基于核空间的多光谱遥感图像分类方法

CLASSIFICATION OF KERNEL BASED ON MULTI-BAND REMOTE SENSING IMAGES

提出了多光谱遥感图像分类方法中解决非线性问题的一种思路。通过引入核空间理论 ,将在输入空间中不能线性分类的问题映射到一个可以进行线性分类的高维空间 ,并利用核函数避免了在高维空间中运算的复杂度 ,较好地解决了非线性分类问题。利用这种思路 ,本文对一种比较简单的分类算法———自适应最小距离分类方法加以改进 ,并将其应用于多光谱遥感图像的分类中 ,提出了一种核函数的选择策略。实验表明 ,这种策略更有利于多光谱遥感图像的分类 ,在训练速度降低较少的情况下 。

In this paper, a method for solving the nonlinear problem of classifying Multi-band Remote Sensing Images is proposed. By introducing the concept of Kernel space, the classification, which cannot be performed linearly in the input space, can be mapped into a high-dimension space, in which the problem can be solved linearly. Moreover, by using Kernel Function, the complex computation in the high-dimension space can be avoided. Based on this method, this paper improved a simple classification method called adaptive min-distance algorithm, and applied it to the classification of multi-band remote sensing images. A choice heuristic is also presented to select an appropriate Kernel Function. Experiments show that, with higher accuracy achieved, the improvements prove to be useful in the classification.