基于多特征融合的细粒度视频人物关系抽取

视频人物关系抽取是信息抽取问题中的重要任务,在视频描述、视频检索,以及人物搜索、公安监察等方面具有重要价值。由于视频数据的底层像素与高层关系语义之间存在巨大的鸿沟,现有方法很难准确地抽取人物关系。现有研究大多通过粗粒度地分析人物共现等因素来抽取人物关系,忽略了具有丰富语义的视频中的细粒度信息。为解决现有算法难以准确、完整地抽取视频人物关系的问题,文中提出了一种基于多特征融合的细粒度视频人物关系抽取方法。首先,为了准确识别视频人物实体,提出了一种基于多特征融合的人物实体识别模型;然后,提出了一种基于细粒度特征的人物关系识别模型,该模型不仅融合了视频中人物的时空特征,而且考虑了与人物相关的细粒度物体信息特征,从而建立更好的映射关系来准确识别人物关系。以电影视频数据和SRIV人物关系识别数据集为实验数据,实验结果验证了该模型的有效性和准确性,与现有同类模型相比,所提模型的人物实体识别F 1值提高了约14.4%,人物关系识别的准确率提高了约10.1%。     

基于BCNN的胎儿颅脑超声横切面识别算法

孕期超声检查是评估胎儿大脑发育、检测生长异常的重要步骤,开展对胎儿早期检查准确高效的诊断研究具有重要的临床价值。文中使用双线性卷积神经网络BCNN进行胎儿颅脑横切面识别,提出了BCNN-R和BCNN-S两种算法。BCNN算法首先对输入的胎儿颅脑超声影像数据进行预处理,去除个人信息等敏感信息;其次,利用两路并行的子网络从影像数据中提取辨识度高、鲁棒性强的横切面特征,并将其融合得到有助于识别的细微特征;最后使用线性连接层进行识别和分类。为了验证算法的有效性,在自建胎儿超声数据集JFU19上进行了对比实验,实验结果表明,所提算法相比常见的深度网络(GoogleNet,DenseNet,SeNet等)在分类性能上有明显的提升,其中BCNN-S算法的总体准确率达到了88.95%,BCNN-R在水平横切面的识别上达到了97.22%的精确度和88.61%的召回率。此外,在公开数据集HC18上进行了实验,BCNN算法的准确率、精确度、召回率分别达到了89.48%,87.66%和87.71%,进一步验证了算法的有效性。     

深度卷积神经网络在心音分类方法中的应用

通过分析心音信号对心脏早期的病理状态进行确诊具有重要的意义。提出了一种基于深度卷积神经网络的心音分类方法。将心音信号转化成具有时频特性的梅尔频谱系数（Mel Frequency Spectral Coefficient,MFSC）特征图,将其作为深度卷积神经网络模型的输入;利用深度卷积神经网络对MFSC特征图进行训练,引入中心损失函数建立最优的深度学习模型;测试阶段,先将心音信号转换成多张二维MFSC特征图,然后利用训练好的深度学习模型对其分类,最后利用多数表决原则判断心音信号的类别。针对人工标注的训练样本有限,导致模型训练正确率不高的问题,以心音的二维MFSC特征图为对象分别从时间域和频率域进行随机屏蔽处理进而扩充训练样本。实验结果表明,该方法在PASCAL心音数据集上进行测试,对正常、杂音、早搏三种心音的分类性能明显优于现有最好的方法。     

基于随机方差减小方法的DDPG算法

针对深度确定性策略梯度算法（DDPG）收敛速度比较慢,训练不稳定,方差过大,样本应用效率低的问题,提出了一种基于随机方差减小梯度方法的深度确定性策略梯度算法（SVR-DDPG）。该算法通过利用随机方差减小梯度技术（SVRG）提出一种新的创新优化策略,将之运用到DDPG算法之中,在DDPG算法的参数更新过程中,加入了随机方差减小梯度技术,利用该方法的更新方式,使得估计的梯度方差有一个不断减小的上界,令方差不断缩小,从而在小的随机训练子集的基础上找到更加精确的梯度方向,以此来解决了由近似梯度估计误差引发的问题,加快了算法的收敛速度。将SVR-DDPG算法以及DDPG算法应用于Pendulum和Mountain Car问题,实验结果表明,SVR-DDPG算法具有比原算法更快的收敛速度,更好的稳定性,以此证明了算法的有效性。     

Differentially private data fusion and deep learning Framework for Cyber–Physical–Social Systems: State-of-the-art and perspectives

The modern technological advancement influences the growth of the cyber–physical system and cyber–social system to a more advanced computing system cyber–physical–social system (CPSS). Therefore, CPSS leads the data science revolution by promoting tri-space information resource from a single space. The establishment of CPSSs increases the related privacy concerns. To provide privacy on CPSSs data, various privacy-preserving schemes have been introduced in the recent past. However, technological advancement in CPSSs requires the modifications of previous techniques to suit its dynamics. Meanwhile, differential privacy has emerged as an effective method to safeguard CPSSs data privacy. To completely comprehend the state-of-the-art developments and learn the field’s research directions, this article provides a comprehensive review of differentially private data fusion and deep learning in CPSSs. Additionally, we present a novel differentially private data fusion and deep learning Framework for Cyber–Physical–Social Systems , and various future research directions for CPSSs.     

