A distributed approach to hyper-spectral image analysis using support vectors

This paper presents a detailed analysis of a new distributed algorithm designed for hyper-spectral image analysis. Hyper-spectral imaging is a valuable technique for detection and classification of materials and objects on the Earth's surface. The conventional approach to hyper-spectral image analysis is based on dimensionality reduction using Principal Component Analysis (PCA). But the results contain more details of the frequently occurred objects compared to the minor objects in the scene. To resolve this, a new algorithm for hyper-spectral image analysis based on Support Vector Clustering (SVC) and Spectral Angle Mapping (SAM) is proposed in this work. The method is found to generate good results, but the calculation of Support Vectors, Spectral Angles and Principal Components are very time-consuming processes and a bulk of data is to be processed to analyse the hyper-spectral images. So the algorithm is designed in a distributed manner and a distributed environment based on Java/RMI is developed to implement it. The algorithm is tested with two Hyper-spectral image datasets of 210 bands each, which are taken with HYper-spectral Digital Imagery Collection Experiment (HYDICE) air-borne sensors. A performance analysis of the distributed environment is also carried out.