二进脊波变换的高光谱遥感图像融合分类

Fusion classification of hyperspectral remote sensing images by dyadic ridgelet transform

将二进脊波变换应用到高光谱遥感图像的数据融合中,并针对该算法的特点,提出了将变换数据分成两部分分别进行融合的融合算法,即将经过二进小波行变换的图像数据进行划分,对于包含图像概貌特征的低频数据进行归一化方差加权融合,对于包含图像边缘、直线等细节特征的高频数据选择各波段数据对应像素点小波系数绝对值最大者作为融合后该像素点的像素值.对标准的AVIR IS高光谱遥感图像实现了数据融合,并在此基础上完成了对高光谱遥感图像的分类.实验结果表明,无论是从直观上还是从数值结果上来看,该方法能有效地实现高光谱遥感图像的数据融合与分类.

Dyadic ridgelet transform can improve fusion classification of hyperspectral remote sensing images. A new fusion algorithm was developed to take maximum advantage of the characteristics of this new transform by dividing image data into two parts by applying dyadic wavelet transform to each row of the image. For low spectrum band da- ta, which contain rough and panoramic characteristics of the image, the norm square error of each band of the hypersprctral data was chosen to weight the fusion algorithm. For high spectrum band data, which contain the salient features （ edges, lines, etc. ）, the largest （ absolute value） wavelet coefficients of the entire bands of hyperspectral data at each pixel point were selected as the fused coefficient at the corresponding pixel point. The proposed fusion algorithm was applied to standard AVIRIS hyperspectral remote sensing images, and the images were classified. The results showed, both visually and numerically, that the proposed transform and data fusion algorithm can effectively achieve fusion classification of hyperspectral remote sensing images.