Please use this identifier to cite or link to this item: http://hdl.handle.net/11455/99612
標題: 由UAV影像自動萃取河床表面粒徑分佈
Automated Extraction of Surface Grain-Size Distribution From UAV Images
作者: 詹勳全
林柏瑋
鄭卉君
Hsun-Chuan Chan
Po-Wei Lin
Hui-Chun Cheng
關鍵字: 無人飛行載具;影像處理;粒徑分佈;Unmanned Aerial Vehicles;image-processing;grain-size distribution
Project: 中華水土保持學報, Volume 50, Issue 3, Page(s) 102-115
摘要: 
河床的粒徑組成為溪流重要的環境因子,舉凡:水流的流動、河床的沖淤、棲地的型態與水質的變化皆會受粒徑分佈而影響,本研究嘗試建立一個安全、機動性高且快速的河道粒徑分析系統,以無人飛行載具(UAV)進行河道底床表面顆粒的影像拍攝,配合自動化的影像處理(image processing)來辨識顆粒的輪廓,經由影像判視出顆粒的代表特徵長度,統計分析特徵長度後獲得粒徑分佈特性。影像處理過程除了採用以往研究常用的邊緣偵測法與集水區分割法進行顆粒輪廓辨識外,本研究亦針對集水區分割法進行修正,改善影像辨識顆時顆粒無法分割與過度分割的情況。河道粒徑分析系統以室內與室外實驗進行應用測試,室內實驗主要檢核影像辨識顆粒之數量與代表長度的正確性,而室外實驗則用於檢核系統分析之河道表面顆粒群體粒徑的分佈特性。室內分析結果顯示,在礫石顆粒為分別為分離、緊密與重疊排列之情況下,修正集水區分割法均能辨識出正確的顆粒數量,且辨識出的粒徑代表長度具有最小之平均誤差;室外分析先以人工調查(包括:卵石採樣分析、方形網格採樣及人工圈繪影像顆粒)進行底床表面粒徑分佈分析,邊緣偵測法、集水區分割法及修正集水區分割法所得之粒徑累積百分比達到50%所對應的顆粒尺寸(D_(50))與人工方式分析比較之誤差百分比分別為2.46~6.73%、23.60~26.94%與0.94~5.28%。由室內與室外實驗分析之結果可知,修正集水區分割法可準確辨識表面顆粒的輪廓與數量,求得的代表粒徑分佈特性與人工方式結果相近,本研究提出的河道粒徑分析系統可有效且正確進行河道粒徑分佈調查。

The composition of riverbed materials is an important environmental factor, which affects the water flow, riverbed variation, river habitat, and water quality. This paper presents a safe and highly mobile unmanned aerial vehicle (UAV)-based image-processing system for determining the grain size in gravel-bed rivers. The image characteristics of a riverbed surface were photographed using a UAV, and automatic image-processing procedures, such as edge detection and watershed segmentation, were used to recognize the gravel's outlines. Subsequently, the grain-size data were analyzed to obtain the grain-size distribution. This study proposed an improved watershed segmentation procedure to eliminate the problems of over- and under-segmentation in the image-processing procedures. The system was tested using indoor and outdoor experiments. The indoor experiments mainly checked the accuracy of gravel numbers and characteristic lengths identified by the system, whereas the outdoor experiments evaluated the grain-size distribution it produced. The results of the indoor experiments indicated that the improved watershed segmentation procedure correctly recognized the gravel number; furthermore, minimal mean errors in characteristic lengths occurred compared with the other methods used in this study on gravel with separated, closed, and overlapping arrangements. In the outdoor experiments, the surface grain-size distribution of the gravel bed was first analyzed through manual investigation, which included pebble counts, grid counts, and manual extraction of information from images. The results were compared with those obtained through edge detection, watershed segmentation, and improved watershed segmentation. The mean absolute percentage errors of the median grain size (D_(50)) were 2.46-6.73%, 23.60-26.94%, and 0.94-5.28%, respectively. According to the indoor and outdoor experiment results, the improved watershed segmentation procedure could accurately recognize the outlines and correct gravel number and could acquire the grainsize distribution with results similar to manual investigation. The UAV-based image-processing system proposed in this paper could effectively and correctly survey grain-size distributions in gravel-bed rivers.
URI: http://hdl.handle.net/11455/99612
Appears in Collections:第50卷 第03期

Files in This Item:
File SizeFormat
1.pdf5.02 MBAdobe PDFView/Open
Show full item record
 

Google ScholarTM

Check


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.