基于分层学习的三维模型兴趣点提取算法

针对基于学习的三维模型兴趣点提取问题,提出一种兴趣点分层学习的全监督算法.提取三维模型表面所有顶点的特征向量后,将人工标注的兴趣点分为稀疏点和密集点,对于稀疏点使用整个三维模型进行神经网络训练,对于密集点则找出兴趣点分布密集的区域进行单独的神经网络训练;然后对2个神经网络进行特征匹配,得到一个用于三维模型兴趣点提取预测的分类器.测试时,提取新输入的三维模型上所有顶点的特征向量,将其输入到训练好的分类器中进行预测,应用改进的密度峰值聚类算法提取兴趣点.算法采用分层学习的策略,解决了传统算法在模型细节处难以准确提取密集兴趣点的问题.在SHREC’11数据集上的实验结果表明,与传统算法相比,该算法提取兴趣点的准确率更高,出现的遗漏点和错误点更少,对解决越来越精细的三维模型的兴趣点提取问题有较大帮助.     

Vertical-dual-source tunnel FETs with steeper subthreshold swing

In order to improve the drive current and subthreshold swing (SS), a novel vertical-dual-source tunneling field-effect transistor (VDSTFET) device is proposed in this paper. The influence of source height, channel length and channel thickness on the device are investigated through two-dimensional numerical simulations. Si-VDSTFET have greater tunneling area and thinner channel, showing an on-current as high as 1.24 μA at gate voltage of 0.8 V and drain voltage of 0.5 V, off-current of less than 0.1 fA, an improved average subthreshold swing of 14 mV/dec, and a minimum point slope of 4 mV/dec.