Theoretical Considerations on Supervised Classification

Study on digital land condition mapPing with remote sensing data (3)

Abstract

A series of supervised technique is the most common procedure in remote sensing data analysis, and is used at the classification step in the digital land condition mapping system (DLC mapping system).
In this paper, theoretical considerations on various types of supervised techniques are reviewed in search of adequate classifiers (decision criteria) in DLC mapping system.
The following items are discussed;
1) Decision criteria and their similarity index
Prerequisite for similarity index and mutual relationships among these indices
Requirements for applicability of decision criteria
2) Refinement of the multi-dimensional feature space
Principal component analysis for information redundancy
Evaluation of feature space by the separability analysis of categories
Dual effects of canonical analysis