Urban Objects Classification from HSR-HTIR Data Using Gaussian and Mahalanobis Distance Classifiers

Detection and classification of objects have been extremely difficult without manual help which was tedious and time consuming. In recent years, High spatial resolution hyper spectral thermal infrared (HSR-HTIR) turned into a novel wellspring of information that ended up accessible for urban object detection. The first part of the process is to perform principle component analysis (PCA) and Fischer Linear Discriminant Analysis (FLDA) for band reduction. The set of features derived from HSR-HTIR images are fed into various classifier algorithms for detecting the urban classes. The image classification is done using two algorithms namely Gaussian and Mahalanobis distance classifier for both raw and dimension reduced datasets. The results obtained by using the above classifiers are compared with each other and a performance evaluation is made based on the level of accuracy.