【摘要】 高光谱遥感由于其精细的光谱分辨率,在定量分析物质成分上独具优势,因此广泛应用于提取蚀变矿物信息。探讨了不同空间分辨率高光谱遥感数据对蚀变矿物信息提取的影响,采用最邻近插值法、双线性插值法和三次卷积插值法3种重采样方法对美国Cuprite矿区空间分辨率为20 m的AVIRIS影像做空间尺度扩展,分别扩展到空间分辨率为25,30,35,40,45,50 m。采用SAM分类方法从不同空间分辨率影像中提取蚀变矿物信息,使用混淆矩阵评价提取结果。一方面比较不同重采样方法对后期蚀变矿物信息提取精度产生的影响;另一方面比较不同空间分辨率对高光谱遥感影像蚀变矿物信息提取精度的影响。结果表明:①采用不同的重采样方法做空间尺度扩展,会影响后期蚀变矿物信息提取的精度,但是数值变化相对较小。相比之下,最邻近插值法重采样下影像蚀变矿物信息提取的精度稍好一些。②在中等空间分辨率(20～50 m)范围内,基于50 m空间分辨率的高光谱影像,蚀变信息提取的总体精度和Kappa系数较20 m的明显下降。其中最邻近插值法重采样下的总体精度和Kappa系数分别下降了7.94%,0.09;双线性插值法重采样下的总体精度和Kappa系数分别下降了6.87%,0.08;三次卷积插值法重采样下的总体精度和Kappa系数分别下降了6.68%,0.08。较高空间分辨率影像的总体精度和Kappa系数整体上均高于较低空间分辨率的情形。更多还原

【Abstract】 Hyperspectral remote sensing has high spectral resolution and has unique advantages in the quantitative analysis of material composition. Therefore, it is widely used to identify alteration mineral information. This paper discusses the effects of different spatial resolution hyperspectral remote sensing data on the extraction of alteration minerals information. The paper applies three resampling methods of Nearest Neighbor interpolation, Bilinear interpolation and Cubic Convolution interpolation to extend the spatial scale of the AVIRIS hyperspectral image with a spatial resolution of 20 m in the Cuprite Mining District in the United States and extend it to spatial resolution image data of 25,30,35,40,45 m and 50 m. On the one hand, the paper compares the effects of different resampling methods for the accuracy of alteration information extraction. On the other hand, the paper employs the SAM(Spectral Angle Mapper) classification method to extract alteration minerals information from the different spatial resolution images and then uses a confusion matrix to verify the extraction results and compares the overall accuracy. The results show that:(1) Different resampling methods used to extent the spatial scale affect the accuracy of alteration information extraction, but the numerical changes are relatively small. Among them, the Nearest Neighbor interpolation method is slightly better.(2) In the medium spatial resolution(20-50 m) range, the Overall Accuracy and Kappa coefficient of the alteration information extraction decreases significantly. With the three resampling methods, the overall accuracy of the alteration information extraction drops by 7.94%, 6.87%, 6.68%, and the Kappa coefficient decreases by 0.09, 0.08,0.08, respectively. The overall accuracy of alteration information extraction from higher spatial resolution image is higher than that of lower spatial resolution. This shows that although the alteration information extraction depends on the fine spectral resolution of the hyperspectral image to quantitatively identify the components of minerals, it is generally affected by the spatial resolution as well.更多还原