一种快速非参数核密度模型及其应用  

New multivariable nonparametric kernel density estimation model and its application

作　　者：员永生[1] 常庆瑞[1] 杨为民[2] 

机构地区：[1]西北农林科技大学资源环境学院,陕西杨凌712100 [2]西南林学院计算机科学系,昆明650224

出　　处：《兰州大学学报（自然科学版）》2009年第2期132-137,共6页Journal of Lanzhou University(Natural Sciences)

基　　金：国家自然科学基金项目(30872073,30571527);西北农林科技大学创新团队支持计划项目(2006)

摘　　要：基于快速高斯变换,提出并建立了一个多维非参数核密度估计监督学习模型并且评估它的性能,为未知分布试验数据的预测和拟合以及土地利用/覆盖变化监测中多维影像数据的预测和识别提供科学依据.给定M个输入特征向量,对应第j类的容量为N_j的训练样本,直接计算的贝叶斯核密度模型复杂度为O(M×N),而本文提出的核密度模型复杂度接近O(M+N),实现了一个高效的基于快速高斯变换的多维非参数监督学习模型和计算信息处理系统.结果表明:提出的模型预测识别精度与支持向量机分类基本一致,速度更快适合大规模的数据处理,有效降低了维数和计算复杂度诅咒.An efficient multi-variable kernel density estimation model was constructed for supervised classification based on improved fast Gaussian transformation. The proposed method can be used for multidimensional image data for the monitoring of land use/cover or for predicting unknown density distribution data sets. Given the summation of a mixture of M Gaussian at Nj evaluation points in j th category, at first, the original kernel density model was designed and the computational complexity was as high as O（M × N） under direct computation. Then, in contrast with the former model, the proposed model achieved a complexity close to O（M ＋ N） using the improved fast Gaussian transformation. Finally, the efficient model and computation system were established based on a multi-variable kernel density estimation. The results show that the proposed method works as well as support vector machines in terms of precision and has lower time consumption. It is suitable for computing large scale and high dimensional data sets. It also effectively decreases the dimensional curse in computation information.

关 键 词：光滑核函数 图像识别 快速高斯变换 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] TP751.1[自动化与计算机技术—计算机科学与技术]