Hyperspectral image classification using MLL and Graph cut methods

Hyperspectral image consists of several spectral bands and all those bands are continuous i.e., each pixel in the image consists of continuous spectrum. But, the number of labeled samples present in these bands that would be used for training is scarce and also the acquisition of the labeled samples is difficult and time consuming. Whereas, the generation of the unlabeled samples is effortless. The Semisupervised Learning (SLL) algorithm helps to obtain the unlabeled samples from a constrained set of labeled samples without much struggle and cost. In proposed method, the semisupervised learning that adapts the Self learning strategies is developed. In the self learning method the machine learning algorithm itself selects the most favorable unlabeled samples for the classification process. Two different approaches are used. First, a segmentation method called the Graph cut technique is used and then the Multilevel Logistic (MLL) classifier is used. The Indian Pines scene captured by the Airborne Visible-Infrared Imaging Spectrometer (AVIRIS) sensor that has sixteen different classes is taken for learning and classification.