Development and analysis of novel spectral and spatial classification approach for airborne hyperspectral images

Abstract

Hyperspectral imagery is becoming emerging trend in remote sensing applications The present and future space missions are primarily based on the hyperspectral image sensors Hyperspectral images contain very detailed information of remotely sensed data Each pixel of the hyperspectral image cube is represented with hundreds of different wavelengths and these contiguous images are formed as three dimensional image cube includes two spatial coordinates and one spectral coordinate Using hyperspectral sensors we can identify the objects minerals vegetation based on their unique spectral imprints Hyperspectral sensors have the ability to acquire images in many narrow spectral bands that are found in the electromagnetic spectrum from visible near infrared medium infrared to thermal infrared Hyperspectral sensors capture energy in 200 bands or more which means that they continuously cover the reflecting spectrum for each pixel in the scene Bands characteristic for these types of sensors are continuous and narrow allowing an in depth examination of features and details on the earth surface The primary objective of thesis is the development of novel approaches for an effective classification of hyperspectral images with emphasis on optimized utilization of spectral and spatial information In the first phase of research work a spectral signature based classification approach is proposed in which each pixel is assigned to the exact classes based on the spectral signatures or similar spectral statistical characteristics for the effective classification of hyperspectral images The spectral signature based hyperspectral image classification approach is developed and compared its assessment parameters with the k means and fuzzy c means classifiers newline