Hyperspectral Image Classification using Spatial Spectral Features and Machine Learning Approach

Hyperspectral imaging is a technique which gathers large number of images at variouswavelength for the same area of Earth. Once Hyperspectral Image (HSI) has been acquired, a meaningful information can be obtained by further processing it. Processing of HSI must aim at achieving of given goals: detect and classify the elementary materials for each pixel. Hyperspectral image classification is an active area to allocate distinct label to each pixel vector so that it is well defined by a given class. For classification of HSI, Support Vector Machine (SVM) is extensively used. To improve the classification accuracy, spectral-spatial preprocessing technique called Multi hypothesis (MH) prediction has been used prior to SVM classifier. This processed HSI will results in less intraclass inconsistency compared to original image and it gives robust classification in the existence of noise. Major contribution of this work is an improvement of classification accuracy and to achieve it, a post processing step after classification has been applied. Spatial domain filter called, Median filter has been used to smooth out wrong classified samples and hence further improvement in classification accuracy has been achieved.