In this study, an efficient spectral similarity method referred to as Weighted Chebyshev Distance (WCD) is introduced for supervised classification of hyperspectral imagery (HSI) and target detection applications. The WCD is based on a simple spectral similarity based decision rule using limited amount of reference data. The estimation of upper and lower spectral boundaries of spectral signatures for all classes across spectral bands is referred to as a vector tunnel (VT). To obtain the reference information, the training signatures are provided randomly from existing data for a known class. After determination of the parameters of the WCD algorithm with the training set, classification or detection procedures are accomplished at each pixel. The comparative performances of the algorithms are tested under various cases.

The decision criterion for classification of an input vector is based on choosing its class corresponding to the narrowest VT that the input vector fits in to. This is also shown to be approximated by the WCD in which the weights are chosen as an inverse power of the generalized standard deviation per spectral band. In computer experiments, the WCD classifier is compared with the Euclidian Distance (ED) classifier and the Spectral Angle Map (SAM) classifier.

The WCD algorithm is also used for HSI target detection purpose. Target detection problem is considered as a two-class classification problem. The WCD is characterized only by the target class spectral information. Then, this method is compared with ED, SAM, Spectral Matched Filter (SMF), Adaptive Cosine Estimator (ACE) and Support Vector Machine (SVM) algorithms. During these studies, threshold levels are evaluated based on the Receiver Operating Characteristic Curves (ROC).


Hyperspectral Images, Supervised Classification, Target Detection, Weighted Chebyshev Distance Algorithm

Date of this Version