final Flashcards
What is an arbitrary or spatial profile?
graph across space of spectral profile (e.g. transect from inland marsh to middle of river)
Spectral profile
looks at individual point, or spectral signature for that pixel
Convolution filtering
window that moves over an image sequence and does a function on the pixels it encounters
NDVI
NIR-red / NIR + red
histogram matching
histogram matching tries match responses of classes
linear contrast enhancement or contrast stretching
expands the original digital values of the remotely sensed data into a new distribution
what would a filter look like that accentuates spacial variability?
a linear edge detector would have some negative and some positive values
principal component analysis
allows you take multiple bands, take the variability and store it in just a few bands. Output number of bands is the same but the crucial information is contained in the first few bands. Thus the benefit is that you can decrease the data dimensionality (number of bands)
data dimensionality
number of bands, size of your data defined by rows and columns.
“features” in image processing (different in GIS)
bands
What is the point of features selection
another way to get rid of data you don’t need. It’s determining which bands are the most important
NDVI stands for
Normalized Difference Vegetation Index
Tassled cap transformation
the first of the new axis becomes brightness (formed from soil line). the axis orthogonal to it is called greenness. Then wetness is sometimes the third axis
supervised vs. unsupervised
unsup clusters elements in the image that look similar. supervised always involves a training data set that you give the algorithm
cluster busing
something that is done in unsupervised classification. we take the classes that didn’t get classified and reclassify them
Why is it a good idea to used predefined classification schemes
a class scheme like USGA or Anderson is useful because it allows us to monitor the change over time if we have a classification scheme that stays consistent. also involves less work.
problems with extending spectral signatures through space and time
you’re telling the algorithm that forests look like this..that may work fine in one single image but the atmosphere information in different. sun was different.
feature space
a graphical representation of the pixels by plotting 2 bands vs. each other (often looks like a tasseled cap)
pros of parallelepiped
simple, makes a few assumptions about the character of the classes
cons of parallelpiped
parallelelpipeds are rectangular, but spectral space is “diagonal” so classes may overlap
minimum distance to mean
finds mean value of pixels of training sets in n-dimensional sapce. All pixels in image classified according to the class mean to which they are closest.
pros of minimum distance
all regions of n-dimensional space are classified. Allows for diagonal boundaries (and hence no overlap of classes)
cons of minimum distance
assumes that spectral variability is same in all directions, which is not the case
maximum likelihood
assume multivariate normal distributions of pixels within classes. calculates the probability that the pixel is a member of that class. takes into account mean and covariance of training set. each pixel is assigned to the class for which it has the highest probability of membership
