Superresolution for classification of hyperspectral datasets

cr

We investigate the use of machine learning techniques such as Bayesian inferencing, Deep learning and Support vector machines for improving the interpretability of hyperspectral datasets. In recent works, we proposed 1) use of coarse image features to improve the optimization formulation of conventional rank based sub-pixel classification approaches 2) an enhanced unsupervised variogram based sub-pixel mapping approach 3) use of inter class compatibility information from coarse images to refine the predicted target distribution 4) inclusion of abundance estimation uncertainty in the unmixing process. Further, we attempted to tailor the prominent classifiers such as SVM, ELM and STM towards supervised spectral unmixing. Effectiveness of deep and shallow learning networks in super-resolving coarse images has been also investigated.




Publications

  • P.V. Arun , B.K. Mohan, and A. Porwal. Integration of Contextual Knowledge in Unsupervised Sub-Pixel Classification, accepted in IEEE WHISPERS 2016, Los Angeles, USA.
  • P.V. Arun , and B.K. Mohan. A Deep Learning Based Spatial Dependency Modelling Approach towards Super-Resolution, accepted in IEEE IGARSS 2016, Beijing, China.
  • P.V. Arun and B.K. Mohan. Classification and Clustering Perspective towards Spectral Unmixing, accepted in IEEE IGARSS 2016, Beijing, China.