Comparison between accuracy measures of images classified by Maximum likelihood and Artificial Neural Networks

Comparison between accuracy measures of images classified by Maximum likelihood and Artificial Neural Networks
Kamilla A. Oliveira1, Antonio Nuno S. Rosa1,2, Reginaldo S. Pereira1,3, Paulo C. Emiliano4, Gloria S. Almeida1,6 and Fabiano Emmert1,6

1.Universidade de Brasília, Campus Universitário Darcy Ribeiro, CEP 70910-900 , Brasília (kamillarbr@gmail.com1,nuno@unb.br1,2, reginaldosp@gmail.com1,3, gloriaf@gmail.com1,6, fabianoemmert@yahoo.com.br1,7)
4.UniversidadeFederal de Lavras, Campus Universitário, Caixa Postal 3037, CEP 37200-000 Lavras – MG (pequenokaiser2002@yahoo.com.br)

Abstract: This work presents a comparison between the performances of the supervised classification with maximum likelihood algorithm and neural networks in the classification of three LANDSAT/TM sensor scenes in the south of the Amazonas State through Envi 4.6. The maxver classifier obtained better results with the parameters default of the computer system not adopting values for the control of the sample standard deviation. The best results obtained followed the contribution of the internal weight with 0.9 level of activation for the point Training Threshold Contribution and 0.2 for the Training Rate. The experimental results show that the maxver algorithm presented better performance with accuracy over 88% in all the scenes.

Keywords: Neural Networks, Maxver, kappa, accuracy.

AttachmentSize
Oliveira1Accuracy2012.pdf231.58 KB