A novel multilevel adaptive pixel classification and detection (AMLCD)
method for detecting pixel and subpixel-size targets for multispectra
l images is presented. The AMLCD method takes into account both spectr
al and spatial characteristics of the data. In the first level of proc
essing, the principal background end members are obtained using the K-
means clustering method, Each pixel is examined next for classificatio
n using a minimum-distance classifier with the principal end members o
btained in the previous level. In the second level, the neighborhood O
f each unclassified pixel is analyzed for inclusion of candidate end m
embers in an unmixing procedure, If the list of candidate background c
lasses is empty, the conditions for their inclusion are relaxed, The f
ractions of neighborhood and target signatures for the unclassified pi
xels are determined by means of a linear least-squares method in the t
hird level. If the results of unmixing are not satisfactory, the list
of candidate clusters is renewed. Target detection within each pixel i
s performed next, The last processing level determines the size and lo
cation of detected targets with a clustering analysis methodology. Tar
get size and location are estimated on the basis of the sum and weight
ed vector mean, respectively, of the mixing fractions of the neighbori
ng pixels, The AMLCD method was successfully applied to both synthetic
and Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) hyperspec
tral imagery datasets. (C) 1996 Society of Photo-Optical Instrumentati
on Engineers.