邱意 陈劲杰



摘 要:由于实际工业中工作池内钢水与表面锌渣存在部分重叠和边界不清以致识别较难的情况,提出了一种基于U-Net(U型神经网络)网络的锌渣识别方法。该方法先是把工业摄像头采集到的工作池图像进行灰度化,均值滤波等多种平滑模糊处理后,再采用完善的U-Net网络进行轮廓提取。接着将所得图像做二值化处理后,通过OpenCV(跨平台计算机视觉库)自带函数获得结果并对其进行分析。实验结果表明,基于U-Net的锌渣识别方法不仅能准确快速地区分钢水与表面锌渣,也能降低人工经验中存在的误差。
关键词:U-Net网络;锌渣识别;图像分割;深度学习
中图分类号:TP391.4 文献标识码:A
文章编号:2096-1472(2021)-01-02-03
Abstract: In industry, the molten steel in working pool and surface slag are partially overlapped and the boundary is unclear, which makes it difficult to visually differentiate them. Aiming at this problem, the paper proposes a zinc slag recognition and segmentation method based on U-Net (U-shaped Neural Network). Firstly, the working pool images collected by industrial camera are smoothed and blurred by grayscale processing and average filtering. Then, perfect U-Net is used for contour extraction. Next, the obtained images are binarized. Finally, the result is obtained and analyzed by the built-in function of OpenCV (Computer Vision). The experimental results show that this zinc slag recognition method based on U-Net can accurately and quickly distinguish molten steel from surface zinc slag, as well as reduce errors in manual experience.
Keywords: U-Net; slag recognition; image segmentation; deep learning
1 引言(Introduction)
目前,眾所周知带钢的耐腐蚀性能需要进行提升的话,工业上很常用的就是在其表面加上一层耐腐蚀的镀层,而热镀是多种手段中最受认可的。在该技术中,不管是热镀什么金属成分一般都用锌锅,具体实施方法是将镀层金属熔融在锌锅内,然后把目标镀件浸没到镀液中进行热镀。在加工过程中,一定会产生锌渣,而这些表面浮渣一旦处理不当,就会对镀件连续热镀锌产生不好的效果,这不仅会降低产品的表面质量,也会造成液面浮渣凝结,以致后续工业生产无法持续进行。因此,在热镀过程中需要对锌锅内的表面浮渣进行捞渣作业,但是捞渣作业的环境非常差,所需的劳动强度又大,容易对工作人员的身体健康造成不良的影响。可见人工捞渣既耗费很多人力物力,准确性和效率又不尽如人意,需要现代先进的解决办法。……