張斐章Chang, Fi-John臺灣大學:生物環境系統工程學研究所鍾昌翰Chung, Chang-HanChang-HanChung2010-05-052018-06-292010-05-052018-06-292008U0001-2307200814295300http://ntur.lib.ntu.edu.tw//handle/246246/181121河床質調查最重要之目的在於暸解河床粒徑分佈資訊，ㄧ般傳統床質調查有體積、網格與面積法等，可依據調查目的不同，選用不同的調查方法，如表層分佈與底層分佈；目前水利界常用的為體積法，然而此法工作量龐大，費時費力，往往需要許多的人力、物力投入其中，造成許多資源的浪費；近年來照相與影像處理技術進步迅速，在辨別與量測方面皆有良好的成果，可作為量測河床粒徑分佈的有利工具；本研究以固定面積拍攝所得影像，透過製作標誌(marker)為基礎，提出四個主要步驟：(1)影像預處理，(2)標誌製作，(3)影像切割，(4)測量最大短軸；上述方式可抑制分水嶺演算法過度切割的缺點並獲得良好的分割結果；我們以台灣北部景美溪所採樣的石頭帶回實驗室隨機排列，並拍攝影像進行分析，藉由上述方法，推算其所得個數及粒徑百分比累積曲線，與實驗室篩分析所得個數及粒徑百分比累積曲線相比有良好的結果，可作為後續研究之參考及快速研判河床粒徑分佈之用。The measurement techniques of river materials are mainly to get surface grain-size distribution information. There are several traditional measurement techniques, such as volume, grid, and area measurement methods. Among them, the volume measurement method is the most common method used by the Hydraulics, but this method needs a hugeorkload, time and energy. Image analysis techniques have been shown to work well in identifying and measuring particles, consequently they can be powerful tools for measuring the grain size distributions. In thisaper we present a rapid image-processing-based procedure for the measurement of exposed fluvial gravels, defining the steps required to minimize the errors in the derived grain size distribution. The main procedure is divided into four steps: (1)image pre-processing, (2)markeraking, (3)image segmenting, and (4)maximum b-axis measuring. The analyzed stones were obtained from Jingmei River and randomly disposed within a square meter grid in the laboratory and taken picture for the analysis. The measurement errors compared with sieve analyses areuite small in all the cases, consequently we can conclude that the image processing method proposed in this study can efficiently and precisely identify the grain-size distribution and can be used in the follow-upesearch.目錄 要................................................... Ibstract ................................................ II錄................................................... III目錄.................................................. VI目錄................................................. VII一章 前言.............................................. 1.1 研究動機......................................... 1.2 研究目的......................................... 5.3 論文架構......................................... 5二章 文獻回顧.......................................... 6.1 表面採樣方法特性比較............................. 7.2 影像處理應用於量測方面........................... 8.3 影像處理應用於河床質調查......................... 8.4 脈衝耦合類神經網路............................... 9三章 理論概述......................................... 10.1 影像預處理...................................... 10.1.1 影像擷取................................... 10.1.2 影像校正................................... 12.1.3 灰階化..................................... 13.1.4 影像縮小................................... 14.1.5 梯度運算................................... 14.2 影像切割........................................ 16.2.1 分水嶺轉換(Watershed transformation) ....... 16.3 標誌標記........................................ 20.3.1 反饋式脈衝耦合類神經網路(Feedbackulse-Coupled Neural Network, FPCNN) ............. 20.3.2 侵蝕(Erosion) .............................. 24.3.3 標記連通成分(Marker Connected Component) ... 25.3.4 合併規則................................... 27.3.5 區域填充................................... 29.3.6 距離轉換................................... 30.4 影像測量........................................ 32.4.1 Hotelling 轉換............................. 32四章 研究方法......................................... 34.1 採樣作業........................................ 35.2 實驗設計........................................ 40.2.1 實驗目的................................... 40.2.2 實驗器材................................... 40.2.3 實驗步驟................................... 44.2.4 實驗樣本................................... 48.3 模式的建立...................................... 51.3.1 FPCNN 的迭代次數........................... 51.3.2 規則運算修正標誌........................... 54.3.3 整體模式的步驟、流程....................... 58五章 結果與討論....................................... 62六章 結論與建議....................................... 75.1 結論............................................ 75.2 建議............................................ 76amp;#63851;考文獻................................................ 77application/pdf3183467 bytesapplication/pdfen-US影像處理河床粒徑類神經網路分水嶺轉換Image processingRiver materialsArtificial neural networkWatershed transformthesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/181121/1/ntu-97-R95622031-1.pdf