登录    |    注册

您好,欢迎来到中国测试科技资讯平台!

首页> 《中国测试》期刊 >本期导读>改进ResNet101网络下渣出钢状态识别研究

改进ResNet101网络下渣出钢状态识别研究

198    2020-11-24

¥0.50

全文售价

作者:李爱莲, 刘浩楠, 郭志斌, 解韶峰, 崔桂梅

作者单位:内蒙古科技大学,内蒙古 包头 014010


关键词:图像识别;下渣检测;空间金字塔池化;ResNet


摘要:

针对人工目测法缺乏严格的科学依据,且受主观因素影响较大,而红外下渣检测方法需对红外图像进行去噪等预处理,易造成图像失真,并且无法提取图像颜色等基本特征,这些缺点造成最终检测准确率的降低;将空间金字塔池化与ResNet101网络相结合,提出一种基于改进ResNet101网络下渣彩色图像的出钢状态分类识别检测方法,首先将下渣检测图像结合专家经验进行分类,将数据集分为训练数据集、交叉验证数据集与测试数据集;其次,分析ResNet101网络模型,加入空间金字塔池化层(SPP-Net),便于任意大小图片的输入;最后采用Softmax分类器进行图像识别实验,钢水下渣图像识别准确率达到99%以上,能够准确实现区分所有下渣情况的功能。


Study on the status identification of slag steel under improved ResNet101 network
LI Ailian, LIU Haonan, GUO Zhibin, XIE Shaofeng, CUI Guimei
Inner Mongolia University of Science and Technology, Baotou 014010, China
Abstract: The manual visual inspection method lacks strict scientific basis and is greatly affected by subjective factors. However, the infrared slag detection method needs to carry out denoising and other pretreatment on the infrared image, which is easy to cause image distortion and unable to extract basic features such as image color. Combined with the spatial pyramid pooling and the ResNet101 network, a method for the classification and detection of the tapping state based on the improved ResNet101 network bottom slag color image was proposed. Firstly, the slag tapping detection image was classified by combining with the expert experience, and the data set was divided into training data set, cross validation data set and test data set. Secondly, the network model of ResNet101 was analyzed and SPP-Net was added to facilitate the input of images of any size. Finally, softmax classifier was used for image recognition experiment, and the image recognition accuracy of molten steel slag reached more than 99%, which can accurately realize the function of distinguishing all slag conditions.
Keywords: image recognition;slag detection;SPP-Net;ResNet
2020, 46(11):116-119,125  收稿日期: 2020-03-07;收到修改稿日期: 2020-04-20
基金项目: 国家自然科学项目(61763039)
作者简介: 李爱莲(1973-),女,河北唐山市人,副教授,硕士,主要从事复杂过程建模、优化控制研究
参考文献
[1] 郑良峰, 潘华国, 王俊国. 杭钢转炉DM7400下渣检测系[J]. 浙江冶金, 2009(4): 16-17
[2] JULIUS E, THEISSEN W, BKOCK F R. Function and application of electromagnetic slag detection system[J]. World Steel and Metalworking, 1987, 107(9): 56
[3] 李培玉, 甘涛, 沈国振, 等. 基于红外热成像原理的嵌入式下渣检测方法[J]. 钢铁研究学报, 2010, 22(7): 59
[4] 陈灵光, 杨晓江, 李培玉, 等. 一种改进的转炉出钢下渣检测方法及其应用[J]. 钢铁研究学报, 2012, 24(11): 56-62
[5] RAU H, ROP V. Application of thermographic slag detection[J]. Ingo MPT Metallurgical Plant and Technology International: Germany, 2002, 25(1): 52
[6] 陈灵光, 李培玉, 孙大成, 等. 基于相位一致性的转炉出钢下渣检测方法[J]. 浙江大学学报(工学版), 2013, 47(2): 216-222
[7] 何伟. 转炉出钢下渣检测方法与挡渣设备的研究[D]. 哈尔滨: 哈尔滨工业大学, 2016.
[8] 姚永宽, 马军. 连铸稀土钢结晶器保护渣的研制[J]. 江苏冶金, 2001, 29(5): 15-22
[9] HE K, ZHANG X, REN S, et al. Deep residual learning for image recognition[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2016.
[10] HINTON G E, SRIVASTAVA N, KRIZHEVSKY A, et al. Improving neural networks by preventing co-adaptation of feature detectors[J]. Computer Science, 2012, 3(4): 212-223
[11] 张博, 张苗辉, 陈运忠, 等. 基于空间金字塔池化和深度卷积神经网络的作物害虫识别[J]. 农业工程学报, 2019, 35(19): 209
[12] 雷丽莹. 基于深度学习的视频异常检测[D]. 杭州: 杭州电子科技大学, 2018.
[13] 梁万杰, 曹宏鑫, 等. 基于卷积神经网络的水稻虫害识别[J]. 江苏农业科学, 2017, 45(20): 241-243, 253
[14] 程曦, 吴云志, 张友华, 等. 基于深度卷积神经网络的储粮害虫图像识别[J]. 中国农学通报, 2018, 34(1): 154-158
[15] BAKER N, LU H, ERLIKHMAN G, et al. Deep convolutional networksdo not classify based on global object shape[J]. PLOS Computational Biology, 2018, 14(12): 1006613
[16] KRIZHEVSKY A, SUTSKEVER I, HINTON G E. Image Net classification with deep convolutional neural networks[J]. Communications of the ACM, 2017, 6(60): 84-90
[17] HE K M, ZHANG X Y, REN S P, et al. Deep residual learning for image recognition[C]//IEEE Conference on Computer Vision and Pattern Recognition(CVPR 2016), 2016.