NEW HYBRID BAYESIAN CLASSIFIER (HBC)

HBC uses the abundance of each class obtained from LS-HSR data by linear unmixing as prior probability while classifying the HS-LSR data using Bayesian classifier of the same geographical area and time frame. That is, given the observation vector  for a pixel, it is classified to fall in class l if , where  is as in (7) calculated using the HS-LSR data and  calculated using the LS-HSR data. The assumptions in this method are (i) if there are r HS-LSR pixels contained in one LS-HSR pixel, i.e. resolution ratio is (r:1), the prior probabilities for all the r HS-LSR pixels are equal corresponding to the same LS-HSR pixel, and (ii) the two data types have a common origin or upper left corner, i.e. the edges of the r x r HS-LSR pixels overlaps exactly with the corresponding LS-HSR pixel. The limitations are (i) (K-1) should be ≥ M in LS-HSR data and (ii) M in HS-LSR should be ≤ M in LS-HSR data.