Comparison of six fuzzy classifiers on unmixing coarse resolution images (SPOT- VEGETATION): A case study in a tropical forest region of southeast Mexico

摘要

We test the suitability of six soft classifiers on unmixing coarse resolution image (SPOT vegetation) in an area of Southeast Mexico, obtaining information about five classes proportions within a pixel on image. The results show the lowest Root Mean Squared Error (RMSE) average by classifier based on Dempster Shafer theory, and the highest (26%) by Radial Basis Function Neural Network. Pearson correlation coefficient results show that Multilayer Perceptron performs better obtaining a correlation average of 0.55 against a value of 0.34 from the Dempster-Shafer classifier. While RMS values and Correlation coefficients show good performance for classifiers, it is inconclusive to assert that these methods are capable of estimating proportions of classes on a pixel.