基于子空间-粗集法的高光谱数据光谱与纹理特征优选

为提高光谱数据光谱信息和纹理信息利用率,提出基于自动子空间划分和粗集理论的光谱与纹理特征优选方法。该方法在传统子空间划分法的基础上,利用粗集约简思想对不同类别地物光谱特征进行约简,得到基于光谱的初选波段,再利用灰度共生矩阵法计算出初选光谱波段的纹理信息,并约简优选,得到基于光谱和纹理信息的终选波段。利用黑河生态水文遥感试验中所获取的机载高光谱数据CASI,开展该方法的实证研究。对原始光谱波段、初选光谱波段和终选波段进行SVM(Support Vector Machine)分类,结果表明:与原始光谱数据相比,经过光谱初选得到的初选波段和增加纹理优选的终选波段,总体分类精度分别提高了0.84%和2.78%,Kappa系数分别提高了0.01和0.035;对地物纹理信息进行深度挖掘可以进一步提高遥感影像分类精度。

Hyperspectral Data Spectrum and Texture Band Selection based on the Subspace-rough Set Method

In order to improve the utilization efficiency of spectral and texture information of hyperspectral data,this paper proposed a method which based on the combination of the automatic subspace division method and rough sets theory for spectrum and texture feature selection.Firstly,we have got the primary selective spectral bands through the traditional subspace division method and rough sets method.Secondly,we calculated the texture characteristics of selective bands as above by the gray level co\|occurrence matrix method.Thirdly,we further completed reduction of selected texture information by rough sets method and then obtained the final selective bands of the spectral and texture.Using the CASI hyperspectral data which was obtained during the eco\|hydrological process experiment in the Heihe River region to validate the method.The original spectral bands,primary selected spectrum bands and final bands were used for classification by the SVM (Support Vector Machine).The results showed that compared with the classification overall accuracy of the original spectral data,that of the primary selected bands and final selected bands increased by 0.84% and 2.78% respectively,and KAPPA coefficient rise 0.01 and 0.035 respectively.The experiment results also indicated that the classification accuracy is able to further improve the extracting of the texture information of the spectral bands.