深度学习的煤矿尘肺病图像识别  

Image recognition of pneumoconiosis of miners based on deep learning

作　　者：刘丹丹 刘玉秋 李德文 郭胜均 汤春瑞 Liu Dandan;Liu Yuqiu;Li Dewen;Guo Shengjun;Tang Chunrui(School of Electrical&Control Engineering,Heilongjiang University of Science&Technology,Harbin 150022,China;China Coal Technology Engineering Group Chongqing Research Institute,Chongqing 400037,China)

机构地区：[1]黑龙江科技大学电气与控制工程学院,哈尔滨150022 [2]中煤科工集团重庆研究院有限公司,重庆400037

出　　处：《黑龙江科技大学学报》2024年第3期457-462,共6页Journal of Heilongjiang University of Science And Technology

基　　金：国家重点研发计划项目(2017YFC0805208)。

摘　　要：针对尘肺病人工诊断率低和诊断误差大等问题,通过深度学习利用神经网络模型对煤矿尘肺病CT图进行辅助诊断,通过比较现有的ResNet原始模型、DenseNet模型、MC-CNN模型训练结果,提出一种基于改进ResNet模型的图像分类方法。将GN正则化嵌入到模型中,将残差网络模型中常规卷积convolution更改为内卷involution。结果表明,ResNet101模型的整体性更强,其准确率达到93.2%,精确率达93.8%,召回率达到93.6%,F 1达到93.7%。该研究融合深度学习算法与正则化的优势验证图像识别模型是可行的。This paper is aimed at addressing the low manual diagnosis rate and big diagnosis error of pneumoconiosis in mine.The study consists of using neural network model to assist in diagnosis of pneumoconiosis CT map;proposing an image classification method based on the improved ResNet model by comparing the existing ResNet original model,DenseNet model and MC-CNN model training result;embeding GN regularization into the model to change the regular convolution to involution in the residual network model.The experimental results show that the ResNet101 model has a stronger integrity,with the accuracy rate of 93.2%and the accuracy rate of 93.8%,the recall rate of 93.6%,and the F 1 of 93.7%.This image recognition model verified by using deep learning algorithms and fusing the advantages of regularization is feasible.

关 键 词：煤矿 深度学习 残差网络 正则化 

分 类 号：TD714.1[矿业工程—矿井通风与安全] TP391.41[自动化与计算机技术—计算机应用技术]