應用快速二維經驗模態分解提升衛星影像中波紋辨識及海底水深推估之研究

Abstract

往昔水深資料多由船隻量測、單波束或多波束測深儀來量測水深，上述方式有經費高、費時及人員安全等缺點，本文提出以光學衛星影像配合二維經驗模態分解法濾除衛星影像中的雜訊，來估算出水深的方法以解決此問題。本文的影像處理技巧包括擷取影像圖中的波紋線，以Otsu演算法區分影像為波紋群集及非波紋群集的二值化影像，由二維經驗模態分解法判斷出合理的波紋線，再由平滑化的波紋線求出波向線及兩波紋線間的長度，即為波長，最後運用波浪理論中的分散關係式由平滑化的修正波長估算海底水深。 本文選擇屏東縣枋寮海域及宜蘭縣頭城海域作為研究基地，選擇六個水深剖面作為評估模式的精度。使用修正波長由本模式推算枋寮水深與實測水深的均方根誤差為0.90m，而推估頭城水深的均方根誤差為1.00m。使用原始的波長推算與實測水深的均方根誤差在枋寮及頭城海域分別為1.42m及為2.54m，此顯示本文之研究方法有效修正推估水深的誤差。未來在沒有實測地形海域需要水深資料時，本方法可提供參考的初估水深資料。

Bathymetric estimation was commonly obtained using single/multi-beam echo sounder on boats. The marine surveying needs high cost, much time and possible risk of staff security. For solving the disadvantage, the method of estimating bathymetry from wave lengths between two crest lines of satellite images using image processing and bi-dimensional empirical mode decomposition (BEMD). Some techniques of image processing include: (1) determining extremals of reflectivity of satellite images regarded as possible crest lines; (2) Using the Otsu algorithm for the threshold of reflectivity and BEMD for reasonable crest lines; (3) determining wave rays normal to smoothed crest lines and calculated wavelength between two adjacent crest lines along a wave ray; (4) calculating the water depth through dispersion relation of the Airy wave theory when the wave period and wave length are obtained in the previous steps. The Fangliao coast of Pingtung county and the north Toucheng coast of Ilan county are chosen for examining the accuracy of the proposed method by the root mean squared error between the measured and calculated water depths. Six selected beach profiles are evaluated to have RMSE of 0.90m and 1.00m, respectively, for the Fangliao coast and the Toucheng coast using corrected wavelengths. The RMSEs are much less than those of 1.42 m and 2.54 m for the Fangliao coast and the Toucheng coast using originally obtained wavelengths. The proposed method is examined applicable for providing estimated bathymetry for coastal planning when measured bathymetry is unavailable.