It is too difficult for a compacted specimen of unsaturated clay prepared and used in a laboratory test to have homogeneous water content and porosity. Although there have been many hydraulic and mechanical studies, t...It is too difficult for a compacted specimen of unsaturated clay prepared and used in a laboratory test to have homogeneous water content and porosity. Although there have been many hydraulic and mechanical studies, there are few papers related to homogeneity of specimen. However, it is clear that homogeneity of specimen influences mechanical properties. In this paper, the use of micro-waves is proposed as method for making homogeneous specimens. The study results indicate that specimens made by micro-waves are more homogeneous than compacted specimens.展开更多
针对现有基于毫米波雷达的人体动作识别方法存在精度低、模型复杂度高等问题,文章提出一种基于多阶段特征协同处理的毫米波雷达人体动作识别(Human Action Recognition with Multi-Stage Feature Collaboration from Radar,HAR-MFC)方...针对现有基于毫米波雷达的人体动作识别方法存在精度低、模型复杂度高等问题,文章提出一种基于多阶段特征协同处理的毫米波雷达人体动作识别(Human Action Recognition with Multi-Stage Feature Collaboration from Radar,HAR-MFC)方法。该方法通过对雷达回波数据进行分析处理,提取每种动作的微多普勒图,并将其作为识别模型的分类特征。首先,特征提取模块负责提取微多普勒图中的动作特征并减少冗余计算;接着,特征融合模块实现局部细节特征与全局语义信息的有效关联;最后,特征优化模块加速模型的收敛过程。实验结果表明,该模型在自建数据集上的识别准确率达到97.66%,参数量仅为0.7599 M;在格拉斯哥公开数据集上的准确率为96.30%,这表明该模型具有较强的泛化能力。展开更多
文摘It is too difficult for a compacted specimen of unsaturated clay prepared and used in a laboratory test to have homogeneous water content and porosity. Although there have been many hydraulic and mechanical studies, there are few papers related to homogeneity of specimen. However, it is clear that homogeneity of specimen influences mechanical properties. In this paper, the use of micro-waves is proposed as method for making homogeneous specimens. The study results indicate that specimens made by micro-waves are more homogeneous than compacted specimens.
文摘针对现有基于毫米波雷达的人体动作识别方法存在精度低、模型复杂度高等问题,文章提出一种基于多阶段特征协同处理的毫米波雷达人体动作识别(Human Action Recognition with Multi-Stage Feature Collaboration from Radar,HAR-MFC)方法。该方法通过对雷达回波数据进行分析处理,提取每种动作的微多普勒图,并将其作为识别模型的分类特征。首先,特征提取模块负责提取微多普勒图中的动作特征并减少冗余计算;接着,特征融合模块实现局部细节特征与全局语义信息的有效关联;最后,特征优化模块加速模型的收敛过程。实验结果表明,该模型在自建数据集上的识别准确率达到97.66%,参数量仅为0.7599 M;在格拉斯哥公开数据集上的准确率为96.30%,这表明该模型具有较强的泛化能力。
