first-hand data - Lidar point clouds. The objective of this research is to analyze the influence of different type of topographic data such as the different resolution grid DSM and discrete point clouds. Firstly, the Quick Atmospheric Correction are performed on the hyperspectral image, and then the parameters required by topographic correction model are obtained and improve the quality of topographic correction of the hyperspectral images. Accordingly, this study discusses and analyzes what kind of topographic correction model is suitable for the topographic data and correct the topography effect of the real high resolution hyperspectral images.