基于VGG19-SSGAN的加热炉内钢坯定位算法研究  

Research on Positioning Algorithm of A Novel Billet in Reheating Furnace Based on VGG19-SSGAN

作　　者：张悦 张昊宇 张娜 ZHANG Yue;ZHANG HaoYu;ZHANG Na(School of Automation,Shenyang Institute of Engineering,Shenyang 110136;State Grid Shenyang Power Supply Company,Shenyang 110000,Liaoning Province)

机构地区：[1]沈阳工程学院自动化学院,辽宁沈阳110136 [2]国家电网沈阳供电公司,辽宁沈阳110000

出　　处：《沈阳工程学院学报(自然科学版)》2025年第2期63-67,83,共6页Journal of Shenyang Institute of Engineering:Natural Science

基　　金：辽宁省教育厅项目(LJKMZ20221719)。

摘　　要：针对加热炉出钢过程中准确定位钢坯位置难的问题,提出一种VGG19-SSGAN的钢坯定位算法。首先,应用VGG19网络对钢坯在炉内图像数据进行特征提取,并将提取后的特征保存,用于后续网络的输入;其次,将提取的特征应用半监督GAN的图像识别网络架构进行半监督训练,可以有效处理人工对图像数据打标签工作量大的问题;最后,将该方法用于实际生产中加热炉内摄像头的数据采集,与核SVM、标准全连接神经网络和标签传播算法进行对比,研究表明所提出的钢坯定位方法在加热炉出钢中有更良好的定位效果。Aiming at the problem that difficult to accurately locate the billet position in the steel discharge process of the reheating furnace,a novel billet location algorithm named VGG19-GAN is proposed.Firstly,VGG19 network is used to extract features from the image data of billet in the furnace,and the extracted features are prepared for the subsequent network.Secondly,considering the heavy workload of manual labeling of image data,the semi-supervised GAN based on image recognition network architecture is applied to train the features extracted from VGG19 network.Batch normalization layer is embedded into the traditional tanh activation function,which can reduce the gradient vanishing.Finally,the method is applied to the data collected by the camera in actual production furnace for modeling and testing.Compared with the kernel SVM,standard fully connected neural network and label propagation algorithm,the proposed method shows more accurate billet positioning effect.

关 键 词：加热炉 钢坯定位 半监督GAN 图像特征提取 

分 类 号：TP6/8[自动化与计算机技术—控制理论与控制工程]