In-Process Classification Assessment of Remotely Sensed Imagery
Title:
In-Process Classification Assessment of Remotely Sensed Imagery
Author:
Eastman, J. Ronald Crema, Stefano Zhu, Honglei Toledano, James Jiang, Hong
Appeared in:
Geocarto international
Paging:
Volume 20 (2005) nr. 4 pages 33-43
Year:
2005-12
Contents:
In-Process Classification Assessment (IPCA) is concerned with the analysis and resolution of uncertainties in the classification of remotely sensed imagery. Using a Bayesian weight of evidence procedure known as Dempster-Shafer theory, a method is presented for the mapping of three types of uncertainty: ignorance (the presence of unknown classes), ambiguity arising from inseparable training statistics, and the presence of impure (mixed) pixels. The procedure yields new images of Unknownness, Inseparability and Mixedness that are combined into a color composite for visual analysis. The color composite allows the analyst to identify classes that need improvements in training site quality and locations for new training sites to incorporate missing or mixed classes. In addition, the individual uncertainty images provide a quantitative assessment of classification quality before a post-classification ground truth assessment is undertaken. A case study and validation show that this IPCA procedure produces an accurate reflection of classification quality.