摘要
摘要 近年來，隨著環保意識的抬頭，因應節能減碳的需求，在燈具上帶動了發光二極體(Light Emitting Diode, LED)的熱潮，也因應此熱潮的趨勢，使得全球對於LED的需求量大幅度的增加，而如何提升LED良率，於各領域的應用上就顯得格外重要，雖然製程流程都在無塵室內進行，廠區也將外在環境的影響降至最低，但製程中仍會因為異物的掉落或是機台校正問題等因素，導致LED晶粒出現缺陷，這對LED廠商而言即是成本的消耗，因此若能在LED製程中即時發現錯誤並加以補救或是停止，即可提升LED之良率與降低廠商的成本消耗。 而LED的品質由電氣特性及表面完整性來決定，所以LED的表面瑕疵檢驗是相當重要的。為了提升LED晶粒良率，多數的公司都設有檢查部門，當製程進行至某一個階段時，以人工進行瑕疵分類工作，但長時間的檢測，會增加檢測人員的疲勞度，進而提高了晶粒誤判的可能性。為了降低人為誤判及縮短檢測的處理時間，以達到全檢的目標，將整個檢測系統自動化是有必要的。 本論文的目標即是發展一個自動化瑕疵檢測系統，於低解析度的晶粒影像上，融合各種數位影像處理技巧，進行LED晶粒外觀瑕疵檢測及分類。此系統主要檢測項目有：晶粒歪斜瑕疵、電極區(Pad)瑕疵、Mesa線斷線瑕疵及發光區瑕疵；若嚴重瑕疵者會直接影響晶粒的電性，使得晶粒短路進而無法運作，所以瑕疵晶粒需由系統全部挑出並做分類，以維護或提升LED之良率。 經由最後的實驗結果可知，本論文輸入了1539張單顆晶粒影像於系統中，瑕疵辨識率高達97%以上，表示此系統對於上述四個瑕疵項目檢測可以確實地將瑕疵晶粒影像分類出來，並做淘汰，於檢測速度方面，單顆晶粒影像之檢測速度為0.18秒，代表系統能快速的分辨瑕疵。
並列摘要
Abstract In recent years, environment awareness issue in respect to energy saving and carbon reduction of the Light Emitting Diode (LED) effect is arising. This trend makes global demand of LED is increasing and how to enhance the yields of LED play a particularly important role on each area. Although the fabrications process in the clean room and factories decrease the impact of external environment, but other objects falling or machine calibration or other factors led to result LED grain defects. This will result in cost consumption of LED manufactures. In addition, the major problem is fabrication. Furthermore, if we can find the error in the fabrication in situ and make compensations or stops, we can enhance the yields of LED and reduce the cost consumptions of manufactures. The quality of LED is determined by electrical characteristic and surface roughness, so the LED surface defect inspections are very important. In order to increase the yields of LED grains, most companies have inspection department, but the process is manual. Manual inspection needs long time, and cause increasing of fatigue of workers, hence enhancing the possibility of wrong judgment. In order to decrease wrong judgment and increase the processing time of detection as well as achieve the objectives, automation inspection system is necessary. The purpose of this thesis is to develop an automation of defect inspection system for low-resolution grains image and integrate with digital image processing technique for defect inspection and classification of LED grains. The grains defect inspection for this system is majored in grains crooked defect, pad defect, mesa disconnection defect and emitting area defect. If there is serious defect that directly influence electrical characteristic or making grains short circuit thus does not work, all defect grains of system must indicate to maintain or increasing the yields of LED. According to the experiment results shows that the defect recognition rate is 97% which has 1539 single grain images in this system. It shows that above-mentioned four defect testing items can classify and eliminate the defect grain image truly. Regarding the speed of detection, the real time of single grain image detection is 0.18 second that means this system can quickly distinguish defect.