High Spatial Resolution Digital Imagery
59
5.2
Invasive Species Mapping at Lake Okeechobee, Florida
Two different image processing techniques were used to develop vegetation
class maps for each of the two image sources:
The CAMIS airborne mosaics were processed using a supervised
classification algorithm, which requires field data.
The IKONOS satellite image was processed using an unsupervised
cluster routine that did not require the use of coincident ground truth
data.
5.2.1
CAMIS Imagery
The goal of most applications of digital remotely sensed data is to assign
each multispectral pixel to predefined, discrete class (or group). This foundation
for classification techniques begins with statistically based multivariate dis-
criminant functions that determine the most likely group membership for each
pixel. There are a variety of supervised classification algorithms. The maximum-
likelihood classifier is one of the most frequently employed discrimination tech-
niques. This algorithm uses the class-conditional probability density functions to
calculate the likelihood that a given pixel, possessing its unique spectral vector,
belongs to each of the pre-defined classes. The equation used with the maximum
likelihood/Bayesian classifier is:
D = ln(ac) [0.5 ln( |Covc| ) [0.5 (X Mc)T(Covc 1) (X Mc)]
Where
D = calculated weighted distance (i.e., the likelihood estimate)
c = unique class
X = measurement vector of the candidate pixel (i.e., the unique spectral
vector)
Mc = mean vector of class c (calculated from the training statistics for
that class)
(commonly the analyst accepts the default value of 1.0)
Covc = covariance matrix of the pixels in class c (calculated from the
training statistics for that class)
|Covc| = determinant of Covc matrix
Previous Page
