將全偏極合成孔徑雷達(Polarimetric Synthetic Aperture Radar；POLSAR) 所能提供之資料作最佳利用，對不同地物之分辨而言，是甚有幫助的。傳統上利用特徵分析法(eigenanalysis)合成一極化配當篩選器(Polarimetric Matched Filter；PMF)，來解決極性對比強化之最佳化問題(Optimization of Polarimetric Contrast Enhancement；OPCE)，以達到各個地物反射強度間的對比為最大之目的，但其中牽涉較多之代數運算，且一次只能獲得兩類別間之對比值，當地物種類增加時，傳統方法便突顯其不切實際之處。 本論文嘗試以三種案例來描述基因演繹法(Genetic Algorithm；GA)於對比強化之應用，(1) 利用限制性的基因演繹法及數值資料，執行點目標參數之最佳化。(2) 利用基因演繹法及影像資料，執行面目標參數之最佳化，並以系統角度建立四種探討方式，分別找出任意兩類別間之最大對比值。(3) 利用基因演繹法及影像資料，執行面目標參數之最佳化，一次同時得到任意兩類別間之對比最大值。 經以國內外地區，分別包含不同類別數之全偏極合成孔徑雷達影像資料作測試驗證後，證明本方法可有效處理極性對比強化之最佳化問題，其可視為另一種甚至更方便之方法。雖然各類別間對比同時一次求解時，精度稍有損失，但其可獲得計算時間上明顯之效益，在實際應用上，當地物類別眾多且時間敏感之情況下，此項功能特別有其意義。 Polarimetric scattering information provided by fully polarimetric synthetic aperture radar (POLSAR) is useful for discriminating between different terrain covers. Conventionally, eigen-analysis is used to synthesize a polarimetric matched filter (PMF) that maximizes the contrast between the features of interest, but this involves many algebraic operations and only one contrast ratio for any two classes can be obtained at one time. It’s impractical when there are so many terrain targets in the image. Polarization synthesis offers a mechanism to find the transmitted and received polarizations that maximize the received power contrast between two terrain types and then to create a maximum class-separation image by applying this knowledge. This thesis describes the application of a genetic algorithm (GA) to the contrast enhancement in three cases. In the first case a constrained GA is used to optimize parameters of point target using numerical data. The second case utilizes GA to find optimal parameters to maximize contrast ratios between any two classes individually and sequentially using real image data. From the systematic point of view, four approaches are devised to investigate the result. The last, nevertheless, the most important case is to use GA to search optimal parameters that can obtain any possible contrast ratio between any two classes simultaneously. Four independent sets of L-band fully polarimetric SAR images that contain different number of terrain class are used as test data. We also analyze the performance under different conditions, including parameters of GA and frequency of test data. It is demonstrated that all the contrast ratios characterizing the best discrimination for any two of all possible classes can be obtained simultaneously, providing another and even more convenient way to solve the problem of optimization of polarimetric contrast enhancement (OPCE). The results can provide excellent preprocessing for subsequent image classification, if desired. The method presented in this thesis is cost-effective in terms of contrast ratios and computation time.
