Remote Sensing

Front 1

Density Slicing

Back 1

simple threshoold methood, anything below a certain number is one class and anything above a certain number is another class

simple but inaccurate

Front 2

Minimum Distance

Back 2

the mean of each class is found in n-dimensional space - the distance between each pixel and the classes means are found - the pixels are assigned to the class to whose mean they are closest

simple and no unclassified pixels but overly simplistic

Front 3

nearest neighbor

Back 3

calculate the distance from each pixel to each training pixel in dimesional space the class associated weith the nearest training pixel is the one to which the pixel is assigned

simple and no unclassfied pixels but overly simplistic

Front 4

K-Nearest Neighbor

Back 4

selects the K nearest training pixels - weights are assigned based on distance - the class with the greatest total weight is the one to which the pixel is assigned

accounts for diepersion, but reduces the effects of outlier training pixels

Front 5

Parallelepiped

Back 5

class means are found in n-dimensional space - class srandard deviations are found for each band - parallelepipeds are drawn about each class mean - pixels are assigned to the class whose parallelepiped they fall

pixels that are outside of all parallelepipeds are unclassified, parallelepipeds freuentyl overlap causing confusion

Front 6

maximum likelihood

Back 6

based on probability rather than distance - assumes normal distributionof pixel values, procedure is done without weights so the pixel has equal probablity of being assigned to each class

most widely used, no unclassified pixels

Front 7

chain method

Back 7

the distance in a 2D spectral plot between each pixel value and each cluster center is compared, of the distance < D the pixel is included, if the distance is > D then a new cluster is created

no advantages over ISODATA

Front 8

Unserpervised

Back 8

a statistical process automatically finds clusters of like pixels patterns, no training required,
class are idetified later, sometimes resullts are difficult to interpret

Front 9

Supervised

Back 9

you train the computer what a class looks like and the computer finds other pixels that are similiar

Front 10

Training Data vs. Reference Data

Back 10

training data are used to perform the image classification

reference data are used to assess the acuracy of the image classification and should be collected after the classification

you cannot use training dats as reference data - can be collected at the same time but they should be spatially autocorrelated - collected as near as possible as the image acquistion date