加权融合核稀疏和协同表示的高光谱影像分类

Hyperspectral image classification by weighted-fusing kernel sparse and collaborative representations

为进一步提高表示分类器中基原子对测试样本的表达能力,提出一种加权融合核稀疏和协同表示的高光谱影像分类算法(WKSCRC)。充分利用核函数处理非线性数据的优势,将高光谱影像数据映射到高维核特征空间;对核稀疏表示系数和核协同表示系数进行加权融合,在核融合表示系数下重构分类测试样本。在ROSIS和AVIRIS两个数据集上的仿真结果表明,该算法在精度与稳定性上优于其它传统分类算法。

To improve the representation ability of the atoms in representation-based classifiers,a hyperspectral image classification algorithm,which named weighted-fusing kernel sparse and collaborative representation was presented.The hyperspectral image data were projected to high dimensional kernel feature space using kernel function which took the advantages of dealing with nonlinear data.The weighted-fusing representation coefficient obtained by fusing two kernel individual representation methods,i.e.kernel SR(KSRC)and kernel CR(KCRC),was used to reconstruction and classification for the test sample.Experiments on two data sets of ROSIS and AVIRIS indicate that the proposed algorithm is better than several traditional representation classification algorithms in precision and stability.