The presence of pixel defects on the screens of LCD-based products (TV, tablet, phone, etc.) is unacceptable given the consumer expectations. Therefore, these defects should be detected before the product reaches the user during the production stage. Visual inspections are mostly performed by human operators in the production. These inspections are error prone and not efficient in terms of consumed time. For this reason, computer visionbased approaches are started to find applications in this kind of problems. This paper presents an image acquisition system and a detailed analysis of deep learningbased object detectors for LCD pixel defect detection problem. Experimental results show that the proposed methods can be a powerful alternative to operator control by providing more efficient use of time, human, financial resources and betterquality standards in TV production industry.

鉴于消费者的期望，基于LCD的产品（电视，平板电脑，电话等）的屏幕上存在像素缺陷是不可接受的。因此，在生产阶段产品到达用户之前，应该检测出这些缺陷。目视检查主要由生产过程中的人工操作人员执行。这些检查容易出错，并且在消耗的时间方面效率不高。因此，基于计算机视觉的方法开始在此类问题中找到应用。本文提出了一种图像采集系统，并针对基于深度学习的目标检测器对LCD像素缺陷检测问题进行了详细分析。实验结果表明，通过更有效地利用时间，人员，