高品質的焦炭對高爐的操作穩定、高爐產率及耗焦率有著重要的影響，其中焦炭的粒度大小及均勻性為品質上必須掌握之項目。因此本研究係建立一套焦炭粒度線上檢測系統，由高速快門CCD監測輸送中的焦炭，利用影像處理技術辨識出焦炭的形狀，以計算焦炭平均粒度。
研究中，利用環型的照明光源搭配CCD取得焦炭影像，運用影像處理的方法，包括影像強化、二值化影像分割、形態學運算…等，可有效的辨識影像中焦炭的二維形貌；並設計了一個邊緣偵測演算法，同時具有提高目標物亮度的效果，幫助提升焦炭影像辨識的正確率；依照焦炭區塊形狀，由主軸運算與旋轉搜尋，可以很快的找出其最小寬度，即為影像平面中的焦炭尺寸。
經由動態量測與現行手工量測之焦炭尺寸分佈比較，尋找焦炭平面尺寸應有的分級界限，並統計大量影像平面量測之焦炭顆粒尺寸與實際三維顆粒尺寸之間的分佈情況，以降低量測的誤差；經焦炭顆粒數的修正，解決影像辨識率的不足，乘上表層焦炭的平均顆粒重後，計算焦炭的平均粒度。
焦炭粒度檢測系統所計算的焦炭平均粒度與手工量測的表層焦炭平均粒度比較，平均誤差約為正負1% 左右；但與焦炭總平均粒度比較，誤差落在3.1%~10.6%，這是因為攝影機只能拍攝到表層焦炭的影像，對底層焦炭的資訊較無法掌握。因此，為了提高焦炭線上量測系統之適用性，統計了表層焦炭的平均粒度與焦炭總平均粒度之間的粒度差與粒度比值，用以修正焦炭平均粒度的量測值；結果顯示，利用粒度差修正方式，焦炭總平均粒度誤差降至-3%~3.4%，利用粒度比值修正方式，焦炭總平均粒度誤差降至-3.6%~3.4%；整體來說，兩種粒度修正方法皆能有效計算出較準確的焦炭總平均粒度。

The high-quality coke greatly influenced the stability, production and coke consumption rate of the blast furnace, the size and uniformity of coke are especially important that should be controlled. Therefore, this research is to build a system to do the coke size inspection online, which employs high shutter speed CCD to monitor the delivering coke and uses the technique of image process to distinguish the shape of coke. The consequence can be utilized to calculate the average size of the coke.
In the research, the ring light and CCD camera are used to acquire coke image and we used the image processing such as image enhancement, threshold and morphology to detect the 2D coke shape. In addition, we design an algorithm of edge detection which helps to elevate the accuracy rate and to light up the object. Furthermore, to search by rotational searching from the principal axis and the shape of the coke block, we can easily find the minimum width. That is the 2D coke size on the image plane.
To compare dynamic measurement with the manual measure, we search the limit of coke distribution on 2D plane and gather statistics the differences between the size of coke on 2D plane and 3D space to decrease the error of measurement. By measuring the weight, we calculate the average weight of every coke to get the data of the average size.
The average error of surface coke size between dynamic measurement and manual measurement is about ±1%. However, the average error of every coke size is about 3.1%~10.6%. The reason is that the video camera only records the image of the surface instead of the coke on the bottom. To increase the stability of the system, we count the difference and the percentage error of coke size between the surface and every coke average size and uses to consequence to modify the statistics of the average coke size. The result shows that using difference modification way, the average error falls to -3%~3.4%. Using rate modification method makes the error fall to -3.6%~3.4%. To sum up, two modification methods all apply to modify the average size of coke.

1. R. M. Carter and Y. Yan, “On-line particle sizing of pulverized and granular fuels using digital imaging techniques”, Meas. Sci. Technol. 14, pp. 1099–1109, 2003.
2. R. M. Carter and Y. Yan, “Measurement of particle shape using digital imaging techniques”, Journal of Physics: Conference Series 15, pp. 177–182, 2005.
3. Kursun, “Particle size and shape characteristics of kemerburgaz quartz sands obtained by sieving, laser diffraction, and digital image processing methods”, Mineral Processing & Extractive Metall. Rev., vol. 30, pp. 346-360, 2009.
4. I. M. B. Martin, D. C. Marinescu, R. E. Lynch and T. S. Baker, “Identification of spherical virus particles in digitized images of entire electron micrographs”, Journal of Structural Biology, vol. 120, pp. 146–157, 1997.
5. M. H. F. Wlkinson, Tsiipke Wjbenga, Gijs de Vries and Michel A. Westenberg, “ Blood vessel segmentation using moving-window robust automatic threshold selection”, IEEE Signal Processing Society, Barcelona, Spain, pp. 1093–1096, 2003.
6. K. Mogireddy, V. Devabhaktuni, A. Kumar, P. Aggarwal, and P. Bhattacharya, “A new approach to simulate characterization of particulate matter employing support vector machines”, Journal of Hazardous Materials, vol. 186, pp. 1254–1262, 2011.
7. S. Outal, D. Jeulin, and J. Schleifer, “A new method for estimating the 3D size-distribution curve of fragmented rocks out of 2D images”, Image Anal Stereol, vol. 27, pp. 97–105, 2008.
8. A. M. Rios, D. Sarocchi, A. L. Valdivieso and Y. Nahmad-Molinari, “Machine vision for size distribution determination of spherically shaped particles in dense-granular beds, oriented to pelletizing process automation”, Particulate Science and Technology, vol. 29, pp. 356–367, 2011.
9. 黃俊銘，“應用數位影像處理於焦炭粒度檢測”，國立成功大學機械工程研究所碩士論文，2009。
10. N. Otsu, “A threshold selection method from gray-level histogram”, IEEE Transaction. on Systems, Man Cybernetics SMC-9, pp. 62–66, 1979.
11. J. N. Kapur, P. K. Sahoo and A. K. C. Wong, “A new method for gray level picture thresholding using the entropy of the histogram”, Computer Vision, Graphics and Image Processing, vol. 29, pp. 273–285, 1985.
12. J. Kitler, J. Illingworth and J. Foglein, “Threshold selection based on a simple image statistic”, Computer Vision, Graphics and Image processing, vol. 30, pp. 125–147, 1985.
13. 廖紹剛，“數位影像處理”，台灣培生教育出版股份有限公司，2003。
14. M. Sonka, V. Hlavac and R. Boyle, “Image Processing, Analysis and Machine Vision”, Thomson, 2008.
15. 許裕騄，“棒鋼表面缺陷自動化檢測系統”，國立成功大學資訊工程研究所碩士論文，2007。
16. 鐘國亮，“影像處理與電腦視覺導論”，台灣東華書局股份有限公司，2008。
17. C.G. Reif, “Image Acquisition and Processing with Labview”, National Instruments, 2004.
18. NI Vision Concepts Manual, 2005 ed., National Instruments, 2005.