Hyper-Spectral (HS) image classification based on hierarchical tree approach with neural networks

Abstract

In this paper, a new algorithm has been designated for classification of satellite remote sensing of hyperspectral image. The classification process is based on the three main categories: (1) image fusion, performed using morphological process of both spatial and spectral information available in the remote sensed images. (2) Clustering, which performed in supervised techniques using thresholding effect of image pixel intensity and (3) segmented and texture based image analysis, in this process to achieve a new textural based image clustering to overcome the problem of multi-label images in satellite remote processing. The method of Neighborhood Cell Pattern (NCP) which provides textural information about a given image has developed. The outcome of NCP, optimization function is implemented to provide best selection and clustering of image intensity. Finally, it gets clustered and result in segmented output. In this proposed research work has to introduce two major new classification algorithms for optimization and image clustering for overall Hyper Spectral Images (HSI). The proposed Path Correlated Ant Colony Optimization (PCACO) algorithms is implemented to provide best selection and the Tree Baggar Algorithm with the Neural Network (HTBANN) image clustering algorithm has given an out labeled image output form classified result. The proposed research contribution is validated by classification experiments using Reflective Optics Spectrographic Imaging System (ROSIS) and Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) image sensors from the results the overall accuracy of single and multi-label of Pavia university dataset is 99.14% and 98.23% .The overall accuracy of single and multi-label of Indian pines of 97.29%, 98.72% respectively. The proposed results indicate that the accuracy of HSI classification is validated effectively by PCACO and HTBANN.