Lecture 10 - Supervised image classification Flashcards
Supervised Classification
- A procedure for identifying spectrally similar areas on an image by identifying ‘training; sites of known targets and then extrapolating those spectral signatures to other areas of unknown targets.
What does it require?
Training data on areas you want to map/classify
Stages of Supervised
- User-defined land cover classes
- Training site selection
- Generation of statistical parameters from the training sites
- Classification
- Accuracy Assessment
Training site selection based on:
Field visits
High spatial resolution data
Previous maps
Investigator/expert knowledge
What should be done when collecting training site data?
Select homogenous training sites to ensure class spectral seperability
Different types of Classification
-Paralellpiped classifier
- Minimum distance to means classifier
- Maximum likelihood classifier
Need rules to decide which class to put a given pixel into
What happens in Parallelepiped classifier (BOX)
Assign boundaries around the spread of class in feature space
All pixels in the image with values within designated parallelepiped will be classified in that spectral class
Bounds are usually determined from training sets
What happens in Minimum distance to means (MDM)
- Same approach as unsupervised classification clustering method.
- Calculate mean spectral value for each training set in each bands
- Put every unknown pixel onto nearest class
- Compute the distance between value of unknown pixel and each category
- Determine a limit beyond which a pixel remains unclassified
What happens in Maximum likelihood classifier?
For each pixel to be classified:
Assumes data in a class are (unimodal) Gaussian (normally) distributed
The probability of classification is calculated for each class based on the training data
The pixel is classified as the class with the largest probability
Theoretically the best classification
Supervised Advantages
Analyst has control
Processing is tied to specific areas of known identitty
Operator can detect errors in training data
Disadvantages of Supervised
Training classes based on field identification not spectral properties
Human error in training data
What is accuracy assessment?
Collecting reference data “Ground Truth”
Compare this to classified map
Ground Truth data sources
High spatial resolution data
Field visits
GIS layers
Accuracy Assessment Reference Data Sampling techniques
Random -
Stratified
Stratified random
Random Sampling
Observations are randomly placed
