Multispectral Classification

Question 1

Multispectral Image classification

Answer 1

pixels assigned to classes

data from a spectral band for each pixel

group together pixels with similar data and put them in a class

these classes will form regions on the image of pixels with similar attributes

Question 2

Multi-spectral Classification: unsupervised mode advantage

Answer 2

-Requires no prior knowledge of area
-Automatic
–no human error
-Small, unique areas are recognised as distinct units

Question 3

Multi-spectral Classification: unsupervised mode disadvantage

Answer 3

• Classes may not be useful to geologist
• Limited control by analyst
• Will only work for that image
• An adjacent image will generate different classes that can’t be matched

Question 4

what is unsupervised mode

Answer 4

plot reflectance in a channels against reflectance in the other channel

2 channels = 2d space
3 channels = 3d space
7 channels = 7d space

Question 5

how does person use unsupervised mode

Answer 5

User selects number of classes they want (here 2)

• Computer looks for that number of clusters in data…
• … then decides which cluster a pixel falls in based on spectra
• … then assigns each class a colour and plots classified image

Question 6

Multi-spectral Classification: supervised mode advantage

Answer 6

• User control over classes
• Classes tailored to specific purpose
• Tied to known identity
• Do not have to match categories to ground truth as with unsupervised mode (as inherently done)
• Check of validity as training areas should be properly classified in final image

Question 7

Multi-spectral Classification: supervised mode disadvantage

Answer 7

• User classes may not match natural classes so may not be distinct in multispectral space (e.g. Rhyolite-Microgranite-Granite)
• Training areas might actually be quite diverse a class may cover a wide range of sub-types (e.g. Sandstone)
• Training data may not represent the conditions across the whole image
• Setting up / verifying training areas time-consuming
• May not recognise special or unique categories not represented in the training data

Question 8

what is supervised mode

Answer 8

samples of known identity (pixels already assigned to a class) to classify other pixels

Samples of known identity are from training areas where the ground is known