Hyperspectral Image Classification using Ensemble Average Method

A Hyperspectral image, has hundreds of spectral bands as opposed to a classical digital image representation of three spectral bands and therefore provides a more complete part of the light spectrum to display and analyze. Hyperspectral imaging provides the possibility of collecting more precise and comprehensive information than other remotely sensed data. Owing to the immense amount of knowledge available in many spectral bands, classification of hyperspectral remote sensing data is a more difficult task. To overcome this problem, the dimensionality reduction-based classification is performed. Ensemble learning is one significant classification method among many classification methods that works by dividing the majority classes into various small-sized groups. In this work, Ensemble Support Vector Machine (SVM) is used to improve the classification results. The experiment utilizes industry-standard hyperspectral data sets, including those from AVIRIS Indian Pines, AVIRIS Salinas and ROSIS Pavia University and inferred that the proposed method shows markable improvement over existing methods. The improvement for Indian Pines image is 15%, for Pavia University 18% and for the Salinas 13%. To ensure accuracy, results are cross validated against ground truth.