分形信号算法的CASI高光谱数据岩性特征提取

高光谱数据具有图谱合一和数据量大的特点，数据降维是主要的研究方向。波段选择和特征提取是目前高光谱降维的主要方法，就高光谱数据图像岩性特征提取的方法进行了试验和探讨。基于高光谱影像的自相似特征, 探索了分形信号算法在CASI高光谱数据岩性特征提取上的应用研究。以CASI高光谱影像数据为研究对象, 将基于地毯的方法进行修正后用于计算高光谱影像中每一像元的分形信号值。试验结果表明, 与其他分类算法相比分形信号算法增强高光谱图像的影像特征从另一个侧面更细致的描述了不同光谱的可区分性。分形信号影像在一定程度上可以更好地突出基岩裸露地区岩性特征, 从而可以实现影像地表岩性特征提取的目的。原始光谱曲线自身形态特征、初始尺度的选择以及迭代步长等对分形信号和分形特征尺度均有影响。目前，光谱曲线的分形信号特征研究还不多，对其物理意义和定量分析尚需要深入研究。

Lithology Feature Extraction of CASI Hyperspectral Data Based on Fractal Signal Algorithm

Hyperspectral data is characterized by combination of image and spectrum and large data volume dimension reduction is the main research direction. Band selection and feature extraction is the primary method used for this objective. In the present article, the authors tested methods applied for the lithology feature extraction from hyperspectral data. Based on the self-similarity of hyperspectral data, the authors explored the application of fractal algorithm to lithology feature extraction from CASI hyperspectral data. The “carpet method” was corrected and then applied to calculate the fractal value of every pixel in the hyperspectral data. The results show that fractal information highlights the exposed bedrock lithology better than the original hyperspectral data. The fractal signal and characterized scale are influenced by the spectral curve shape, the initial scale selection and iteration step. At present, research on the fractal signal of spectral curve is rare, implying the necessity of further quantitative analysis and investigation of its physical implications.