Abstract: Due to the unparalleled advantage at the quality control of industrial products,intelligent detection of surfacedefect based on vision sensing has attracted ever-increasing attention and been applied in a wide range of industries,such as automotive industry,semiconductor manufacturing,glass fabrication,steel metallurgy.The rapid development of AI learning algorithms and the vision sensing technology has brought new opportunity and challenges to the detection of surface defect.A survey of methodologies and trends in the vision based surface detection was summarized,with special focus on modern image processing,geometric deep learning and deep learning methods for object detection,which could prompt a technological breakthrough to the intelligent detection of surface defect.The applications of industrial image detection were discussed by three typical fields including steel metallurgy,air pollution monitoring and defect detection of aircraft engine.Several challenging issues were envisioned for future research.

Key words: vision sensing, intelligent detection, deep learning, image processing, surface defect detection, few-shot object detection, artificial intelligence

摘要： 由于在工业产品质量监控中不可比拟的优势,基于视觉感知的表面缺陷检测近年来得到了很多研究者的持续关注,且已广泛应用于不同工业领域,包括汽车工业、半导体加工、玻璃制造、钢铁冶金等。AI学习算法与视觉传感技术的飞速发展为表面缺陷检测研究带来了新的机遇与挑战。综述了基于视觉感知的表面检测研究中的主要方法与进展,重点介绍了图像处理、几何深度学习、面向目标检测的深度学习等方向的研究现状,这些研究有望为表面缺陷智能检测技术的发展带来突破。讨论了工业图像检测与识别在钢铁冶金、大气污染监测以及航空发动机缺陷检测3个领域的应用。最后,提出了值得研究的挑战性问题。

关键词: 视觉感知, 智能检测, 深度学习, 图像处理, 表面缺陷检测, 小样本目标检测, 人工智能