Multi-Level Manifolds for Remote Sensing Imagery Classification

A systematic approach for supervised classification of remote sensing images is introduced in this letter. The proposed method deals with the Multi-Level Manifolds, which primarily deals by preserving the local information inside a class along with the class label information. The sharing features are also considered while training the data to represent the parent manifold. The out of sample problem is solved by using Pulse Coded Neural Network which potentially reduces the computational cost. The proposed method solves the major problems of supervised learning systems such as out of sample and preserving local structure. The proposed system is tested in the standard data sets and the results are appreciable.