针对传统电磁频谱地图构建方法感知节点分布不均匀、监测数据存在异常值等问题,将基于密度的噪声应用空间聚类(Density-based Spatial Clustering of Applications with Noise,DBSCAN)算法与反距离加权(Inverse Distance Weighting,IDW...针对传统电磁频谱地图构建方法感知节点分布不均匀、监测数据存在异常值等问题,将基于密度的噪声应用空间聚类(Density-based Spatial Clustering of Applications with Noise,DBSCAN)算法与反距离加权(Inverse Distance Weighting,IDW)算法相结合,提出一种城市环境电磁频谱地图构建方法。该算法首先通过DBSCAN减轻极端值的影响,并分离核心点、边界点和噪声点,将每个核心点的局部密度作为权重,计算簇的加权中心点。其次,运用IDW对聚类簇的加权中心点进行插值估计,以显著减少需要进行插值的数据点数量,从而构建精度更高的电磁频谱地图。仿真结果表明:与IDW算法和反障碍距离加权算法相比,所提算法重构得到的电磁频谱地图的平均绝对误差和归一化均方误差分别降低了10%和23%以上。展开更多
子空间聚类在最近几年受到了广泛的关注,新近提出的自适应图卷积子空间聚类方法取得了很好的效果。但是该方法仅适用于单一视图的子空间聚类问题。本文将该方法拓展到多视图上,提出了多视图图卷积子空间聚类。该方法构建了F范数正则项...子空间聚类在最近几年受到了广泛的关注,新近提出的自适应图卷积子空间聚类方法取得了很好的效果。但是该方法仅适用于单一视图的子空间聚类问题。本文将该方法拓展到多视图上,提出了多视图图卷积子空间聚类。该方法构建了F范数正则项以便更有效地挖掘每个视图中数据之间的关系,还构建了不同视图之间的加权机制来融合不同视图之间的信息。大量的实验证明,我们的方法是有效的。Subspace clustering has received extensive attention in recent years. Although the recently proposed adaptive graph convolutional subspace clustering performs well, it can only be applied to subspace clustering problem with a single view. This paper proposes multi-view graph convolutional sub-space clustering to extend this method to the multi-view situation. This method not only constructs F-norm regularization to more effectively mine the relationships between data in each view, but also builds a weighting strategy between different views to integrate their information. A large number of experiments have proved that our method is effective.展开更多
文摘针对传统电磁频谱地图构建方法感知节点分布不均匀、监测数据存在异常值等问题,将基于密度的噪声应用空间聚类(Density-based Spatial Clustering of Applications with Noise,DBSCAN)算法与反距离加权(Inverse Distance Weighting,IDW)算法相结合,提出一种城市环境电磁频谱地图构建方法。该算法首先通过DBSCAN减轻极端值的影响,并分离核心点、边界点和噪声点,将每个核心点的局部密度作为权重,计算簇的加权中心点。其次,运用IDW对聚类簇的加权中心点进行插值估计,以显著减少需要进行插值的数据点数量,从而构建精度更高的电磁频谱地图。仿真结果表明:与IDW算法和反障碍距离加权算法相比,所提算法重构得到的电磁频谱地图的平均绝对误差和归一化均方误差分别降低了10%和23%以上。
文摘子空间聚类在最近几年受到了广泛的关注,新近提出的自适应图卷积子空间聚类方法取得了很好的效果。但是该方法仅适用于单一视图的子空间聚类问题。本文将该方法拓展到多视图上,提出了多视图图卷积子空间聚类。该方法构建了F范数正则项以便更有效地挖掘每个视图中数据之间的关系,还构建了不同视图之间的加权机制来融合不同视图之间的信息。大量的实验证明,我们的方法是有效的。Subspace clustering has received extensive attention in recent years. Although the recently proposed adaptive graph convolutional subspace clustering performs well, it can only be applied to subspace clustering problem with a single view. This paper proposes multi-view graph convolutional sub-space clustering to extend this method to the multi-view situation. This method not only constructs F-norm regularization to more effectively mine the relationships between data in each view, but also builds a weighting strategy between different views to integrate their information. A large number of experiments have proved that our method is effective.