Hyperspectral Image Classification Based on Discriminative Joint Sparse Model and Label Refinement

摘要

In this work, we develop an effective classification framework to classify Hyperspectral Image (HSI), which consists of three basic components: dictionary learning, joint sparse model and label refinement. First, we exploit a class-specific KSVD method to learn a set of sub-dictionaries and concatenate them into an overall dictionary. Next, a joint sparse model is introduced for representing neighboring samples around each pixel simultaneously, which can take the spatial information into account in the feature level. The joint sparse model can be solved effectively by using the Simultaneous Orthogonal Matching Pursuit (SOMP) method and therefore the soft-classification scores can be defined based on the reconstruction errors. After that, we design a label refinement scheme to model the spatial information in the label level. Finally, we evaluate the proposed methods by comparing them with other competing algorithms on the Indian Pines and University of Pavia datasets. Both qualitative and quantitative results demonstrate that the proposed methods perform favorably against other algorithms.