多焦點鏡片扭曲檢測與分類系統

Automated Distortion Detection and Classification on Multifocal Lenses

李東昕(Tung-Hsin Lee)

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(2019年)

摘要

隨著網路世代的興起與行動裝置的普及，人們使用電子產品的時間越來越長，對於眼睛的依賴與造成的負擔也越來越大，導致輔助視力的眼鏡需求量逐漸上升。眼鏡的鏡片因應近遠視的需求都是曲面的形狀，多焦點鏡片是一種新型的鏡片類型，可以同時兼具看較遠視野與看較近視野，但缺點是當鏡片在磨製的過程中，由於鏡片中不同區域的曲度有些微的變化，而導致鏡片在成像上容易出現光學變形。如果光學變形太嚴重，消費者在使用上會造成生活上的不便，因此鏡片的品質需要嚴格把關。現行檢測儀器只有量測鏡片之度數、透光率等鏡片數據，目前無儀器針對扭曲瑕疵進行量化與檢測，僅依賴專業人員的經驗判斷，因此本研究開發自動化扭曲瑕疵檢測系統，將具有一貫性、準確性、可循環檢測等優點。
本研究針對多焦點鏡片的扭曲瑕疵提出使用同心圓標準板成像在鏡片上，並計算每一圈同心圓邊點的質心半徑描述子，接著使用EWMA管制法將有扭曲瑕疵的區域找出來。之後藉著瑕疵影像與標準板影像進行線條差異比對，並將根據瑕疵發生的位置將影像分三個區域，再分別輸入三個區域的扭曲量特徵值，透過建立模糊歸屬函數與規則庫，最後使用GA-ANFIS模式進行瑕疵嚴重程度的分類。本研究初步使用小樣本實驗確定各程序之較佳參數設定，由較佳參數進行大樣本實驗，共使用150張測影像進行扭曲瑕疵偵測與分類。實驗結果顯示，本研究方法之扭曲瑕疵誤判率(α')為10.94%、檢出率(1-β)為81.09%且瑕疵嚴重程度之正確分類率(1-γ)達94%。

關鍵字

GA-ANFIS模式 ； 微量偏移偵測 ； 扭曲瑕疵 ； 光學檢測 ； 多焦點鏡片

並列摘要

The using time of electronic devices is becoming longer for people’s daily lives, and the burden of people’s eyes is getting larger, while we approaching the rising of internet generation and the popularity of mobile devices. Thus, the demand for eyes-glasses is gradually increasing. The lenses of glasses are made to be curved to meet requirements of myopia and hyperopia. Multifocal lens is a new type of lens that is designed to provide focus of both distance and near objects. The main risk of producing multifocal lens is that the lens is easy to cause optical distortion in imaging while different regions of the lens have different curvatures when the lens is in the process of grinding. If the optical distortion defects are too serious on lens, the consumers will have distorted scenes in their sights and cause inconvenience in lives. Therefore, the quality of the lens needs to be strictly controlled. This study develops an automated distortion defect detection system for multifocal lenses to replace professional inspectors relying on experienced judgments.
In this paper, we propose a novel approach based on a standard pattern of concentric circles to inspect optical distortion defects on multifocal lenses. First, we calculate edge points of each circle of the concentric circles based on a centroid-radii model. Second, we find distortion defect areas through the exponentially weighted moving average (EWMA) control scheme. Third, we make difference comparison between defect image and standard pattern image. Fourth, we divide the image into three levels of severity according to defect locations on the lens. Fifth, we input the distortion deviations for the three levels and set up the fuzzy membership functions and inference rule base. Finally, the GA-ANFIS model is applied to classify the levels of severity for the distortion defects. Experimental results show that the proposed methods achieves a high 94% correct classification rate of severity of distortion defect, 81.09% distortion defect detection rate and 10.94% false alarm rate.

並列關鍵字

Small-shift control scheme ； Distortion defects ； Optical inspection ； Multifocal lens ； GA-ANFIS model

參考文獻

[1] Agnieszka, B., Damian, S., “Optical performance of the eye with progressive addition lens correction,” Optik, Vol. 121 pp. 1937-1940 (2010).

[2] Banerjee, A., Abu-Mahfouz, I., “Crack Detection and Identification Using Vibration Signals and Fuzzy Clustering,” Procedia Computer Science, Vol. 114, pp. 266-274 (2017).

[3] Barbosa, E. A., Silva, D. M., Nascimento, C. E., Galvão, F. L., Mittani, J. C. R., “Progressive power lens measurement by low coherence speckle interferometry,” Optics and Lasers in Engineering, Vol. 51, No. 7, pp. 898-906 (2013).

[4] Bibiloni, P., González-Hidalgo, M., Massanet, S., “General-purpose curvilinear object detection with fuzzy mathematical morphology,” Applied Soft Computing, Vol. 60, pp. 655-669 (2017).

[5] Boullosa-Falces, D., Gomez-Solaetxe, M. A., Sanchez-Varela, Z., Trueba, A., “Validation of CUSUM control chart for biofouling detection in heat exchangers,” Applied Thermal Engineering, Vol. 152, pp. 24-31 (2019).