本研究欲研發一套專門對風機葉片表面剝落裂紋辨識之演算法，適用於葉片影像後處理，進而發現其剝落裂紋損傷以提供後續維修風機葉片之參考。此一套辨識系統大致可劃分為：現場圖像、圖像預處理、圖像分割、分析影像組件等四大部份。其中，對於圖像預處理的步驟中，包括去除背景、像素直方圖、二值化圖像等許多步驟。
透過比較各式演算法的優缺點，吾人可針對剝落型態之裂紋選擇適合的預處理演算法；在背景去除部分，吾人選擇以較簡單化之單一顏色背景與葉片實體進行運算，並求得適合的矩陣；而在二值化分析及中，可得知圖像在灰階程度區間像素直方圖比例，並初步估算出灰階圖像之閾值(Threshold)，分析影像組件的部分則須先進行上述的二值化閾值建立，導出葉片剝落型態裂紋與葉片主體之像素相對關係後，針對剝落裂紋邊緣進行計算，進而抓取此剝落裂紋區域於剝落裂紋邊緣上色驗證。本研究所開發的辨識系統，包含了辨識系統及分析功能，再最後的辨識成果，條件是要於影像形成灰階前景(破損)與背景(葉片)灰階差異程度高於27%，其辨識範圍區域設定為正投影之情況，不考慮拍攝模糊、天色昏暗、光線反射、低解析度之因素。

This thesis is to develop a vision algorithm which is used to detect and recognize surface cracks on wind turbine blades for image post-processing. This research proposed a method to filter out background and image segmentation based on recognition for surface crack of wind turbine blades. This algorithm is divided into four steps: image acquisition, image preprocessing, image segmentation, and image analysis. Image preprocessing include removing background, pixel histogram, image binarization, etc.
At first, it is proposed to set the appropriate environment for photography. The best way is to align the blade, sun and the camera with each other called orthographic projection, so we don’t consider the influence of blurring, darkness, light reflection, etc.
By comparing advantages and disadvantages of other algorithms, we select a suitable preprocessing algorithm for the crack of the peeling damage type. We choose simple color as background to extract blades by calculating proper matrix from every image.
In the image binarization analysis, the pixel histogram scale of the image in the grayscale extent can be known, and the threshold of the grayscale image is preliminarily estimated.
To analyze the image, the above-mentioned binarization threshold is established first. After comparing the blade peeling damage pixels with the blade pixels, we calculate the edge of the peeling crack that is used to detect the peeling crack region and verify the algorithm via drawing color on the edge of cracks.

[1] H. Zhang, “Feasibility of Automatic Detection of Surface Cracks in Wind Turbine Blades,” Iowa State University Industrial and Manufacturing Systems Engineering, 2014.
[2] 萬瑨, 利用UAV 影像進行牡蠣棚之辨識與災損評估, 國立中興大學木工程學系碩士論文, 2016.
[3] 林云媚, 廣域應變量測應用於大型風機葉片損傷診斷之研究, 國立台北科技大學木工程系土木與防災碩士班碩士論文, 2016.
[4] 陳柔安 曾千育, 基於貝氏定理的影像透明度分析與效能評估, 2015年台灣國際科學展覽會, 2015.
[5] Wind Turbine Blade Damage,
https://www.awfullive.site/wind-turbine-blade-damage/
[6] 阿洲的程式教學,
http://monkeycoding.com/
[7] 郭彥甫, “Matlab的工程應用,” 國立台大生機系講義.
[8] J. Canny, “A computational approach to edge detection,” IEEE Transactions on pattern analysis and machine intelligence, vol. 6, pp. 679-698, 1986.
[9] 繆紹綱, 數位影像處理活用Matlab， 全華圖書股份有限公司, 第二版，2015.
[10] I. Sobel, and F. Gary, “A 3x3 Isotropic Gradient Operator for Imageprocessing,” a Talk at the Stanford Artificial Project, pp. 271-272, 1968.
[11] M. Sharifi, M. Fathy, M. T. Mahmoudi, “A Classified and Comparative Study of Edge Detection Algorithms,” Information Technology: Coding and Computing, Proceedings. International Conference on IEEE, 2002.
[12] J. M. S. Prewitt, “Object Enhancement and Extraction,” Picture Processing and Psychopictorics, pp.15-19, 1970.
[13] H. Shi, and W. Rabab, “Canny Edge Based Image Expansion,” Circuits and Systems, ISCAS, IEEE International Symposium, vol. 1, pp.I-785- I-788, 2002.
[14] C. Tomasi, and T. Kanade, “Detection and Tracking of Point Features,” Carnegie Mellon University Technical Report CMU-CS, pp.91-132, 1991.
[15] Jr. Weih, C. Robert, and D. Norman, Jr. Riggan, “Object-based Classification vs, Pixelbased Classification: Comparative Importance of Multi-resolution Imagery,” The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 2010.
[16] Matlab Research: Correcting Nonuiform Illumination,
http://andypb.byethost24.com/andypb/?p=178&i=1
[17] Matlab forum,
https://www.ilovematlab.cn/forum.php?mod=home