分形维数作为高光谱遥感数据波段选择的一个指标

Fractal Dimensions Being An Index of Bands Selection for Hyper-spectral Remote Sensing Data

在分析最大值、最小值、标准差等传统统计量作为高光谱遥感数据波段选择方法的优缺点后,将分形维数作为波段选择的一个指标,弥补了传统统计量不能获取图像空间结构信息及其变化规律的缺点。在研究中采用分线法和三角棱柱法两种方法计算了研究地区OMIS-I成像光谱仪各波段沙地、植被的分形维数。分析表明,第 、 两个光谱波段区各波段分形维数变化相对平缓,图像质量及空间结构较好,是研究中重点考虑的波段;而 、 、 各区分形维数较高,且波动性大,图像质量和空间结构差。另外,高光谱数据分形维数计算结果表明,分形维数的变化反映了高光谱数据各波段空间结构信息变化,定量地表示了不同波段间的差异,因此,传统统计方法结合分形维数将为高光谱遥感应用研究中选择最佳波段提供新的技术支持。

Fractal is an excellent tool for researching and exploring spatial structure and its complexity. In this paper, the advantage and disadvantage of traditional statistical method such as maximum, minimum and standard deviation etc for bands selecting of hyper-spectral remote sensing data has been analyzed. Traditional method can't acquire the spatial information of images, and fractal dimensions served as an index of bands selecting can remedy for it. Fractal dimensions values for all selected bands of the two OMIS-I scenes sand and vegetation were computed by Matlab program using the line-divider (isarithm) method and triangular prism method. It shows that the difference of the fractal dimensions inⅠand Ⅴ spectral regions are light, on the contrary, Ⅱ, Ⅲ and Ⅳ regions have bigger change and fluctuation. Basically, fractal dimensions values in Ⅱ, Ⅲ and Ⅳ regions are higher than Ⅰand Ⅴ too. So the spatial structural information and the image quality in Ⅰand Ⅴ regions are better thanⅡ, Ⅲ and Ⅳ regions. The traditional method in combination with the fractal method will be a new technique for the bands selection of hyper-spectral remote sensing data.