資訊萃取於高解析度影像在紋理資訊對海岸漂流木之辨識差異研究

The Study on the Different Texture Information of Coastal Driftwood Based on Information Extraction through High-Resolution Image Data

漂流木(Driftwood)一直以來算是台灣一項珍貴的資產，然而漂流木的發生是一種自然現象，台灣原本就位在板塊交界處、地震頻繁地層較不穩定，一旦遇到颱風或豪大雨，雨量超過森林植物與土壤的負荷，地表流水流量大增，就可能產生洪水及崩塌，森林裡的樹木也因此被沖下。本研究以遙測技術高解析度影像分類在海岸垃圾的影像識別，以亂度基礎分類法(Entropy-based Classification)在原始波段(B,G, R, IR)和紋理資訊上的篩選，進行平行研究並比較8種紋理資訊的選取之識別差異，並與實際地貌進行差異化分析。本研究以支援向量機(SVM)的分類比較，再與人工檢核地貌的成果進行比較，製作出誤差矩陣表和主題圖，並針對海岸漂流木的特徵波段，並分析 有/無紋理資訊與最佳的紋理資訊選取之辨識成效。

Driftwoods have always been regarded as a precious asset in Taiwan. However, the driftwood is produced by a natural phenomenon. Taiwan was originally located at the junction of earth-structure plates in which the stratum was unstable with frequent earthquakes. Once encountered a typhoon or heavy rain, the rainfall run over the forest and driftwoods are flush to the coastal. Hence, the load on plants and soil, and the large increase in surface water flow, may cause flooding and collapse, and the trees in the forest will be washed down as a result. This study used high-resolution remote sensing image classification in image recognition of coastal garbage, and used Entropy-based Classification to screen the original bands (B, G, R, IR) and 8 texture information, and conduct parallel analysis. The study compared and identify the differences in the selection of different texture information, and conduct a differentiated analysis with the actual landform. In this study, the classification and comparison of support vector machines (SVM) were used to draw the thematic map of landforms of coastal. An error matrix and a thematic map were produced. The classification performance of the best texture information selection is also discussed.