基于IFCM聚类与变分推断的遥感影像分类  

Remote sensing image classification based on IFCM clustering and variational inference

作　　者：向泽君 黄磊[1] 楚恒[1,2,3] XIANG Ze-jun;HUANG Lei;CHU Heng(School of Communication and Information Engineering,Chongqing University of Posts and Telecommunications,Chongqing 400065,China;Chongqing Urban Planning Bureau,Chongqing 401121,China;Chongqing Survey Institute,Chongqing 400020,China)

机构地区：[1]重庆邮电大学通信与信息工程学院,重庆400065 [2]重庆市规划局,重庆401121 [3]重庆市勘测院,重庆400020

出　　处：《计算机工程与设计》2019年第7期2059-2063,共5页Computer Engineering and Design

基　　金：重庆市2013西南大学博士后科研基金项目(Rc201336);重庆高校创新团队建设计划基金项目(CXTDX201601020)

摘　　要：针对高分影像地物繁多,特征混杂导致现有模糊C均值算法稳定性差、分类精度低的问题,提出一种IFCM(improved FCM)聚类与变分推断法结合的遥感影像分类算法。在聚类分割目标函数计算阶段,考虑像素区域特征的同时,邻域像元采用吸引力模型进行距离测度;特征提取阶段使用空间像素模板法提取像斑特征点,基于贝叶斯统计中的变分推断法逼近参数后验分布,获取影像分类结果。实验结果表明,所提方法能提高影像分类精度。Due to the high number of high-resolution imagery features and the complexity of the features,the existing fuzzy C-means algorithm has poor stability and low classification accuracy. A remote sensing image classification algorithm based on IFCM (improved FCM) clustering and variational inference method was then proposed. In the calculation stage of target function in clustering segmentation,while taking the characteristics of the pixel region into account,the neighborhood pixel employed an attractive model for distance measurement. In the feature extraction stage,the feature points of the image spot were extracted using the spatial pixel template,and the parameter posterior distribution was approximated based on the variational inference in Bayesian statistics to obtain the image classification results. Experimental results show that the proposed method can improve the accuracy of image classification.

关 键 词：改进模糊C均值 变分推断 吸引力模型 像素模板 影像分类 

分 类 号：TN911.73[电子电信—通信与信息系统]