Weighting fidelity versus classified area in remote sensing classifications from a pixel and a polygon perspective

Pere Serra 1, Gerard Moré 2 and Xavier Pons 1,2
1 Department of Geography 
Edifici B, Campus de la Universitat Autònoma de Barcelona, 08193-Cerdanyola del Vallès (SPAIN).
Tel.: 00 34 93581 3273; Fax: 00 34 93 581 2001
pere.serra@uab.cat
2 Center for Ecological Research and Forestry Applications (CREAF)
Edifici C, Universitat Autònoma de Barcelona, 08193-Cerdanyola del Vallès (SPAIN).
Tel.: 00 34 93 581 1312; Fax: 00 34 93 581 4151
g.more@creaf.uab.cat; xavier.pons@uab.cat

Abstract
This paper summarizes the consequences in the area classified and in the thematic accuracy of being more or less conservative in a hybrid classifier. The most important parameter of that classification consists in fidelity (the introduction of the threshold proportion at which to accept a spectral class as being a part of a thematic category). Two options have been tested: the first less conservative, the second more conservative. These fidelities have been applied to ten Mediterranean crops and tested using error matrices. Thematic accuracies were quantified following the classical approach (number of pixels correctly classified), a polygon approach (number of polygons correctly classified)  and, finally, area approach (area correctly classified). Results showed that the most restrictive fidelity produces less area classified but with more thematic accuracy when unclassified pixels are not included in the quantification of the accuracy. This fact occurred in all the options (pixel, polygon and area) although did not affect all the crops equally.

Keywords: hybrid classifier, fidelity, producer’s accuracy, Mediterranean crops, area classified

In: Caetano, M. and Painho, M. (eds). Proceedings of the 7th International Symposium on Spatial Accuracy Assessment in Natural Resources and Environmental Sciences, 5 – 7 July 2006, Lisboa, Instituto Geográfico Português

AttachmentSize
Serra2006accuracy.pdf738.41 KB