PROJECT TITLE :

Unsupervised Hyperspectral Band Selection by Dominant Set Extraction

ABSTRACT:

Unsupervised hyperspectral band selection has been an necessary topic in hyperspectral imagery. This technique aims at choosing some critical and decisive spectral bands from an imaginative image for compact illustration while not compromising and distorting the raw information within the relevant spectral bands. Though several efforts have been created to the present topic, the structural data has not nonetheless been well exploited during band choice, and there are still many deficiencies in search strategies, leaving area for more improvement. This paper tackles the unsupervised hyperspectral band selection drawback from a world perspective and proposes a novel technique claiming the subsequent main contributions: 1) structure-aware measures for band informativeness and independence; and a pair of) a graph formulation of band selection permitting for an economical integrated search by suggests that of dominant set extraction. Experiments on 3 real hyperspectral images demonstrate the superiority of the proposed band selector compared with benchmark methods.