運用監督式分類技術辨識桃園藻礁露出範圍之研究初探;Preliminary study on identification of algae-reef areas using supervised classification methods in Taoyuan coast

NCU Institutional Repository > 地球科學學院 > 水文與海洋科學研究所 > 博碩士論文 > Item 987654321/89252

jsp.display-item.identifier=請使用永久網址來引用或連結此文件: http://ir.lib.ncu.edu.tw/handle/987654321/89252

题名:	運用監督式分類技術辨識桃園藻礁露出範圍之研究初探;Preliminary study on identification of algae-reef areas using supervised classification methods in Taoyuan coast
作者:	林宜瑾;Lin, Yi-Chin
贡献者:	水文與海洋科學研究所
关键词:	影像分析
日期:	2022-09-21
上传时间:	2022-10-04 11:03:01 (UTC+8)
出版者:	國立中央大學
摘要:	桃園藻礁為海岸重要生態棲地，其範圍易受漂沙覆蓋而改變且位置緊鄰第三天然氣接收站，因此藻礁辨範圍辨識為重要研究議題。本研究透過監督式分類法辨識藻礁範圍，並將分類結果與人工辨識結果進行準確度評估，藉以檢視不同分類方法之成效。本研究利用 RGB影像及數值地表模型影像及數值地表模型 (DSM)兩種資料進行影像分析，別運用三種方法設定訓練樣本的閥值，藉以分類礁體與沙區域。其方法分別為：(1)色彩強度：運用歐幾里得距離計算組成的相色彩強度：運用歐幾里得距離計算組成的相似程度；(2)影像梯度：利用灰階影像梯度：利用灰階影像(gray-scale image)在過渡不同區域所產生的在過渡不同區域所產生的連續性，檢測其梯度變化進行區域劃分；(3)粗糙度粗糙度 (Rugosity)：計算實際地表起伏的表面積與正交投影後平面表面積的比值。另將上述三種分類之結果進行交集，以驗證同時採用多種分析方法是否能提升影像類的準確度。三種的分類方法中，使用色彩強度進行時其準確約為 0.49左右，無法有效的區分礁體與沙之域；灰階梯度為三種方中較佳類，其準確度可達 0.8左右；粗糙度分類之準確落於0.34-0.71之間，雖能辨識部分礁體區域，但其準確度變化幅較大。此三種方法交集結果之於0.54-0.71之間，雖無法高於灰階梯度準確但優色彩強或粗糙個別分類之結果。;Taoyuan algal reef is an important coastal ecological habitat, and the area of the reef is easily changed by the coverage of drifting sand. Recently, due to the construction of the third liquefied natural gas (LNG), the algal reefs in Taoyuan are at high-risk of being threatened. Therefore, monitoring the variation of algal reef areas is necessary and an important research topic. This study uses supervised classification to identify the area of algal reefs. Besides, we evaluate the accuracy of the classification methods by comparing the consistent areas between methods to the manual identification results. Specifically, to identify the reef area and compare it with the manual results, we applied three methods using the orthophotograph and numerical surface model (DSM) data to classify images. Three methods were applied in this study are called as: (1) Color intensity: using the intensity of color composition (RGB) to classify reefs and sand ; ( 2) Image gradient: the threshold value of detection classification through the variation of the grayscale caused by the transition of heterogeneous regions; (3) Rugosity: Calculating the ratio of the surface area of the undulating terrain to the plane after orthogonal projection. In addition, the results of the three classified methods above are intersected to verify whether the compilation of multiple analysis methods can improve the accuracy of image classification or not. The results show that the color intensity accuracy is 0.49. Compared to other methods, this accuracy value is not as well as other methods for classification. Thereby, the color intensity is not the perfect method to distinguish the reef and sand areas. Whereas, the grayscale method gives the highest accuracy of about 0.8. The accuracy of the rugosity method is range between 0.34-0.71. Although rugosity can identify some parts of reef areas, but the accuracy varied with the mixing areas between reef and sand. Moreover, the accuracy of the intersection of the three classified methods ranges between 0.54-0.71. Even can not be higher than the accuracy of the gray-scale gradient method but better than the color intensity and rugosity method.
显示于类别:	[水文與海洋科學研究所] 博碩士論文

文件中的档案:

档案	描述	大小	格式	浏览次数
index.html		0Kb	HTML	82	检视/开启

在NCUIR中所有的数据项都受到原著作权保护.

社群 sharing

数据加载中.....