Post-classification of Misclassified Pixels by Evidential Reasoning: a Gis Approach for Improving Classification Accuracy of Remote Sensing Data

Abstract

This paper discusses an approach for extracting supporting evidence from multisource spatial data and by rule-based models to incorporate the evidence with pre-classified Landsat TM data for improving classification accuracy. The process was focused on the extracted "possibly misclassified pixels" (PMPs) only. Based on Dempster-Shafer's theory of evidence, the concepts of homogeneous, heterogeneous. and conflicting evidence and the rules for evidential combination are discussed. Boolean logic, conditional statement, and spatial relationship operations were employed in the models. By running the models, correct labels for the PMPs were judged by pooled evidence from multitemporal Landsat TM data and multisource spatial data.

7 Figures and Tables

Cite this paper

@inproceedings{Wang2010PostclassificationOM, title={Post-classification of Misclassified Pixels by Evidential Reasoning: a Gis Approach for Improving Classification Accuracy of Remote Sensing Data}, author={Yeqiao Wang}, year={2010} }