影像分析應用於河床表面粒徑分佈計算

Image Analysis Applied to the Calculation of Grain Size Distribution on Riverbed Surface

劉家瑋(Jia-Wei Liou)

指導教授 ： 劉格非

國立臺灣大學/工學院/土木工程學研究所/碩士(2018年)

https://doi.org/10.6342/NTU201802515

摘要

河床剪應力(bed shear stress)、粒子沉降速度(grain settling velocity)、河床運移(bed load transport)等計算公式皆以粒徑(grain size)作為參數。而傳統求取河床表面粒徑分佈主要依靠人工進行採樣並量測，此方法雖然較精確且適用於各種環境，但需要耗費大量的人力及時間成本。近期隨著數位相機的普及與蓬勃發展，許多研究開始朝向以影像分析取代人工量測的自動化粒徑分析(Automated Grain Sizing, AGS)技術。縱使自動化粒徑分析技術於室內實驗中已得到很好的驗證；然而對於現地環境亮度分佈不均勻、石頭表面雜訊過多等因素，皆使得此技術離全自動監測的目標還有許多需要克服的地方。
本研究改進了自動化粒徑分析的架構，並將分析流程分為四大步驟。首先，從影像修正開始將影像修正為正射影像並進行環境亮度的修正，接著分別透過基於面與基於線的影像萃取方法萃取出影像中的石頭本體、邊界與背景區域，再將兩種萃取方法進行結合與修正；然而此時影像中部份區域的石頭仍然相互連通，故藉由標記分水嶺法對二值影像進行強制切割，使影像辨識的石頭數量與人眼辨識的結果達到相同數量級，最後利用橢圓擬合的方式進行粒徑量測，並透過分析與驗證可知道本研究分析一顆石頭長度的基本分析誤差為±10個像素寬。此外，本研究利用影像中的最大粒徑，取代分析程式中五項參數的設定，使分析流程達成半自動分析的目標。
最後本研究於三種不同的現地環境中，將方形網格採樣(Grid Sampling)之結果與本研究結果相比可發現，各粒徑累積百分比所對應的粒徑長度誤差皆小於預期的±10個像素寬以內。而本研究平均計算一張兩千四百萬畫素的影像所需花費的時間為12秒，相對於人工採樣量測的方法來說減少了約300倍的分析時間。

關鍵字

粒徑分佈 ； 影像分析 ； 影像分割 ； 自動化粒徑分析

並列摘要

Grain-size distributions (GSDs) often be using as a parameter in bed shear stress, grain settling velocity, bed load transport, etc. The traditional measurement of GSDs mainly relies on manual sampling and measurement. Although this method is more accurate and suitable for various environments, it requires a lot of manpower and time cost. Recently, with the popularity and flourishing of digital cameras, many studies have begun to develop automated grain sizing (AGS) based on image analysis. Even though AGS has performed well in indoor experiments, the unevenness light effect and grains with excessive noise in outdoor environment make the technology still have many problems to be overcome.
This study improved the architecture of AGS and divided the analysis process into four major steps. First, rectify the image to an orthophoto and adjust the unevenness light. Second, the image segmentation including line-based extraction and area-based extraction were used. Then we integrated these two extraction results to increase the degree of extraction and extracted a binary image. With this binary image, each grain boundary were separated by the well-known watershed segmentation method. Finally, the representative grain-sizes value were calculated after ellipse fitting. Through analysis and verification, we know that the basic analysis error of a grain size in this study is ±10 pixels. In addition, this study uses the maximum grain size in the image to replace the five parameters in the analysis program, so that the analysis process achieves the goal of semi-automatic analysis.
The results of grid sampling were compared with the results of this study in three different outdoor environments. The error corresponding to the cumulative percentage of each grain size was less than ±10 pixels. In this study, the average calculation time for a picture of 24 million pixels is 12 seconds, which is about 300 times less than the manual sampling method.

並列關鍵字

grain-sizes distribution ； image processing ； image segmentation ； automated grain sizing

參考文獻

Beucher, S., & Meyer, F. (1992). The morphological approach to segmentation: the watershed transformation. Optical Engineering-New York-Marcel Dekker Incorporated-, 34, 433-433.

Borgefors, G. (1986). Distance transformations in digital images. Computer vision, graphics, and image processing, 34(3), 344-371.

Bunte, K., & Abt, S. R. (2001). Sampling surface and subsurface particle-size distributions in wadable gravel-and cobble-bed streams for analyses in sediment transport, hydraulics, and streambed monitoring. Gen. Tech. Rep. RMRS-GTR-74. Fort Collins, CO: US Department of Agriculture, Forest Service, Rocky Mountain Research Station. 428 p., 74.

Butler, J. B., Lane, S. N., & Chandler, J. H. (2001). Automated extraction of grain-size data from gravel surfaces using digital image processing. Journal of hydraulic Research, 39(5), 519-529.

Chang, F.-J., & Chung, C.-H. (2012). Estimation of riverbed grain-size distribution using image-processing techniques. Journal of Hydrology, 440, 102-112.