基于机器视觉和深度学习的材料缺陷检测应用综述

Survey of Applications of Material Defect Detection Based on Machine Vision and Deep Learning

因材料缺陷而导致的安全问题一直是备受人们关注的热点问题,如何实现对材料缺陷的快速准确识别与定位是当今材料缺陷问题的研究重点。传统的无损检测方法主要通过超声波、X射线等实现对材料缺陷的识别与定位,这种方法虽然解决了人工检测效率低等问题,但是难以满足智能化、自动化与高精度的多重要求。计算机领域的进步刺激了机器视觉在材料缺陷检测方面的飞速发展,机器视觉检测技术的特点主要是将无损检测、自动化与智能化相结合,不仅安全性好、效率高,而且检测精度高。但是大多数机器视觉设备存在对不同缺陷识别任务需要不同图像处理算法的问题,意味着设备的通用性低,而且设备的生产与维护的成本也较高。近年来,深度学习的崛起推动人工智能领域迅猛发展,并且解决了传统的机器视觉对不同任务进行分类需要不同图像处理算法的问题,深度学习也成为材料缺陷检测方面的一个热门研究方向,已经有很多学者将深度学习技术应用到材料缺陷检测方向,而且无论是对材料的缺陷类别进行分类,还是对材料的缺陷进行定位与分割,相关研究都取得了不错的成果。本文首先对传统方法和机器视觉方法在材料缺陷方面的应用进行了概述与分析,叙述了深度学习在材料缺陷检测中的原理,然后分别系统地阐述了国内外深度学习在缺陷的分类、定位以及分割技术方面最新的应用,并对深度学习在材料缺陷检测领域应用的未来发展趋势进行了展望。

The safety problem caused by material defects has always been a hot issue,and how to realize the rapid accurate identification and locationof material defects is the focus of current research on material defects. The traditional non-destructive testing method mainly uses ultrasonic,X-ray and other advanced technologies to realize the identification and positioning of material defects. Although this method solves the problem of low manual testing efficiency,it is still difficult to achieve the requirements of intelligentization,automation and high precision.Advances in the computer technology has stimulated the rapid development of machine vision in the detection of material defects. The advantages of machine vision inspection technology are mainly the combination of non-destructive inspection,automation and intelligentization,which has not only good safety and high efficiency,but also high detection accuracy. However,an image processing algorithm usually can be effective to the recognition of only one specific kind of defect,which causes low versatility of the equipment and elevated cost of production and maintenance.In recent years,the rise of deep learning has promoted the rapid development of the field of artificial intelligence,and solved the problem that traditional machine vision needs different image processing algorithms to classify different tasks. Deep learning has also become a popular research direction in material defect detection. Many scholars have applied deep learning technology to the field of material defect detection,and they have achieved good results whether they are related to the classification of material defects,or the location and the segmentation of material defects.The article summarizes the application of traditional methods and machine vision methods in material defects,introduces the principles of deep learning in material defect detection,and systematically describes the application of deep learning in the classification,positioning and segmentation of material defects at home and abroad,and the future development trend of the application of deep learning in the field of material defect detection is prospected.