一种新型基于插值的高光谱图像亚像元定位算法

A novel interpolation based hyperspectral image subpixel mapping algorithm

在高光谱图像亚像元定位领域中,基于插值的亚像元定位算法具有无参数、无迭代、快速运算等优点,得到了广泛的应用。传统的基于插值的亚像元定位算法由于原始高光谱图像分辨率低和现有的光谱解混技术的局限性,使得丰度图像不能充分携带原始高光谱图像的先验信息。针对这一问题,提出了一种新型基于插值的高光谱图像亚像元定位算法。将插值算法直接应用于原始低分辨率高光谱图像,得到高分辨率图像。将高分辨率图像通过光谱解混得到具有丰富原始图像先验信息的丰度图像,利用丰度图像提供的信息将类别标签分配到每个亚像元中,获得亚像元定位结果。仿真实验证明,该方法可以充分地携带原始图像的先验信息,获得比传统的基于插值的亚像元定位算法精度更高的结果。

In the hyperspectral image subpixel mapping field,the subpixel mapping algorithm based on interpolation is nonparametric,noniterative and fast,which can obtain good mapping results. The low resolution of tradional hyperspectral image and the limitation of the spectral unmixing techniques in traditional subpixed mapping do not make the fraction images fully carry the original hyperspectral image prior information. To solve this problem,a novel subpixel mapping based on interpolation is proposed. An interpolation algorithm is applied directly in the original low resolution hyperspectral image to derive a high-resolution image. The fraction images with full original image prior information will be derived from the high-resolution image by spectral unmixing. The fraction images are used to allocate hard class labels to subpixels to achieve subpixel mapping result. Experiment results show that the proposed method can utilize more prior information of the original image and produce higher accuracy result than the existing subpixel mapping based on interpolation.