Classification approach

Algorithm

Advantages

Disadvantage

Reference

Pixel-Based

Artificial

Neural

Networks

Manage well large feature space;

Indicate strength of class membership;

Resistant to training data deficiencies-requires less training data than DT

Needs parameters for network design;

Tends to overfit data; Black box (rules are unknown); Computationally intense; Slow training

[46]

Sub-pixel Based

Spectral Unmixing

Clear physical meaning and being able to estimate fractional distribution

Hard to find a proper endmember in larger scale

[60]

Object Based

Support

Vector

Machines

(SVM)

Manages well large feature space; Insensitive to Hughes effect; Works well with small training dataset and does not overfit

Needs parameters: regularization and kernel; Poor performance with small feature space; Computationally intense

[46]