Computer Engineering and Applications ›› 2021, Vol. 57 ›› Issue (16): 269-275.DOI: 10.3778/j.issn.1002-8331.2005-0142

Previous Articles     Next Articles

Multi-class Deep Convolutional Generative Adversarial Networks for Belt Tear Detection

MENG Xiaojuan, ZHANG Yueqin, HAO Xiaoli, LYU Jinlai   

  1. College of Information and Computer, Taiyuan University of Technology, Jinzhong, Shanxi 030600, China
  • Online:2021-08-15 Published:2021-08-16



  1. 太原理工大学 信息与计算机学院,山西 晋中 030600


Belt tearing is one of the most common faults of belt conveyor, which directly affects the safe and stable operation of belt conveyor. In view of the fact that most of the existing methods only detect one type of damage, a belt tear detection method based on two time-scale update rule and multi-class deep convolutional generative adversarial network is designed. The image of the belt surface is captured by CCD camera and transmitted to the decision subsystem by the data transmission subsystem. In the processing module of the decision subsystem, by removing the batch normalization operation of the generator, the damage location and type can be quickly obtained by multi-class deep convolutional generative adversarial network. The introduction of two time-scale update rule makes the model converge faster. The experimental results show that the mean average precision on the MS COCO data set is 95.7%, the mean average precision is 96.9% on the belt image data set.

Key words: belt conveyor, two time-scale update rule, multi-class, deep convolutional generative adversarial networks, belt tear detection


皮带撕裂是皮带机出现的最常见故障之一,直接影响皮带机的安全稳定运行。针对现有的方法大多仅对一种破损类型进行检测的情况,设计了一种基于双时间尺度的多分类深度卷积生成对抗网络的皮带撕裂检测方法。利用CCD相机捕获皮带表面图像,并经数据传输子系统将图像传送到决策子系统;在决策子系统的处理模块,通过去掉生成器的批量归一化操作,由多分类深度卷积生成对抗网络快速得到破损位置和类型;引入双时间尺度更新规则使得模型更快地收敛。实验结果表明,在MS COCO数据集上,多类别平均精确率为95.7%;在皮带图像数据集上,多类别平均精确率为96.9%。

关键词: 皮带机, 双时间尺度更新规则, 多分类, 深度卷积生成对抗网络, 皮带撕裂检测