高光谱遥感数据光谱特征提取算法与分类研究

Study on feature extraction and experiment of hyperspectral data

针对高光谱数据的特点,探讨了高光谱数据特征提取的若干算法,重点研究了导数光谱和光谱编码技术,并从地物光谱曲线中提取了其光谱吸收特征。对同类曲线特征求交得到识别地物的有效特征;对不同类曲线特征求交得到区分不同类地物的有效特征。最后基于提取的特征建立了地物识别决策树,从而达到快速识别分类地物的目的,能够实现依据地物光谱特征的地物识别与分类。

This paper discussed some feature extraction algorithms for hyperspectral data process, and put forward the quad encoding algorithm as the improved binary encoding algorithm. In particular, extracted spectral absorption parameter including absorption position, absorption depth, absorption width, slope, absorption asymmetry, spectral absorption index（SAI） by the derivative spectrum algorithm and binary/quad encoding spectrum algorithm. Using these feature, got the valid feature to recognize the same objects and to recognize different objects. At last, based on the experiment constructed the decision tree to recognize objects.