【摘要】 针对单一使用联合稀疏表示分类(JSRC)或支持向量机(SVM)对高光谱遥感图像进行分类时,单级分类器不能很好适应高光谱遥感图像所具有的维度高、像元信号相似度高和线性混杂的特点导致分类精度差的问题,提出一种两级分类器方法对高光谱图像进行分类。首先将JSRC作为前级分类器进行分类,然后选出重构残差最小的两个类计算辨识系数,当系数大于预设阈值时,直接采纳JSRC分类结果,否则用这两个基原子对应类的训练样本去训练后级SVM分类器,再输出SVM的分类结果。实验结果表明,在帕维亚大学(University of Pavia)数据集上两级分类器算法的总分类精度与JSRC相比提高了3.26%;与SVM相比总分类精度提高了2.33%。所提算法克服了JSRC和SVM对高光谱图像信号适应性不稳定的缺点,在高光谱遥感图像的分类精度上有较大的优势。更多还原

【Abstract】 When applied to classification of Hyperspectral remote Sensing Image( HSI) data separately,common methods such as spatial Joint Sparse Representation Classification( JSRC) and Support Vector Machine( SVM) do not perform well with the HSI signals with high dimension,high similarity and linear hybridization,leading to low classification accuracy in HSI classification.A novel hybrid approach combining JSRC as the first stage and SVM as the second stage for HSI classification was proposed.First,JSRC was used to select two main bases.Then a threshold was set,only when the condition was fulfilled,JSRC could be used to determine the classification.Otherwise signals trained by the two selected bases were fed to SVM and the output of SVM would be the final result.The experimental results on dataset of University of Pavia show that the overall accuracy of the proposed method is improved by 3.26% compared with JSRC and 2.33% compared with SVM.The proposed algorithm can overcome JSRC and SVM’ s drawback to adapt to complex signal of HSI,and it has better classification performance in accuracy.更多还原