基于子空间字典偶学习的高光谱图像分类

Subspace-based dictionary pair learning for hyperspectral image classification

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。

In view of the problem of huge data amount from hyperspectra＇ s high resolution and the mixed pixels problem from the spatial resolution, a subspace-based dictionary pair learning （DPL） algorithm, abbreviated to DPLsub algorithm, was presented. The DPL algorithm is an improvement of the dictionary learning, which reflects the high efficiency in pattern recognition through learning a synthesis dictionary and an analysis dictionary, while the subspace projection method better characterizes noise and highly mixed pixels. The fusion of spectra and spatial char- acteristics was applied to the classification experiment, and two hyperspectral images were used as the experimental data to compare the classification result of the DPLsub model with that of the other three classifiers of least squares support vector machine （ LS-SVM）, sparsemultinomial logistic regression （SMLR） and dictionary learning （ DL- OMP）. The experimental results verifies the feasibility and effectiveness of the proposed DPLsub algorithm in classification of hyperspectral images, and show that it outperforms other current algorithms in time and accuracy.