基于子空间字典偶学习的高光谱图像分类

发布时间：2019-02-22 08:35

【摘要】：针对高光谱高分辨率带来巨大数据量和空间分辨率引起混合像元的问题,提出了基于子空间(subspace)的字典偶学习(DPL)算法,简称DPLsub算法。DPL算法是对字典学习的改进,它通过学习得到综合字典和分析字典,在模式识别中体现了高效性,而子空间投影的方法能更好地表征噪声和高度混合的像元。将光谱和空间特征融合的方法用于分类研究试验。实验数据是两幅高光谱影像,比较了子空间字典偶学习(DPLsub)模型和其他三种分类器即最小二乘支持向量机(LS-SVM)、稀疏多分类回归(SMLR)和字典学习(DL-OMP)的分类结果。实验结果显示,DPLsub算法无论在时间上还是精度上都优于其他算法,证明了这种子空间字典偶学习方法对高光谱图像分类的可行性与高效性。
[Abstract]:In order to solve the problem that hyperspectral and hyperresolution brings huge data and spatial resolution leads to mixed pixel, a dictionary even learning (DPL) algorithm based on subspace (subspace) is proposed, which is referred to as DPLsub algorithm. DPL algorithm is an improvement on dictionary learning. By learning comprehensive dictionaries and analytical dictionaries, it has high efficiency in pattern recognition, and subspace projection can better represent noise-highly mixed pixels. The method of spectral and spatial feature fusion is used in classification research experiment. The experimental data are two hyperspectral images. The subspace dictionary-even-learning (DPLsub) model and the other three classifiers, least squares support vector machine (LS-SVM), are compared. Sparse multiple classification regression (SMLR) and dictionary learning (DL-OMP) classification results. The experimental results show that the DPLsub algorithm is superior to other algorithms in both time and accuracy. It is proved that this subspace dictionary couple learning method is feasible and efficient for hyperspectral image classification.
【作者单位】：燕山大学信息科学与工程学院;
【基金】：国家自然科学基金(61273019,61473339) 河北省自然科学基金(F2013203368) 中国博士后科学基金面上项目(2014M561202) 河北省博士后专项项目(B2014010005) 河北省青年拔尖人才支持计划([2013]17)资助项目
【分类号】：TP751

【相似文献】