Title | ||
---|---|---|
Hyperspectral Endmember Extraction And Unmixing By A Novel Spatial-Spectral Preprocessing Module |
Abstract | ||
---|---|---|
Integration of spatial context and spectral features of neighborhood pixels in preprocessing modules prior endmember (EM) extraction algorithms has been recently studied in hyperspectral images processing as a result of their capability in enhancing EMs signatures recognition and computational performance. In this paper, we propose an autonomous preprocessing module using incorporation of a novel over-segmentation and unsupervised k-means clustering algorithms. This novel scheme can generate spatially small and homogenous regions with high spatial correlation and minimum local spectral variability. Over-segments located at transition areas between two or more cluster regions are identified and eliminated. Average process and spectral purity computations are applied to find suitable candidates for the subsequent EM extraction algorithm. The efficiency of our proposal is validated on a synthetic hyperspectral image in terms of RMSE reconstruction and average minimum SAD while comparing with state-of-the-art spatial-spectral preprocessing modules. |
Year | DOI | Venue |
---|---|---|
2016 | 10.1109/IGARSS.2016.7729874 | 2016 IEEE INTERNATIONAL GEOSCIENCE AND REMOTE SENSING SYMPOSIUM (IGARSS) |
Keywords | Field | DocType |
clustering, endmember, over-segmentation, spatial, spectral | Endmember,Spectral purity,Computer vision,Full spectral imaging,Pattern recognition,Computer science,Segmentation-based object categorization,Image segmentation,Hyperspectral imaging,Artificial intelligence,Pixel,Cluster analysis | Conference |
ISSN | Citations | PageRank |
2153-6996 | 1 | 0.35 |
References | Authors | |
13 | 3 |
Name | Order | Citations | PageRank |
---|---|---|---|
Fatemeh Kowkabi | 1 | 7 | 1.80 |
Hassan Ghassemian | 2 | 396 | 34.04 |
ahmad keshavarz | 3 | 22 | 5.84 |