透明玻璃之自動化可視瑕疵檢測

Automated visual defect inspection of transparent glasses

陳韋吉(Wei-Ji Chen)

指導教授 ： 林宏達

朝陽科技大學/理工學院/工業工程與管理系/碩士(2011年)

摘要

航空科技、建築材料和電子產業等經常使用到透明玻璃相關產品，使得玻璃瑕疵檢測變得相當重要。傳統的透明玻璃表面瑕疵檢測是採用人員目視檢測之方式，其人員目視檢測容易發生主觀意識判斷、漏判和誤判，且由於車用透明玻璃表面具有高穿透性與反射性，與其他類型之玻璃更加容易沾附灰塵及指紋等髒污造成檢測上的干擾，讓自動化檢測系統容易產生誤判，故本研究將針對車用透明玻璃之表面瑕疵檢測開發一套檢測系統，並對整張影像中個別瑕疵面積大於10個像素以上之瑕疵進行檢測。
本研究所開發之車用透明玻璃表面瑕疵檢測系統是先將待測影像由空間域使用離散哈特利正轉換至頻率域中，接著使用熵值計算出整張頻率值的資訊量，然後搭配模糊理論方法建立模糊歸屬函數，並利用此歸屬函數推論出盈縮係數矩陣，再將此矩陣之係數與對應位置的原始頻率值相乘，之後利用離散哈特利反轉換將濾波後的頻率值轉換至空間域影像內，最後使用簡單閥值切割方法即可分離出瑕疵與背景，達到去除背景中的干擾且能突顯出瑕疵之目的。實驗結果顯示本研究所提方法之平均瑕疵檢出率為98.08%，平均正常區域誤判率為1.68%，優於傳統玻璃瑕疵之檢測方法。

關鍵字

離散哈特利轉換 ； 車用透明玻璃 ； 表面瑕疵檢測 ； 熵 ； 模糊推論

並列摘要

Transparent glass products have become necessities in our daily life and major materials for construction, optical and electronic industries. Since the surface defects directly affect the quality of the transparent glass products, the detection of surface defects is very important for manufacturers. Human inspection is easy to be interfered by the external object images reflected or transmitted on the surface of transparent glass and results in making erroneous judgments of defect detections. Moreover, the surface of transparent glass product is easily attached to dust, dirt, fingerprints, and so on and make the defect inspection tasks more difficult. Therefore, this research aims at exploring the automated inspection of surface defects of car transparent glasses.
This study develops an automated visual defect inspection system of car transparent glasses. We first take Hartley transform of a testing image to frequency domain. The entropies of different radiuses of frequency components are calculated in Hartley domain. The fuzzy inference algorithm combining fuzzy rules and fuzzy membership functions is applied to choose adequate sizing factors (shrink coefficient matrix) based on entropy changes for modifying frequency components. After the frequency components are modified, the frequency domain image is transferred back to the spatial domain and yields the desired enhanced image. Then, the visual defects with low intensity contrast can be easily detected. Experimental results show that the defect detection rates achieve up to 98.08% and the false alarm rates lower to 1.68% by the proposed method and outperform the traditional defect detection methods.

並列關鍵字

Car transparent glass ； Surface defect detection ； Discrete Hartley transform ； Fuzzy inference ； Entropy.

參考文獻

[1] Aziz, Mohammed Yunis, “Radix-2*2*2 Algorithm for the 3-D Discrete Hartley Transform”, IEEE Transactions on Signal Processing, Vol. 49, No. 12 (2001).

[2] Bracewell, R. N., "Computing with the Hartley Transform," Computers in Physics Vol.9, No.4, pp.373-379 (1995).

[4] Bracewell, R. N., "The Fast Hartley Transform," Proceedings of IEEE, Proc. IEEE 72 (8), pp. 1010-1018 (1984).

[5] Bracewell, Ronald N. “The Hartley Transform” New York: Oxford university press (1986).

[6] Chang, Long-Wen, and Lee, Shen-Wen “Systolic Arrays for the Discrete Hartley Transform”, IEEE Transactions on Signal Processing, Vol. 39, No. II , November (1991).