一种鲁棒性的遥感图像地域分类新方法

目前遥感图像分类算法面临的主要问题是分类精度与算法复杂度的矛盾及算法缺乏鲁棒性。为此,提出了一种基于特征空间重采样的非参数化核密度估计聚类与边缘检测相融合的多模型鲁棒性遥感图像分类方法。首先对遥感图像进行边缘检测以获取图像中每个像素的边缘梯度和方向信息;然后利用重采样策略,在联合域中对新的样本集合进行加权均值平移滤波,找到图像各区域的核密度函数局部最大值,通过迭代移动附近的数据点到此局部最大值;最后对各个分割区域进行合并,得到最终的分类图。实验结果表明,算法可获得高精度的遥感图像分类结果,且具有很强的鲁棒性。

A New Robust Approach for Remote Sensing Image Regional Classification

The main problem of remote sensing image classification is the contradiction of classification precision and algorithm complexity, and algorithm lacking of robust. Therefore, a multi - model robust approach of remote sensing image classification based on non- parameter kernel density estimation of resampling strategy in feature space and edge detection is proposed in this paper. The edge gradient and direction information are obtained by edge detection of remote sensing. Then the new samples sets are weighted mean shift filtering to find kernd density function local maximum of image each region using resampling strategy in the joint spatial- range domain and data points are shifted the local maximum by iterative shifting. Last, the classification image is obtained by combining each region. Experimental results illustrate that it is able to classify remote sensing image effectively and robustly.