滑鼠觸控板光學瑕疵檢測系統之開發

Development of an Optical Flaw Defection System for Touch Pad

練建良(Chien-Liang Lien)

指導教授 ： 吳明川

國立臺北科技大學/機電學院/機電整合研究所/碩士(2006年)

摘要

本研究針對筆記型電腦上的滑鼠觸控板(Touch Pad)的表面瑕疵，利用機器視覺的技術建立一套自動化的檢測系統。在滑鼠觸控板的製造過程中，常常因為機器的油墨塗佈不均、或是在運輸的過程中產生瑕疵。而主要的瑕疵種類包括：尺寸不良、刮痕、油墨汙點與油墨汙漬這四類。而在黑色的滑鼠觸控板中，瑕疵具有低對比性與非均勻性的特質，所以現階段還是以人工方式來作瑕疵判斷，而這樣的方式不僅使得檢測的成本提高，且因為人工檢測的因素，無法達到固定的品質標準。在本文研究中，係針對這些問題，發展一套檢測技術來取代人工檢測，以期達到減少人力資源，提高生產品質、生產效率及降低生產成本。
在研究中所使用的機器視覺技術，主要有影像對比增強，將瑕疵的對比性提高，搭配實數型基因演算法(Real-Valued Genetic Algorithm, GA)找出最佳的對比閥值，利用空間域濾波方式消除雜訊，經由邊界特性的局部臨界法與形態學濾波方式將瑕疵從樣本中清晰的分割出來，再經由倒傳遞類神經網路(Back-Propagation Neural Network)分類出瑕疵種類，本研究使用的影像處理方式，可以將瑕疵部份快速且精確的檢測出來，而最後的總檢測平均時間約為0.19~0.35秒，比用人工檢測需花上3~4秒要快上許多。

關鍵字

滑鼠觸控板 ； 實數型基因演算法 ； 倒傳遞類神經網路

並列摘要

This paper develops an automatic inspection system for touch pad of notebook using machine vision. The surface of the touch pad is often unsmooth during painting process and some defects are easily produced in the course of transportation. There are four kinds of main defects including inaccurate size, scratch, spots and stain. As to black touch pad, the features of defects are low contrast of image and not homogeneous. So the inspect factory heavily depend on human vision. Using human vision does not only raise the cost of inspection, but also does not reach the quality standard. In this study, we develop an automatic inspection system to replace human vision and increase the inspection efficiently, the quality of products and the quantity of output.
In this research, image processing such as image contrast enhancement, spatial filter, edge detection, morphology and image classification are used to inspect the defects. In this paper we use Real-Valued Genetic Algorithm (RVGA) to find out the best threshold value of contrast. In the image segmentation, we use edge detection to separate the defects from pad, and uses Back-Propagation Neural Network to recognize kinds of the defects. It is hope for the proposed inspection system can detect the defects quickly and accurately. It is shown that the system only takes 0.19~0.35 seconds for image process in standard of 3~4 seconds by human vision. So the proposed automatic inspection system is better than human vision.

並列關鍵字

Touch Pad ； Real-Valued Genetic Algorithm ； Back-Propagation Neural Network

參考文獻

[32] 泰洛科技股份有限公司，「增進視覺系統辨識之光源選擇與技巧」，機電整

[16] 鄭光宏，應用影像視覺於超薄型表面載式電感器之線上自動檢測，碩士論文，

金禾資訊，2004。

所，高雄，2000。

科技大學機械工程系，屏東，2001。