基于机器视觉的表面缺陷检测方法研究进展

Review of surface defect detection methods based on machine vision

基于机器视觉的表面缺陷检测以无接触、无损伤、自动化程度高及安全可靠等突出优点被广泛应用于各种工业场景中,尤其随着深度学习技术的快速发展,视觉缺陷检测有助于提高产品及装备的智能化水平。综述分析了表面缺陷检测的常用方法、通用数据集、检测结果评价指标和现阶段面临的关键问题。首先,将缺陷检测方法分为传统基于图像处理的缺陷检测、基于传统机器学习模型的缺陷检测及基于深度学习的缺陷检测,并对各种方法进一步细分归类和对比分析,总结了每种方法的优缺点和适用场景;然后,对目前常用的缺陷检测结果评价方法做出了描述,进一步探讨了表面缺陷检测应用在实际工业产品检测过程中关键问题——小样本问题,重点剖析了小样本问题的解决方法和无监督学习在解决这类问题上的优势;最后,从提高缺陷检测方法的工业适用性角度展望了下一步研究方向。

Surface defect detection based on machine vision is widely used in various industrial scenarios.It has outstanding advantages such as no contact,no damage,high automation,safety and reliability,etc.Especially with the rapid development of deep learning technology,visual defect detection helps to improve the intelligence level of products and equipment.The review analyzes the common methods of surface defect detection,common data sets,evaluation indexes of detection results and key issues faced at this stage.First,the defect detection methods are divided into based on traditional image processing,traditional machine learning model and deep learning.These methods are further subdivided into categories and comparative analysis,summarizing the advantages,disadvantages and applicable scenarios of each method.Then,a description of the current commonly used methods for evaluating defect detection results is given.The key problem of surface defect detection applied in the process of actual industrial product inspection,the small sample problem,is further discussed.The solution of the small sample problem and the advantages of unsupervised learning in solving it are focused on.Finally,the next research directions are prospected in terms of improving the industrial applicability of the defect detection methods.