节点文献
顾及梯度的高斯混合模型在三维属性场空间聚类中的应用
Application of the gradient-based Gaussian mixture model to spatial clustering of three-dimensional attribute field
【摘要】 针对高斯混合模型(GMM)在空间聚类中由于忽视目标对象之间的空间关联性而导致的高误判率等问题,本文提出了一种顾及梯度的高斯混合模型:GMM-G,并将其应用在三维属性场的空间聚类中。GMM-G用反映标量场最大属性变化方向的梯度因子来定义邻域规则,设定梯度正交平面所通过的邻域体元更倾向于与中心体元归属于相同或相近的类别;并据此设计了符合归一性和空间连续性的空间邻域信息函数,来定义中心体元属于各类别的具有空间领域规则约束的后验概率。通过对由蒙特卡洛随机抽样构建的实验场的空间聚类结果进行对比表明,相对GMM方法,GMM-G具有更优的聚类精度及效率。最后,把GMM-G方法用于红透山铜矿区可控源音频大地电磁法(CSAMT)三维视电阻率场的空间聚类,得到了与已知岩性划分具有较高匹配度的分类结果,该方法可为物性属性场的岩性划分及地质推断提供相关的依据和参考。
【Abstract】 To solve the problem of high rate of misjudgment due to ignoring the target objects’ spatial correlation in the Gaussian mixture model(GMM)spatial clustering this paper proposes a new spatial clustering method,i.e.the gradient-based Gaussian mixture model(GMM-G)and it is applied to spatial clustering of three-dimensional attribute field.In this method the gradient factor,which reflects the biggest attribute changing direction of scalar field is selected to define the neighborhood rules and is assumed that the neighboring voxels,which is intersected by the gradient orthogonal plane,are more inclined to belonged to the same or similar class as the central voxel.Accordingly,the spatial neighborhood information function which make need of unitary and spatial continuity is designed to define the posteriori probability of central voxel belonging to various class with rules of the spatial constraints.The experimental results show that the GMM-G method has better clustering accuracy and efficiency than the GMM method in the spatial clustering of experimental filed which is created by the Monte Carlo random sampling technique.In the end,the GMM-G method is applied in spatial clustering of the controlled source audio-frequency magnetotelluric method(CSAMT)three-dimensional resistivity field in Hongtoushan copper mining area,NE China.The clustering result has the high matching ratio with the known lithology classification,therefore,this method can provide the lithology classification and geological inferences of geophysical attribute field with the basis and references.
【Key words】 Gaussian mixture model; spatial clustering; gradient; spatial neighboring information function; attribute field;
- 【文献出处】 地质找矿论丛 ,Contributions to Geology and Mineral Resources Research , 编辑部邮箱 ,2019年03期
- 【分类号】TP311.13;P631.325
- 【被引频次】2
- 【下载频次】108