綠化生態模型系列研究：比較隨機森林演算法和支援向量機在都市多類別影像判釋的差異

都市環境由於綠地不足，地表水泥地和柏油路造成熱度集中，人工發熱大，地表高蓄熱化，使都市有如一座發熱的島嶼，因此分析一個都市綠化程度是相當重要的議題。人工智慧目前已經廣泛被應用在很多地方，如果能透過機器學習和大數據分析技術將都市地貌進行有系統的分析，是綠環境重要的課題。本研究的目的是透過資料探勘辨識都市高光譜影像，來探討都市高光譜影像辨識度較差或較混淆的地貌有哪些，比較支持向量機、隨機森林在都市高光譜影像的分類效能，比較在都市高光譜影像的判釋效果。本研究的研究地區是台中市北區太原北路附近，其擁有道路、河堤、人行道、河、陰影、樹、草皮、操場、籃球場紅、籃球場綠、水泥屋頂、鐵皮屋頂12種地貌，各取90個點為訓練資料，45個點為測試資料。研究成果顯示之支持向量機（98.70%）高於隨機森林（93.89%）約4～5%。

關鍵字

影像判釋；支持向量機；隨機森林

並列摘要

Owing to the lack of green space in the urban environment, the surface cement and asphalt roads cause the heat concentration. It also produce artificial heat is high and the ground surface suffers to highly heat-storing. This makes the city like a hot island. Therefore, the urban greening degree analysis is a very important topic. Artificial intelligence has been widely used in image classification. A systematically analysis of the urban landscape through machine learning and big data analysis through technology is well-accept. It becomes an important issue to have better understanding the green environment. The purpose of this study is to identify urban hyperspectral imagery through data exploration to explore the urban landscape hyperspectral image classification. It compares the support vector machine, random forest in urban hyperspectral image classification outcomes and efficiency. The interpretation effect of hyperspectral images is used as image material. The research area of this study is near Taiyuan North Road, North District of Taichung City. It has 12 kinds of landforms including roads, river embankments, sidewalks, rivers, shadows, trees, turf, playground, basketball court red, basketball court green, cement roof and tin roof. It takes 90 points for training data and 45 points for testing data. The research results show that the support vector machine (98%) is about 4~5% higher than the random forest approach (94%).