Non-destructive testing and pre-warning analysis on the quality of bolt support in deep roadways of mining districts

The bolt support quality of coal roadways is one of the important factors for the efficiency and security of coal production. By means of a self-developed technique and equipment of random non-destructive testing, non-destructive detection and pre-warning analysis on the quality of bolt support in deep roadways of mining districts were performed in a number of mining areas. The measured data were obtained in the detection instances of abnormal in-situ stress and support invalidation etc. The corresponding relation between axial bolt load variation and roadway surrounding rock deformation and stability was summarized in different mining service stages. Pre-warning technology of roadway surrounding rock stability is proposed based on the detection of axial bolt load. Meanwhile, pre-warning indicators of axial bolt load in different mining service stages are offered and some successful pre-warning cases are also illustrated.The research results show that the change rules of axial bolt load in different mining service stages are quite similar in different mining areas. The change of axial bolt load is in accord with the adjustment of surrounding rock stress, which can consequently reflect the deformation and stability state of roadway surrounding rock. Through the detection of axial bolt load in different sections of roadways, the status of real-time bolt support quality can be reflected; meanwhile, the rationality of bolt support design can be evaluated which provides reference for bolting parameters optimization.     

Deep mine cooling,a case for Northern Ontario:Part Ⅰ

Cooling energy needs,for mines in Northern Ontario,are mainly driven by the mining depth and its operation.Part I of this research focusses on the thermal energy loads in deep mines as a result of the virgin rock temperature,mining operations and climatic conditions.A breakdown of the various heat sources is outlined,for an underground mine producing 3500 tonnes per day of broken rock,taking into consideration the latent and sensible portions of that heat to properly assess the wet bulb global temperature.The resulting thermal loads indicate that cooling efforts would be needed both at surface and underground to maintain the temperature underground within the legal threshold.In winter the air might also have to be heated at surface and cooled underground,to ensure that icing does not occur in the inlet ventilation shaft-the main reason why cooling cannot be focussed solely at surface.     

合肥地铁深基坑开挖对邻近高速公路沉降的影响

针对合肥地铁深基坑开挖对邻近高速公路沉降的影响,采用MidasGTS有限元数值计算软件进行分析,建立了考虑高速公路及基坑施工过程的三维有限元模型,研究了邻近高速公路在不同开挖阶段的最大沉降变化以及不同开挖阶段高速公路的路面沉降分布规律,研究表明:路面最大沉降位置始终在离基坑最近的一侧路面,最终沉降量为7.77mm,对路面安全没有造成影响;随着开挖的进行,沉降量和沉降影响范围不断增加,但增加速率越来越缓慢;研究所得到结论可用来指导未来的相似工程。     

深空通信系统中高阶LDPC码的软定时同步算法

深空通信具有信号衰减严重,接收机信噪比低的特点,针对深空通信系统中高阶LDPC码定时同步实现困难的问题,提出一种码辅助定时同步算法．首先使用高阶LDPC码的译码软信息构造一种代价函数,进行大定时偏移的粗同步,其次基于最大似然准则,使用EM算法进行定时偏移的细同步．该算法将定时同步器、解调器与译码器联合迭代,利用高阶LDPC译码器输出的软信息辅助定时同步,并通过数字插值方式对过采样信号进行补偿,从而得到接近修正克拉美罗限的定时估计．仿真结果表明,在低信噪比条件下,算法能够在较大定时偏移范围内实现有效的定时同步,并以较低的复杂度获得近似理想的系统误码率性能．     

Experimental investigations of machining characteristics and removal mechanisms of advanced ceramics in high speed deep grinding

Machining characteristics and surface integrity of advanced ceramics, including alumina, alumina–titania, and yttria partially stabilized tetragonal zirconia, were studied under high speed deep grinding conditions. Material removal mechanisms involved in the grinding processes were explored. The material removal in the grinding of alumina and alumina–titania was dominated by grain dislodgement or lateral cracking along grain boundaries. The removal for zirconia was via both local micro fracture and ductile cutting. It was found that under a feed rate of 500 mm/min and for all the wheel speeds used, an increase in the wheel depth of cut (DOC) from 0.1–2 mm slightly improved the ground surface finish, but greatly prolonged the wheel life. This increase did not deepen the subsurface damage layer for the alumina and alumina–titania, but resulted in a slightly deeper damage layer for the zirconia.