In this paper, a novel wear particle online images multi-label classification based on semantic segmentation (WPC-SS) is proposed. In this model, both semantic labels and class labels are applied to guide network training, which make the regions with wear particles attain more attention during the process of network training. It solves the problem that it is difficult to classify the small wear particles when they are compared with the background in the online image. In addition, chain channel attention and class attention unit are added to optimize the network to improve the recognition accuracy. The important channels of the network are monitored by chain channel attention, so that the extracted features can be better prepared for the subsequent classification work. Class attention unit can refine the segmentation results and further optimize the classification results. Comparison experiments are executed with the standard image classification method (multi-CNN). The results of experiments show that WPC-SS model surpasses the standard image classification methods in solving the problem of multi-label classification of online wear particle images.

在本文中，提出了一种基于语义分割的新型磨损粒子在线图像多标签分类（WPC-SS）。在该模型中，语义标签和类标签都被用来指导网络训练，使得有磨损颗粒的区域在网络训练过程中得到更多的关注。解决了在线图像中小磨损颗粒与背景对比时难以分类的问题。此外，还增加了chain channel attention和class attention unit来优化网络，提高识别准确率。通过chain channel attention对网络的重要通道进行监控，使得提取的特征可以更好地为后续的分类工作做准备。类注意力单元可以细化分割结果，进一步优化分类结果。使用标准图像分类方法（multi-CNN）执行比较实验。实验结果表明，WPC-SS模型在解决在线磨损颗粒图像的多标签分类问题上优于标准图像分类方法。