运用边缘保持滤波和改进核模糊聚类的遥感影像分类  

Remote Sensing Image Classification Method Based on Edge Preserving Filtering and Improved Kernel Fuzzy Clustering

作　　者：薛琳 XUE Lin(Qinghai Remote Sensing Center for Natural Resources,Xining 810000,China)

机构地区：[1]青海省自然资源遥感中心,青海西宁810000

出　　处：《地理空间信息》2024年第12期42-45,64,共5页Geospatial Information

基　　金：高分专项黄河流域青海段生态保护和高质量发展应用产业化示范基金资助项目(94-Y50G36-9001-22/23)。

摘　　要：针对高分辨率遥感影像存在边界模糊、“同物异谱”的特点,使得核模糊聚类算法难以获取理想的分类结果,提出边缘保持滤波和改进核模糊聚类的遥感影像分类方法。该方法首先采用边缘保持滤波算法对高分辨率遥感影像处理,融合影像光谱与空间信息,突出影像边缘特征;然后将混合高斯核函数引入核模糊C均值聚类方法中,将低维数据映射到高维特征空间上;最后为了增强算法抗噪性,利用马尔科夫先验概率对目标函数进行修正。实验表明:该方法能有效保留影像边缘特征,提高影像分类精度。The characteristics of boundary blur and“same object but different spectrum”in high-resolution remote sensing images make it difficult for kernel fuzzy clustering algorithms to obtain ideal classification results.Aiming at this problem,we proposed a remote sensing image classification method based on edge preserving filtering and improved kernel fuzzy clustering.This method uses edge preserving filtering algorithm to process high-resolution remote sensing images,fuses spectral and spatial information of the images,and highlights image edge features at first.Then,the method introduces a mixed Gaussian kernel function into the kernel fuzzy C-means clustering method to map low dimensional data onto high-dimensional feature spaces.Finally,this method uses Markov prior probabilities to modify the objective function,enhance the noise resistance.Experimental result shows that the method can effectively preserve image edge features and improve image classification accuracy.

关 键 词：联合双边滤波 引导滤波 混合核函数 粒子群算法 马尔科夫随机场 

分 类 号：P237[天文地球—摄影测量与遥感]