基于深层迁移学习的DR胸片肺结核病灶检测

针对基于传统机器学习方法设计的DR胸片肺结核检测器存在着泛化能力不强,实际检测精度低等问题,提出了一种基于Focal Loss的深度学习检测方法Tuberculosis Neural Net(TBNN).医学图像的特殊性,存在带标注的数据量小导致无法充分训练深层网络模型等问题.该方法利用肺炎和肺结核同为呼吸道感染疾病且在DR胸片上有相似表征的特点,基于迁移学习原理训练特征提取子网络,减少肺结核胸片样本不足对模型训练造成的影响.首先在大型的肺炎胸片数据集上训练特征提取网络,以获取DR图像中丰富的深层图像语义信息,然后使用样本较少的肺结核数据集微调网络参数,并将多层卷积的输出作为TBNN分类子网络的输入,得到基于DR胸片的肺结核病灶检测模型.实验结果表明,该方法生成的检测模型在分类精度和性能上均优于基于传统机器学习的肺结核检测器.在同等训练数据量和训练周期下,模型性能高于其他采用传统数据增强方法的深层网络肺结核检测算法,且能标识病灶区域,准度上有不低于放射科阅片医生的表现.     

高瓦斯特长高铁隧道通风方案设计及监测分析

成都至自贡高铁白云山隧道全长13 340 m,为特长高瓦斯隧道,多工区、多阶段施工通风问题严重制约着隧道运营施工与人员安全。通过工作面最大通风量、最大供风量及沿程风压损失对不同工区施工阶段进行通风计算,设计单工区通风方案及隧道总体贯通顺序,并采用水气分离方法处理地下水及自动安全监控系统实时监测隧道瓦斯浓度。分析表明：施工的大部分时间内,2#～6#斜井的瓦斯浓度均偏低,说明合理的通风方案及工区贯通顺序可以有效降低隧道瓦斯浓度;采用水气分离方法对地下水进行处理,将地下水和瓦斯分别排出,可进一步降低隧道内瓦斯浓度;监控系统对2#、3#和6#斜井出现的瓦斯排放异常情况及时报警提示并断电保护,保证了隧道施工进度及人员安全。针对特长高瓦斯隧道多工区、多阶段施工通风,基于隧道通风方案、贯通顺序、水气分离处理及瓦斯监测等方面建立一套完整的通风监测设计方法,为类似工程提供指导。     

Modeling and Assessment of Buoyancy-Driven Stratified Airflow in High-Space Industrial Hall

In industrial environment,heat sources often are contaminant sources and health threatening con-taminants are mainly passive,so a detailed understanding of airflow mode can assist in work environment hy-giene measurement and prevention.This paper presented a numerical investigation of stratified airflow scenario in a high-space industrial hall with validated commercial code and experimentally acquired boundary conditions.Based upon an actually undergoing engineering project,this study investigated the performance of the buoyancy-driven displacement ventilation in a large welding hall with big components manufactured.The results have demonstrated that stratified airflow sustained by thermal buoyancy provides zoning effect in terms of clean and polluted regions except minor stagnant eddy areas.The competition between negative buoyant jets from displace-ment radial diffusers and positive buoyant plume from bulk object constitutes the complex transport characteris-tics under and above stratification interface.Entrainment,downdraft and turbulent eddy motion complicate the upper mixing zone,but the exhaust outlet plays a less important role in the whole field flow.And the corre-sponding suggestions concerning computational stability and convergence,further improvements in modeling and measurements were given.     

新型自动捕风排风装置排风口性能优化

基于自然通风原理,对所研制的自动捕风排风装置的通风性能进行了优化分析.通过数值模拟和实验研究,重点分析了排风口几何形状与其表面自然通风风压系数的关系,以及提高其排风性能的关键参数.研究结果表明,圆锥形渐扩口的综合排风性能为最佳,较无渐扩口的排风口可提高排风量14.3％;所研制的自动捕风排风装置在北京地区冬季典型天利用自...