| 摘要: | 由於感測器(sensor)中各元件(detector)的製造及時間的光學衰減等微小變化與差異，使得在拍攝均勻輻射光源時各頻道中的每個感測器元件會有不相同的反應而造成在影像中所呈現出的條紋現象。影像中的條紋須在應用前進行合理地校正，以避免誤差之產生，此類校正稱為相對輻射校正。經驗模態分解法(Empirical Mode Decomposition, EMD)是希爾伯特-黃轉換(Hilbert-Huang Transform, HHT)中的基礎部份，複雜的訊號資料在經過使用經驗模態分解法後，皆會分解出有限數量的本質模態函數(Intrinsic Mode Functions, IMF)。分解出的本質模態函數為感測器(sensor)各單位元件(detector)原始訊號間在時間軸和空間軸相對差異所構成的特性分布訊號，可藉此分析感測器中各元件之差異(條紋現象)，以進行正規化的修正。因此本研究嘗試應用HHT於福爾摩沙衛星五號(福衛五號)光學酬載遙測照相儀測試影像的相對輻射校正，這也是國際上首次的測試與應用。
福衛五號主要光學酬載RSI可提供2公尺地面解析之全色態影像和4公尺多光譜影像，包含紅光、綠光、藍光以及近紅外，並具有不同的電子增益值(G1, G2, G4)選項功能，可觀測特定地區的細微變化。研究首先使用經驗模態分解法校正影像中時間軸的差異，其後再對空間軸的差異進行校正。本研究使用經驗模態分析法得到各個本質模態函數所對應的校正係數，並透過福衛五號輻射轉換方程求出感測器各元件之等化係數(relative response of each detector)，對原始影像進行相對輻射校正。校正結果以飛行模式增益值為G2的結果為例，校正後之影像標準差有明顯的降低(全色態: 66.31 - 1.85, 紅光: 54.19 - 1.90, 綠光: 36.50 - 1.49, 藍光: 32.43 - 1.56, 近紅外: 37.67 - 1.20)，並透過工程體實拍影像的校正，模擬建構升空後相對輻射校正流程，因此本研究所提出之經驗模態分解法對於福衛五號遙測照相儀之相對輻射校正的方法具高度的可行性。
;Due to minute variations in manufacturing and temporal degradation, every detector in each band may exhibit a different behavior when exposed to uniform radiance, causing a noticeable striping artifact in recorded image. The empirical mode decomposition (EMD) method is the fundamental part of the Hilbert-Huang Transform. With EMD method, the complicated data set can be decomposed imagery into a finite number of components which are described as intrinsic mode functions (IMF). These IMFs can form a complete and nearly orthogonal basis for the original signal, thus the relative discrepancy between detectors can be corrected in both time and space domains accordingly. In this study, the EMD method is applied to calibrate FORMOSAT-5 (FS-5) pre-flight data for the relative radiometric calibration of RSI which is the international debut test and application of EMD method on the relative radiometric calibration.
The primary payload of FS-5 is an optical Remote Sensing Instrument (RSI), which can provide 2-meter high-spatial-resolution imagery with one panchromatic and four multi-spectral bands (blue, green, red, NIR) with different electronic gain settings (G1, G2 and G4). First of all, we used the EMD method to analyze the variation in the time domain then the space domain between each detector will be calibrated. In the EMD method analyzing, we get the correction coefficient for response of each IMF components and derive the relative response of each detector by the RSI radiometric conversion function to calibrate the original images. Taking G2 result for example, standard deviation of calibrated imageries can be dramatically decreased (PAN: 66.31 to 1.85, B1: 54.19 to 1.90, B2: 36.50 to 1.49, B3: 32.43 to 1.56, B4: 37.67 to 1.20), and estimate the step of relative radiometric calibration by the result from the digital mode, we can suggest the potential practicability of EMD for the relative radiometric calibration of FS-5 RSI sensor. |
