應用非線性擴散模式於異質性紋路之多晶太陽能晶片表面微裂瑕疵檢測

Micro-crack inspection in heterogeneously textured of solar wafers using anisotropic diffusion

張志傑(Chih-Chieh Chang)

指導教授 ： 蔡篤銘

元智大學/工程學院/工業工程與管理學系/碩士(2008年)

https://doi.org/10.6838/YZU.2008.00016

摘要

本研究利用機器視覺的技術檢測多晶太陽能晶片(Solar wafer)表面微裂(Macro-crack)瑕疵，由於多晶矽太陽能製程所生產之晶片表面晶格分佈都不相同，造成每一片太陽能晶片表面影像具備獨一無二的晶格結構，且微裂瑕疵與正常晶格之輪廓都呈現細長線條的形狀，使得檢測微裂瑕疵的工作非常困難。

本研究使用LED正向光源取得微裂瑕疵影像，由於微裂瑕疵具備灰階值較週遭正常區域小(偏黑)且灰階梯度值較大的特性，因此本研究根據瑕疵之特性建構一非線性擴散(Nonlinear diffusion)模式，針對多晶太陽能晶片表面影像中梯度值較大且灰階值較小之瑕疵邊緣進行平滑處理，而對梯度值較小或灰階值較大之背景紋路則抑制平滑處理，再將平滑處理後影像與原始影像進行相減(Difference)運算以凸顯出瑕疵，接著就可以使用二值化(Thresholding)技術偵測瑕疵的位置。本研究透過此策略檢測多晶太陽能晶片表面之微裂瑕疵，可以有效的將背景晶格濾除同時保留瑕疵原有之輪廓。因為非線性擴散檢測多晶太陽能晶片的處理時間會受到擴散參數及疊代處理次數的影響，所以本研究使用較少的疊代次數搭配適當的擴散參數，並且建立擴散函數值對照表與只對邊緣點進行擴散處理，以縮短擴散處理時間；本研究測試50張瑕疵與45張正常影像後，實驗結果都能將微裂瑕疵偵測出來而無誤判，且處理一張 像素之影像僅需0.28秒。

關鍵字

機器視覺 ； 瑕疵檢測 ； 多晶太陽能晶片 ； 微裂瑕疵 ； 線性擴散函數

並列摘要

This research proposes a machine vision scheme for detecting micro-crack defects in solar wafer manufacturing. Micro-cracks affect the structural integrity of solar wafers. They should be detected in an early stage of the manufacturing process so that the overall production yield can be substantially improved. The surface of a polycrystalline wafer shows heterogeneous textures, and the shape of a micro-crack is similar to the background pattern. They make the inspection task extremely difficult.
The low gray-level and high gradient are two main characteristics of a micro-crack in the sensed image with front-light illumination. An anisotropic diffusion scheme is proposed to detect the subtle defects of micro-crack embedded in heterogeneous textures of solar wafer. The proposed anisotropy diffusion model takes both gray-level and gradient as features to adjust the diffusion coefficients. Only the pixels with both low gray-levels and high gradients will generate high diffusion coefficients. The diffusion model acts as an adaptive smoothing filter. It triggers the smoothing process in defective areas to change pixel gray-levels by assigning a large diffusion coefficient value close to one, and stops the diffusion process in faultless areas to preserve pixel gray-levels by assigning a small diffusion coefficient value close to zero. By subtracting the diffused image from the original image, the micro-cracks can be distinctly enhanced in the difference image. Then, a simple binary thresholding followed by morphological operations can be carried out to segment the micro-cracks and remove noise in the difference image.
By precalculating the diffusion coefficient function values in a look-up-table and triggering the smoothing process only for edge pixels, the proposed method can significantly improve the computational efficiency of the diffusion process. It can achieve a fast computation of 0.28 seconds for a 1K 1K image. Experimental results have shown the efficacy of the proposed method for numerous solar wafer images that contain various defect sizes and shapes.

並列關鍵字

Machine vision ； Defect detection ； Solar cell ； Micro-crack ； Anisotropic diffusion

參考文獻

林明獻，2007年，「太陽電池技術入門」，全華圖書股份有限公司。

蔡欣儒，2007年，「太陽能電池板的尺寸量測與線路瑕疵檢測」，碩士論文，國立中央大學資訊工程研究所。

陳心怡，2007年，「太陽能電池板表面瑕疵檢測」，碩士論文，國立中央大學資訊工程研究所。

Acton, S. T., J. P. Havlicek and A. C. Bovik, 2001, “Oriented texture completion by AM-FM reaction-diffusion,” IEEE Transactions on Image Processing, Vol. 10, pp. 885-896.

Bakalexis, S. A., Y. S. Boutalis and B. G. Mertzios, 2002, “Edge detection and image segmentation based on nonlinear anisotropic diffusion,” IEEE International Conference on Digital Signal Processing, Vol. 2, pp. 1203-1206.