Spectro-Spatial Co-training based Transductive Classifier

S2CoTraC Learner





 
A short description 
The distribution package 
Related publications 
Authors & Aknowledgement 
Contact 


A short description

Remotely sensed hyperspectral image classification is a very challenging task. This is due to two different aspects: the spatial correlation of the spectral signature and the high cost of true sample labeling. In this situation, the collective inference paradigm allows us to manage the spatial correlation between spectral responses of neighboring pixels, as interacting pixels are labeled simultaneously. On the other hand, the transductive inference paradigm allows us to reduce the inference error for the given set of unlabeled data, as sparsely labeled pixels are learned by accounting for both labeled and unlabeled information. Both these paradigms contribute to the definition of a spectro-spatial classification methodology for imagery data.
S2CoTraC (Spectro-Spatial Co-Training based Transductive Classifier) is a novel hyperspectral image classification algorithm to assign a class to each pixel of a sparsely labeled hyperspectral image. It integrates the spectral information and the spatial correlation through a co-training system. Spatial information is iteratively extracted at the object (set of pixels) level rather than at the conventional pixel level. Spatial neighborhoods are defined and used to construct relational features. Classification is performed with iterative co-training strategy using the available spectral information and the extracted spatial information. The more reliable labels predicted by the co-training are fed back to the labeled part of the image.

The distribution package

The S2CoTraC algorithm is implemented in a Java system. It integrates the inductive Support Vector Machine (SVM) as a base classifier of the transductive ensemble system. It uses the Java implementation of SVM included in the WEKA toolkit.

jar Description
S2CoTraC This rar bundle contains (1) s2cotrac.jar that allows us to classify sparsely labeled hyperspectral images by integrating spectral information and the spatial correlation through a co-training strategy in the transductive setting; (2) config file and (3) benchmark images (Indian Pines, Pavia University and Salinas Valley)

Warning: The S2COTRAC Learner is free for evaluation, research and teaching purposes, but not for commercial purposes.
Please Acknowledge
 

Top of this page

Related publications

 
A. Appice, P. Guccione, D. Malerba A Novel Spectral-Spatial Co-Training Algorithm for the Transductive Classification of Hyperspectral Imagery Datał Pattern Recognition, Elsevier, accepted for publication, DOI information: 10.1016/j.patcog.2016.10.010

Top of this page

Project team

  • Dr. Annalisa APPICE
  • Dr. Ing. Pietro GUCCIONE
  • Prof. Donato MALERBA


    Contact

    Name Email address Tel. number Fax
    Annalisa Appice annalisa.appice@uniba.it +39 080 5443262 +39 080 5443262
     

    Top of this page