模糊粒子群演算法與自我組織映射圖判釋之研究-以萬大水庫崩塌地為例

本論文研製之影像分類判釋方法，依前人文獻，從衛星影像的原始光譜值再加入八種輔助因子，作為影像之分類資訊，再運用模糊粒子群演算法(FCM+PSO)與自我組織映射圖(SOM)作為影像之分類器，建構崩塌的判釋規則，以此規則達到快速判釋崩塌地、有效降低時間及人力之目的。本研究首先在萬大水庫附近抽取60筆訓練資料的光譜值與輔助因子，建立訓練樣本的資料庫，運用模糊粒子群演算法(FCM+PSO)與自我組織映射圖(SOM)進行影像分類判釋，分別對驗證區進行判釋與準確度分析，最後透過坡地門檻值之類別轉換，以提高判釋率。本研究判釋結果顯示出，兩種影像分類器在還未加入坡地門檻值轉換時，判釋率可達到84.38%，再經坡地門檻值轉換後，判釋率增加了約 10%，判釋度有明顯提升。

關鍵字

模糊分群法；粒子群演算法；判釋規則；植生指標

並列摘要

The reservoirs are generally constructed on the mountain area at Taiwan. The earthquake results in the soil distributed and typhoon will bring a huge amount of water to the reservoir zone. The movement of rock and soil of landslide into the reservoirs will produce soil deposit which influence seriously on the delivery of water. Accordingly, the landslide surrounding the reservoir will also dominate its life-time. The multi-scenario remote sensing data can effectively monitor the reservoirs. Recently, the Linear Discriminant Analysis (LDA) is a well-known method to classify image categories. However, few studies have been made to optimize the classification function. That is, the ancillary information is adopted easily by new technology. Unfortunately, the ancillary information requires to be examined to apply efficiently into the landslide decision system. The proposed method includes (a)Fuzzy-C-means + Particle Swam Optimization (FCM+PSO) can find the core factors (b)Self Organization Map (SOM) to construct the knowledge rule. Both of the classifier can approach about 84% of landslide image classification accuracy. Then the translation scheme of category amend is developed to enhance about 10% of accuracy.