基于局部空间信息的可变类模糊阈值光学遥感图像分割  

Fuzzy Threshold Optical Remote Sensing Image Segmentation With Variable Class Number Based on Local Spatial Information

作　　者：杨蕴 李玉[1] 赵泉华[1] YANG Yun;LI Yu;ZHAO Quan-Hua(Institute for Remote Sensing Science and Application,School of Geomatics,Liaoning Technical University,Fuxin 123000)

机构地区：[1]辽宁工程技术大学测绘与地理科学学院遥感科学与应用研究所,阜新123000

出　　处：《自动化学报》2022年第2期582-593,共12页Acta Automatica Sinica

基　　金：国家自然科学基金(41301479,41271435)资助。

摘　　要：阈值法分割在光学遥感图像分析中被得到广泛的应用,然而传统阈值法也存在诸多局限性,如对噪声敏感,需人为设定类别数,计算复杂度高等.针对传统阈值法的局限性,提出一种基于局部空间信息的可变类模糊阈值光学遥感图像分割方法.首先,以图像光谱的一阶矩为初始类中心,利用二分法原理和区域间最大相似度准则来快速确定类别数及其中心.然后,通过岭形模糊隶属函数计算各像素点对不同类的隶属程度,同时考虑到像素点的隶属度局部空间信息,在隶属度域中定义一个模糊加权滤波器对各类的隶属度矩阵进行滤波,以滤波后的隶属度集合为依据,按照最大隶属原则确定图像的标号场.最后,对标号场中的局部异常标号进行替换,将修正后的标号场由对应的类中心赋色得到分割图像.视觉和统计分析评价结果表明,与传统阈值法相比,该方法能在减少计算时间的同时获得更好的分割结果,可适用于光学遥感图像的多阈值分割.Threshold segmentation has been widely used in optical remote sensing image analysis.However,traditional threshold methods also have many limitations,such as sensitivity to noise,artificially setting the number of classes,high computational complexity and so on.Aiming at the limitation of traditional threshold methods,a fuzzy threshold optical remote sensing image segmentation method with variable class numbers based on local spatial information is proposed.Firstly,taking the one-order moment of the image spectrum as the initial class center,the dichotomy principle and the maximum similarity criterion between regions are used to quickly determine the number of classes and their centers.Then,through the ridge-shaped fuzzy membership function,the degree of membership of each pixel to different classes is calculated.Meanwhile,considering the local spatial information of the membership of each pixel,a fuzzy weighted filter is defined in the membership domain to filter the membership matrix of each class.Based on the filtered membership set,the label field of the image is determined according to the maximum membership principle.Finally,the local abnormal labels in the label field are replaced,and the corrected label field is colored by the corresponding class center to obtain the segmented image.The results of visual and statistical analysis show that compared with the traditional threshold method,the proposed method can obtain better segmentation results while reducing the computation time.It can be applied to multi-threshold segmentation of optical remote sensing images.

关 键 词：遥感图像分割 局部模糊阈值 可变类 隶属度域 标号场 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]