Section 7 Flashcards
What are the four steps of a classification?
Select training areas
Build spectral signatures
Classification
Accuracy assessment
Name the two approaches to digital classification.
Supervised
Unsupervised
What is a supervised classification?
You determine the class boundary based on training sites
What is an unsupervised classification?
The class boundaries are determined automatically
What is the main source of error of classification?
Miss class identification by user
What is a training site?
The sample areas a user selects for the classes to be based off of.
How do you decide to use supervised or unsupervised?
If there is previous knowledge of the site use supervised. If not unsupervised
In an unsupervised classification what are the two inputs by the user?
The preferred cluster algorithm
Number of classes
Name the three parametric classification algorithms.
Minimum distance to mean
Parallelpiped
Maximum likelihood
Name the three non parametric classification algorithms.
Spectral mixture analysis
Spectral angle mapper
Support vector machine
How does the minimum distance to mean algorithm work?
Relies on the straight line (Euclidean) distance from the class means to the unclassified pixel. It is simple and efficient
What is the downside to the minimum distance to mean?
Insensitive to different degrees of variation in spectral response of data
How does the parallelepiped algorithm work?
It is the simplest method, known as the box classifier because it uses one to classify minimum and maximum ranges
What are the problems with the parallelepiped?
When boxes overlap the box the is over top will claim the entire area. Like coding
What is a stepped parallelepiped?
It is an improvement in the boxes are stepped to reduce overlapping and make a tighter classification
What is the maximum likelihood classifier?
It uses normal distribution to make a statistical probability of the pixels location. It assigns pixels a class based on the max probability
What is a negative of the maximum likelihood classifier?
The assumption that data is normally distributed and it is computationally intense
Describe the spectral mixture analysis (SMA) algorithm.
It decomposes a pixel to look at the spectral signature. It can be linear or non linear
Describe the SMA linear.
All reflective elements contribute to its representation. It is additive
Describe the SMA non linear.
Looks at reflective components proportionally
Describe the spectral angle mapper (SAM) algorithm.
It compares spectral signatures, it is insensitive to intensity
Describe the support vector machine (SVM) algorithm.
Similar to maximum likelihood, it yield high classification results
What is an error confusion matrix?
An accuracy assessment that does a class by class basis for relationships.
Where is the optimal place reference data should be collected?
The field
What are the five sampling strategies?
Systematic Stratified Random Stratified random Cluster resampling
What is the systematic sampling strategy?
Uniform sampling that uses a pattern like a grid. This is not very useful.
What is stratified sampling strategies?
Sample based on predetermined strategy based on location or class
Describe random sample strategy.
Yields statistically selected sample with no bias
Describe the stratified random sample strategy.
Most effective
What are the downsides of cluster sampling?
It is only used when limited access to site and there will be bias spatial results
What is a kappa coefficient?
A normalization process used to standardize error matrix results so they can be compared
What is the key goal of a classification?
Simplification or reduction in the complexity of a system into something meaningful for the observer
