Image classification- supervised Flashcards
What is supervised classification?
A method used for identifying spectrally similar areas on an image by identifying ‘training’ sites of known targets and then extrapolating this to other areas of unknown targets
What are the requirements for training data?
Areas representing the land cover types you want to map
Similar land cover type (homogenous)
Large enough to capture the variation in the image
What is a classification method?
A model that uses different statistical rules to classify the data
What are the 3 different classification methods?
Parallelepiped
Minimum distance to means
Maximum likelihood
What is the main difference between supervised and unsupervised classification?
Supervised classification uses training data that organises the data into specific classes e.g. land cover type, tree species type
What are the 5 stages of supervised classification?
- User-defined land cover class
- Training site selection
- Generation of statistical parameters from the training site
- Classification
- Accuracy asssessment
How do you ‘train’ data?
Select data based on field visits, high spatial resolution data (google earth), and previous maps
Select multiple areas for each cover type in the image
What do you do after you obtain the training data?
Generate the spectral statistical parameters by:
Getting the mean spectral profile for all the land cover types
They need to have spectral separability- no overlaps of pixels so the training data can accurately classify the data in the image
Why use the parallelepiped classifier (BOX)?
Assigns boundaries around the spread of a class feature space i.e. TAKE ACCOUNT OF SPECTRAL VARIANCE
Pixels of the same values cluster together
Look at cluster mean and classifies accordingly
Why use the maximum likelihood (ML)?
MOST COMMON
Assumes data in a class is normally distributes
Probability of classification is calculated for each class based on the training data
Pixel is classified as the class with the LARGEST PROBABILITY
What are the advantages of supervised classification?
Analyst has control- provides training data
Processing is tied to specific area of known identity
Dont need to match categories with field info on final map
What are the disadvantages of supervised classification?
Training classes are BASED ON FIELD IDENTIFICATION NOT SPECTRAL PROPERTIES
Analyst may not choose training data that is representative
Unable to recognise unique categories that aren’t in the training data
What is the purpose of an accuracy assessment?
Assess how well land cover map represents the real environment
