請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/90723
標題: | 電腦視覺輔助面板光學檢測 Computer Vision Assisted Optical Inspection at Panel Displays |
作者: | 陳平軒 Ping-Xuan Chen |
指導教授: | 陳中平 Chung-Ping Chen |
關鍵字: | 電腦視覺,光學檢測,MURA,影像校正, Computer Vision,Optical Inspection,MURA,Distortion Correction, |
出版年 : | 2023 |
學位: | 碩士 |
摘要: | 在 2019 年時我國總人口數達最高峰 2,360 萬人,至今高齡人口逐漸攀升出生人數逐漸下降,工作人口日益減少。在面板產業,長時間持續監控面板生產品質是必要的,目前主要監控方式由工人定時檢查,但受限人力分配,導致檢查頻率無法提升,也難以保證檢查品質的一致性,因此高可靠度且長時間穩定持續監控的系統開發有其必要性,這系統可以降低工廠人員的人力需求,提供可靠的檢測品質。
合作廠商根據目前工廠的情況提出四個待解決的問題,1. 影像校正技術,2.影像計算加速技術,3. 影像辨識技術,4. 影像瑕疵檢測技術。針對四個問題,我們團隊應用電腦視覺中的演算法,結合特定面板影像進行處理。根據我們的研究結果,在影像校正技術方面可以達到 99.5% 以上的準確率; 在影像計算加速技術方面,掌握了 CPU 與 GPU 分配計算的數據; 在影像辨識技術上,能夠辨識目標影像,但運算速度還需要進一步提升; 對於影像瑕疵,目前能夠凸顯瑕疵特徵,明顯的特徵能夠檢測出瑕疵。 現今隨著機器視覺影像演算法與設備技術的突破,自動化檢測系統的開發將不再是夢想,分擔勞力的自動化設備已用於現實或許就在不久的將來。 In 2019, the total population of Taiwan reached its peak of 23.6 million people. Since then, the elderly population has been gradually increasing, the number of births has been gradually decreasing, and the working population has been decreasing. Continuous monitoring of panel production quality is necessary for the display panel industry. The current primary monitoring method is for workers to check regularly. However, due to human resources allocation, the inspection frequency cannot be increased, and it isn't easy to ensure the consistency of inspection quality. Therefore, developing a system with high reliability and long-term stable continuous monitoring is necessary. This system can reduce the workforce demand of factory personnel and provide reliable detection quality. Cooperative manufacturers have proposed four problems to be solved based on the current situation of the factory: 1. Image correction technology, 2. Image computation acceleration technology, 3. Image recognition technology, 4. Image defect detection technology. In response to these four problems, our team applies algorithms in computer vision and processes specific panel images. According to our research results, the image correction technology can achieve an accuracy rate of more than 99.5%; in the image computation acceleration technology, we measured the time required for CPU and GPU computation; in the image recognition technology, we can recognize the target image, but the operation speed needs to be further improved; for image defects, we can currently highlight defect features, and prominent features can detect defects. Nowadays, with the breakthrough of machine vision image algorithms and equipment technology, the development of automatic detection systems will no longer be a dream. Automation equipment to share labor has been used in reality and maybe soon. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/90723 |
DOI: | 10.6342/NTU202300857 |
全文授權: | 未授權 |
顯示於系所單位: | 電子工程學研究所 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-111-2.pdf 目前未授權公開取用 | 17.39 MB | Adobe PDF |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。