摘要： 在转炉炼钢吹炼过程中,要求对转炉终点做出准确且实时的判断。为提升转炉终点判断的准确率,提出一种基于火焰图像卷积神经网络识别建模的转炉炼钢吹炼终点判断方法。利用卷积神经网络自行从样本图像中分层递阶地学习相应特征,减少或避免人工经验的误导,从而实现转炉终点判断准确度的提升。将火焰图像在HSI空间下采用最大类间方差法进行分割,寻找出模型最佳参数,并在5个炉次的火焰数据上验证算法性能。实验结果表明,与灰度共生矩阵和灰度差分统计方法相比,该方法识别率分别提升29%和4%,模型准确性与实时性较高,可应用在实际转炉炼钢终点判断中。

关键词: 转炉炼钢, 最大类间方差法, 卷积神经网络, 有监督式训练, 梯度下降法

Abstract: In the process of basic oxygen furnace blowing,it is required to make accurate and real-time judgment of the furnace endpoint.In order to enhance the accuracy of blowing endpoint judgment,a method for basic oxygen furnace endpoint judgment based on flame image Convolution Neural Network(CNN) recognization modelling is proposed.The related features are learned by CNN from the sample image,avoiding misleading from hand-crafted features,so as to realize the improvement in the accuracy of endpoint judgment.The OSTU algorithm is used for flame image segmentation in HSI color space to find out the best parameters of the model.The performance of the algorithm is verified with the flame data of five furnaces.Experimental results show that the recognition rate is respectively improved by 29% and 4% compared with that of the gray level co-occurrence matrix and gray scale difference statistics method.Therefore,the proposed model meets the real-time needs with high accuracy and can be used for actual basic oxygen furnace endpoint judgment.

Key words: basic oxygen furnace, Otsu method, Convolution Neural Network(CNN), supervised training, gradient descent method

中图分类号: 

• TP391.41