- 學位論文
利用資料探勘技術分析彩色濾光片瑕疵之自動分類模型
Automatic Classification of Color Filter Defects by Data Mining
摘要
隨著資訊科技產品不斷的推陳出新，如電視機、平板電腦、及智慧型手機等，這些產品生命週期已逐漸的被縮短，因此對於產品的產能與品質的掌控就顯得相當重要。在這些資訊科技產品的供應鍵中，薄膜電晶體液晶平面顯示器(Thin Film Transistor - Liquid Crystal Display, TFT LCD)產業扮演著重要的角色，如何有效的確保面板產能供給與品質良率已是相當重要的課題。本研究以TFT LCD的彩色濾光片(Color Filter, CF)進行研究，目前在CF生產線的後段檢查點仍舊是以人工檢查作業為主，人工檢查方式有著不確定的因素存在，因此利用自化檢測方式是有其必要性。 本研究以影像處理與資料探勘技術進行彩色濾光片瑕疵之自動分類。在影像處理階段分別處理彩色濾光片的穿透光影像與反射光影像，找出影像上有瑕疵的區域，提取其影像特徵屬性並組成特徵向量。在自動分類階段，基於資料探勘的方法，從訓練樣本中找出16種瑕疵類型的特徵向量，再與測試樣本的特徵向量計算相似度，並依此找出測試樣本的分類結果。 最後，實驗結果分析使用準確率(Precision)與召回率(Recall )來進行驗證，所得到的實驗數據顯示，本方法具有高度的準確性。
並列摘要
Nowadays, life cycles of electronic products, such as televisions, tablet computers, and smart phones, have become shorter and shorter, due to the renewed and fast developed information technologies. Therefore, it is important to control the productivity and quality of these products. The industry of TFT LCD plays an important role in the supply chain of these IT products. Therefore, it is a challenge to keep high productivity and high quality of TFT LCD for the flat display industry. In the quality control process, the last checking of color filter (CF) of TFT LCD are still mainly performed manually. However, there are uncertain factors in the manual checking process, so it is needed to be replaced by an automatic checking process. In this research, we developed a method of automatic defect classification of CF by image processing and data mining techniques. In the stage of image processing, we processed the transparent-light images and reflective-light images, detecting the defect areas, extracting the image features, and then formed a feature vector of the defect. In the stage of automatic classification, based on data mining techniques, we discovered the feature vectors of the 16 defect classes from training images, then we calculated the similarity of the feature vectors of testing images to get the results of classification. Finally, we evaluated the method from the experiment results by calculating the precision and recall. The experiment results indicate that our method has average high precision and high recall.
參考文獻
延伸閱讀
- 陳暎仁、簡禎富、黃馨滿（2016）。建構彩色濾光膜及微透鏡缺陷樣型分析之資料挖礦架構。資訊管理學報，23(2)，129-153。https://www.airitilibrary.com/Article/Detail?DocID=16085752-201604-201605110007-201605110007-129-153
- 廖育昇（2008）。利用紋理濾除方式於彩色濾光片瑕疵檢測〔碩士論文，國立中央大學〕。華藝線上圖書館。https://www.airitilibrary.com/Article/Detail?DocID=U0031-0207200917354011
- 梁榮隆、鄭淵明（2012）。彩色濾光片之缺陷於光學檢查機檢出最佳化研究。機械技師學刊，5(4)，33-38。https://doi.org/10.6208/JPME-2012-04-06
- 范祖源（2015）。整合相似度匹配功能之彩色濾光膜缺陷分類模式與實證研究〔碩士論文，國立清華大學〕。華藝線上圖書館。https://doi.org/10.6843/NTHU.2015.00139
- 黃文鴻（2008）。Two Efficient Demosaicking Algorithms for Color Filter Array〔碩士論文，國立清華大學〕。華藝線上圖書館。https://www.airitilibrary.com/Article/Detail?DocID=U0016-1410200814313010