基于机器视觉的皮带撕裂检测综述  

A Survey of Belt Tear Detection Algorithms Based on Machine Vision

作　　者：李旭东 廖婷婷 曾小信[1,2] 李曦 李宗平 

机构地区：[1]中冶长天国际工程有限责任公司,湖南 长沙 [2]国家烧结球团装备系统工程技术研究中心,湖南 长沙 [3]中冶长天(长沙)智能科技有限公司,湖南 长沙

出　　处：《计算机科学与应用》2023年第12期2191-2197,共7页Computer Science and Application

摘　　要：本文综述了机器视觉在皮带撕裂检测领域的发展,并讨论了现有的检测方法。首先介绍了皮带撕裂的形式,包括纵向撕裂和横向撕裂。据调研数据显示,纵向撕裂占据了绝大部分的撕裂情况。然后介绍了传统机器视觉方法和基于深度学习的方法。传统方法主要是通过人工设计特征和选择合适的分类器算法进行图像分析和提取特征,但需要大量经验和调试。而基于深度学习的方法通过训练神经网络进行端到端的学习,能够提高检测精度和鲁棒性。同时归纳了近三年来应用到实际生产中的技术。最后,指出了需要解决的算法设计和评价指标问题,并展望了未来的研究方向,包括解决复杂环境的影响、提高实时性要求、优化算法适应不同情况和制作公开测试数据集等。总体来说,机器视觉技术在皮带撕裂检测领域具有广阔的应用前景,但还需要进一步研究和探索。This article reviews the development of machine vision in the field of belt tear detection and dis-cusses existing detection methods. First, the forms of belt tears are introduced, including longitu-dinal tears and transverse tears. According to survey data, longitudinal tears account for the vast majority of tears. Then traditional machine vision methods and deep learning-based methods are introduced. Traditional methods mainly perform image analysis and extract features by manually designing features and selecting appropriate classifier algorithms, but this requires a lot of experience and debugging. Methods based on deep learning can improve detection accuracy and robust-ness by training neural networks for end-to-end learning. At the same time, the technologies ap-plied in actual production in the past three years are summarized. Finally, it points out the algorithm design and evaluation index issues that need to be solved, and looks forward to future research directions, including solving the impact of complex environments, improving real-time re-quirements, optimizing algorithms to adapt to different situations, and creating public test data sets. Overall, machine vision technology has broad application prospects in the field of belt tear de-tection, but further research and exploration are needed.

关 键 词：机器视觉 传统检测方法 深度学习检测方法 皮带 撕裂检测 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]