臺灣位於西太平洋上，四面環海，處於熱帶季風與副熱帶季風氣候中，這樣的環境中使得臺灣成為鋼鐵鏽蝕的溫床。鏽蝕會使得鋼橋強度逐漸下降，而臺灣又常受到颱風與地震災害，橋梁的維護逐漸受到重視。而目前臺灣所使用之檢測法為「目視法」，檢測結果受檢測者的經驗影響過大。故本文擬藉由數位影像識別發展一套自動化、系統化的檢測程序。
本研究首先收集100張鏽蝕影像並利用HSI之H圖層與影像在灰階下之COV將數位影像區分為三類，並利用傳統的K-means法(在H圖層)與自行開發之雙園心半徑法(在RGB與HIS色彩空間下)對上述三類影像進行鏽蝕區域的識別；其中，並利用LS-SVM提升影像分類的準確度與雙半徑法中半徑的預測，雖然本文所提的演算法在影像分類上仍有改善的空間，但其對於具有陰影、流鏽與不均勻光源影像的識別均優於先前的演算法(AEA)，有助於實務上鋼橋鏽蝕區域面積的識別。

Because Taiwan is surrounded by the sea, the humidity level in most regions is notably high, and steel structures are easily invaded by sea breeze. Steel corrosion will decrease a structure’s strength. Moreover, probability of having a natural disaster such as typhoons and earthquakes is not low in Taiwan. As a result, a good maintenance plan for steel structures is expected. The most common method of detecting a defect in steel bridge in Taiwan is an eye-examination. Such method greatly depends on the experience of the detector, resulting in a subjective approach. Therefore, this article aims to develop an objective approach in which the technique of digital image analysis is adopted. First, we collected 100 images of rust, and then use ranges of H layers and cov of grayscale of images to further divide images into three groups. Three corresponding techniques are developed to identify the corroded area for each group and they are: the traditional K-means method in the H layer, double centers-radiuses algorithm in RGB color space and in HSI color space. The LS-SVM is used here to predict the radius in the double centers-radiuses approach. In addition, the LS-SVM is also adopted here to categorize images into three groups. Although the algorithm developed in this article still have a room to improve, its performance in identifying corroded area is better than some existing algorithms such as AEA, especially, for an image of having a shadow area or an uneven light streaming.

[1] 彭國瑞，「智慧型影像處理於鋼構橋樑表面塗裝檢測之應用」，碩士論文，國立交通大學，民國 89 年 6 月。(指導教授：黃世昌博士)
[2] P.H. Chen, Y.C. Yang, L.M. Chan, ”Automated bridge coating defect recognition using adaptive ellipse approach” Automation in Construction, Vol. 18, No. 5,. August 2009, pp. 632-643.
[3] 楊雅晴，「運用智慧型彩色影辨識於鋼橋生鏽檢測」，碩士論文，國立台灣大學，民國 98 年 7 月。(指導教授：張陸滿教授)(共同指導教授：陳柏翰教授)
[4] 色彩空間，維基百科，http://en.wikipedia.org/wiki/Color_space。
[5] HSI色彩屬性模式，維基百科，http://en.wikipedia.org/wiki/HSL_and_HSV。
[6] 彩色座標系統YIQ Color Space，http://cmf-resource.blogspot.tw/2009/02/yiq-color-space.html。
[7] K-means clustering, WIKIPEDIA, http://en.wikipedia.org/wiki/K-means_clustering.
[8] K-means 分群演算法，陳鍾誠，http://ccckmit.wikidot.com/。
[9] 薛丞佑，「Cauchy-Riemann方程式與影像邊緣偵測」，碩士論文，國立中正大學，民國 99 年 8 月。(指導教授：賴振耀博士)
[10] Edge detection, WIKIPEDIA, http://en.wikipedia.org/wiki/Edge_detection.
[11] 交通部運輸研究所，「臺灣地區大氣腐蝕劣化因子調查研究(1/2)」，民國 97 年 3 月。(著者：陳桂清、柯正龍、羅俊雄、羅建明、陳新北、劉益雄、翁榮洲)
[12] 交通部運輸研究所，「臺灣地區大氣腐蝕劣化因子調查研究(2/2)」，民國 98 年 4 月。(著者：陳桂清、柯正龍、羅建明、羅俊雄)
[13] 彭健賓，「鐵塔鍍鋅層及塗膜劣化影像分析及資料庫建立」，碩士論文，國立台灣科技大學，民國 95 年 6 月。(指導教授：鍾國亮博士)
[14] Lab color space, WIKIPEDIA, http://en.wikipedia.org/wiki/Lab_color_space.
[15] Support vector machine, WIKIAPEDIA, http://en.wikipedia.org/wiki/Support_vector_machine.
[16] William J. Palm III,”MATLAB 7 在工程上的應用”，高立圖書有限公司。(王晉中譯)
[17] A. McAndrew, J.H. Wang, C.S. Tseng,”Introduction to digital image processing with MATLAB Asia Edition”。(劉震昌審譯)
[18] 廖國偉、呂胡忠、Doddy Prayago、Alvaron pratomo，「橋梁基礎耐洪能力可靠度分析」，投稿於「結構工程」審稿中。
[19] 王勝均、蔡俊明、葉榮木、許溥鑫，「應用於停車場之即時動態車牌偵測」。
[20] 張智星，「MATLAB程式設計進階篇」，鈦思科技出版。
[21] Sobel operator, WIKIPEDIA, http://en.wikipedia.org/wiki/Sobel_operator.
[22] 交通部運輸研究所，「港灣地區碼頭腐蝕資料及設施結構資料更新建置(1/2)」。(著者：張道光、謝明志、柯正龍、單誠基)
[23] 交通部運輸研究所，「港灣地區碼頭腐蝕資料及設施結構資料更新建置(2/2)」。(著者：張道光、謝明志、柯正龍)
[24] P.H. Chen, Y.C. Chang, L.M. Chang, Peter C. Doerschuk, “Application of multiresolution pattern classification to steel bridge coating assessment”, ASCE 0887-3801~2002, 16:4~244.
[25] H.D. Cheng, X.H. Jiang, Y. Sun, Jing Li Wang, “Color Image Segmentation: Advances & Prospects”, Utah State University.
[26] Y.I. OHTA, T. Kanade, O. Sakai, “Color Information for Region Segmentation”, Vol. 13, NO. 3,. July 1980, pp. 222-241.