Landcover classification of remotely sensed data has found many useful applications in industries such as forestry, agriculture, and defense. With the push toward end users, class maps are often incorporated directly into geographical information systems for use in solving large, complex problems. However, errors are inherent in the classification process. The importance of assessing the thematic accuracy of data derived from remote sensing platforms is universally recognized and has motivated much research. Classification accuracy assessment is often required to determine the "fitness of use" or suitability of a data set for a particular application. Failure to identify the magnitude of inaccuracies in classified data can result in errors cascading into subsequent exploitation and eventually result in false conclusions or flawed products. Many different techniques have been developed and utilized by the remote sensing community for performing thematic accuracy assessment. To date, no one procedure has been adopted as an industry-wide standard. The purpose of this research was to evaluate the effectiveness and compare the results of several state-of-the-art assessment techniques. Synthetically generated imagery, along with real multispectral line scanner data, served as the baseline for the comparison. Synthetic imagery is uniquely suited for this task because the exact classification accuracy can be determined.
Library of Congress Subject Headings
Geography--Remote sensing--Data processing; Remote sensing--Data processing; Cartography--Data processing; Classification--Data processing; Pattern recognition systems--Evaluation
Department, Program, or Center
Chester F. Carlson Center for Imaging Science (COS)
Allen, Jeffrey R., "Methods of digital classification accuracy assessment" (1997). Thesis. Rochester Institute of Technology. Accessed from
RIT – Main Campus