基于低秩结构提取的高光谱图像压缩表示

Low-rank Structure Based Hyperspectral Compression Representation

为实现高效、精准的高光谱图像分类,该文利用低秩矩阵恢复从原始数据中提取低维特征,实现高光谱图像的压缩表示。针对高光谱应用的特殊性,该文算法基于结构相似性度量(Structural Similarity Index Measurement,SSIM)对矩阵恢复过程提出了信噪分离约束,有助于选择更优的模型参数,增强表示的准确性。实验证明,相比现有相关方法,该文算法能够有效去除高光谱图像中的噪声,表示结果更为鲁棒;在仅使用低维特征时,仍能达到较高的分类精度。

A method which makes use of structure information abstracted from hyperspectral data via low-rank matrix recovery for hyperspectral image classification is proposed in this paper. The principle of maximizing structure information based on Structural Similarity Index Measurement（SSIM） is proposed to restrain the process of matrix recovery as well, which facilitates the separation of the signal and the noise. The experiments show that the proposed algorithm can effectively eliminate the non-linear noise in hyperspectral image and abstract the low-rank characteristics of hyperspectral image, which achieves better performance in classification.