Post-processing and band selection for hyperspectral image data classification with AdaBoost.MH

This research proposed a method to improve classification performance on hyper spectral image data. The method consists of three phases, i.e., first phase is band selection, second phase is hyperspectral image data classification, and third phase is post-processing. AdaBoost.MH classifier is used for hyperspectral image data classification. The Accuracy of AdaBoost.MH classification without post-processing is 96.3%. After post-processing, accuracy increased by 3.698%. In addition to improving the accuracy of classification, post-processing also can remove the small misclassified pixel (speckle noise). Band selection was used in this research to reduce the number of hyperspectral bands. Band selection method used was contribution ratio value of AdaBoost.MH classifier. The maximum accuracy in this research is 99.9982%, only 1 misclassified pixel. This research improves the classification performance of previous research using Local Binary Pattern (LBp) and Extreme Learning Machine (99.63%).