一种改进的高光谱数据自适应波段选择方法

高光谱遥感数据具有的波段数目多、波段宽度窄、数据量庞大等特点给图像的进一步解译带来了困难。为了解决这一问题,对自适应波段选择的降维方法进行了改进,不仅考虑了高光谱遥感图像波段的信息量和波段间的相关性,更重要的是考虑了各地物连续光谱间的可分性。光谱间的可分性距离越大,表明类间的可分性越大,地物越清晰。首先选出了能有效区分图像上任意两类别的理想波段子集,再根据波段子集中任意3波段的相关系数之和最小和它们的均方差最小两个指标,选出任意两类对间那些包含信息量大、相关性又小、谱间差异又大的3波段组合(且不唯一),最后对整幅影像选出的最佳3波段45、75、85合成的假彩色图像用光谱角度制图法(SAM)进行了分类,总体分类精度达到91.7%,Kappa系数达到0.82。

An Improved Method of Hyperspectral Remote Sensing Data Adaptive Band Selection

In comparison with multispectral remote sensing,the characteristics of hyperspectral remote sensing data with more channels, narrower bandwidth and larger amount of data have caused many problems to further management. In order to solve these problems, this paper introduced an improved
adaptive band selection method of dimensional reduction, which can' t only obtain the bands with the high information content and the low correlativity, but also the bands with the longer divisibility distance between the continous spectrals of all ground objects. It shows that the divisibility distance between the spectrals of the arbitrary two ground objects is more long, these two kinds of ground objects appear more clearly. At first, the ideal band subsets are choiced, from which the optimal three bands combination(not unique) with high information content, smaller correlation coefficient and longer divisibility distance are further choosed based on the lowest correlation coefficient sum and the lowest mean square deviation. In order to testify the effect of the improved adaptive band selection method, the spectral angle mapping classification method is implemented on the false color synthesis image of the channels 45、75 and 85.The results of the classification show that the total classification accuracy percentage of the choiced bands image is 91.7%, Kappa coefficient 0.82.