基于自适应波段聚类PCA的高光谱图像压缩

Hyperspectral Image Compression Based on Adaptive Band Clustering PCA

对高光谱图像进行有效压缩已经成为高光谱遥感领域的研究热点。针对现有高光谱图像压缩算法谱间特性利用不够充分的问题,提出了一种自适应波段聚类PCA(principal component analysis)与JPEG2000相结合的高光谱图像压缩算法。算法采用基于吸引力传播聚类的方法进行自适应波段聚类,对聚类后的各个波段组分别进行PCA运算,最后利用JPEG2000标准对所有主成分进行编码压缩。对高光谱图像进行波段聚类,不仅能更有效地利用谱间相关性,提高压缩性能;还可以降低PCA的运算量。实验结果表明,该算法在相同压缩比下,其信噪比、异常检测、光谱角性能相比对比算法均有所改善。

Efficient compression of hyperspectral images has been the focus in the field of hyperspectral remote sensing. Aiming at the problem of inadequate usage of spectral characteristics of existing hyperspectral image compression algorithms. A new compression algorithm of hyperspectral images using adaptive band clustering principal component analysis （PCA） in conjunction with JPEG2000. The proposed algorithm using affinity propagation for adaptive band clustering, PCA is performed on the each band group respectively after clustering. At last JPEG2000 algorithm is employed in coding the principle components. In the proposed scheme, band clustering not only can further utilize spectral correlation of hyperspectral images and improve the compression performance but also reduce the calculation of PCA. Experimental results investigate that the proposed algorithm can achieve a better performance on signal-to-noise ratio （SNR）, anomaly detection, and spectral angle compared with the state of the art algorithms in the same compression ratio.