Union of Class-Dependent Collaborative Representation Based on Maximum Margin Projection for Hyperspectral Imagery Classification

This article proposed a novel spectral-spatial classification framework for hyperspectral image (HSI) through combining collaborative representation (CR) and maximum margin projection (MMP). First, class-dependent CR classifier (CDCRC) is used on HSI classification to fully make use of self-information contained in each class. Second, the MMP is included into the framework to discover local manifold structure. Combined with CDCRC, it formed the classifier named CDCRC based on MMP (CMCRC), which aims to reduce band redundancy. Finally, a comprehensive spectral-spatial classifier, called union of CMCRC, is proposed to optimize the classification map through integrating cumulative probability of residuals instead of applying strong constraints to maintain the spatial consistency. Experimental results on three real hyperspectral datasets demonstrate the effectiveness and practicality of the proposed methods over other related models for HSI classification tasks.

Automated summary

This article introduces a novel spectral-spatial classification framework for hyperspectral images by combining collaborative representation and maximum margin projection. Experimental results demonstrate the effectiveness and practicality of the proposed methods for HSI classification tasks.