Protective effects of baffles with different positions,row spacings,heights on debris flow impact

The baffle effectively slowed down debris flow velocity,reduced its kinetic energy,and significantly shortened the distance of debris flow movement.Consequently,they are widely used for protection against natural hazards such as landslides and mudslides.This study,based on the threedimensional DEM(Discrete Element Method),investigated the impact of different baffle positions on debris flow protection.Debris flow velocity and kinetic energy variations were studied through single-factor experiments.Suitable baffle positions were preliminarily selected by analyzing the influence of the first-row baffle position on the impact force and accumulation mass of debris flow.Subsequently,based on the selected baffle positions and four factors influencing the effectiveness of baffle protection(baffle position(P),baffle height(h),row spacing(S_(r)),and angle of transit area(α)),an orthogonal design was employed to further explore the optimal arrangement of baffles.The research results indicate that the use of a baffle structure could effectively slow down the motion velocity of debris flows and dissipate their energy.When the baffle is placed in the transit area,the impact force on the first-row baffle is greater than that when the baffle is placed in the deposition area.Similarly,when the baffle is placed in the transit area,the obstruction effect on debris flow mass is also greater than that when the baffle is placed in the deposition area.Through orthogonal experimental range analysis,when the impact on the first row of baffles is used as the evaluation criterion,the importance of each influencing factor is ranked asα>P>S_(r)>h.When the mass of debris flow behind the baffle is regarded as the evaluation criterion,the rank is changed to P>α>S_(r)>h.The experimental simulation results show that the optimal baffle arrangement is:P_(5),S_(r)=16,α=35°,h=9.

多时间尺度一致性的弱监督时序动作定位

由于弱监督时序动作定位模型使用视频级的标签作为监督信号,模型在识别出动作实例中最具区分性的视频片段时,也会将和视频级标签有关的背景片段误认为是动作,难以产生完整的动作提议。为了进一步检测动作片段,通过分析动作片段在多时间尺度上标记的一致性,提出了一种多时间尺度一致性的弱监督时序动作定位方法。对输入的视频帧提取RGB和光流的特征,设计一种多时间尺度的模块,使用不同尺寸的卷积核建模视频的时序关系。通过估计多时间尺度特征的时间类激活图,并对多分支的时间类激活图进行融合,获得多时间尺度一致性的动作预测标签。为了进一步优化模型预测的动作标签,采用迭代优化策略,在每次迭代中更新预测标签,并为模型训练提供有效的帧级监督信号。在THUMOS14和ActivityNet1.3数据集上进行实验验证,实验结果表明,方法性能优于现有弱监督时序动作定位方法。

基于聚类关联网络的群组行为识别

目前群组行为识别方法没有充分利用群组关联信息而导致群组识别精度无法有效提升,针对这个问题,提出了基于近邻传播算法(AP)的层次关联模块的深度神经网络模型,命名为聚类关联网络(CRN)。首先,利用卷积神经网络(CNN)提取场景特征,再利用区域特征聚集提取场景中的人物特征。然后,利用AP的层次关联网络模块提取群组关联信息。最后,利用长短期记忆网络(LSTM)融合个体特征序列与群组关联信息,并得到最终的群组识别结果。与多流卷积神经网络(MSCNN)方法相比,CRN方法在Volleyball数据集与Collective Activity数据集上的识别准确率分别提升了5.39与3.33个百分点。与置信度能量循环网络(CERN)方法相比,CRN方法在Volleyball数据集与Collective Activity数据集上的识别准确率分别提升了8.7与3.14个百分点。实验结果表明,CRN方法在群体行为识别任务中拥有更高的识别准确精度。

从城市更新的角度解析斯德哥尔摩文化中心

本文以时间顺序叙述,以20世纪60年代初的城市更新为切入点,分析斯德哥尔摩文化中心的设计,该建筑所承担的政治职能、社会文化职能与公共职能之间的关系和变化,以及在周边环境不断发展的过程中,该建筑如何配合城市发展,在改造的过程中与城市互相适应。文章涉及建筑设计、建筑改造和保护立法等多个层面,由此引发对此类大型公共建筑在设计之后如何管理运营的思考。