The application of the method combine prior knowledge and fuzzy adaptive resonance theory map in remote sensing classification

Abstract

The paper combines prior probability and a fuzzy adaptive resonance theory map to remote sensing classification. Compared to the method using prior knowledge as an additional band in the fuzzy adaptive resonance theory map, both of the methods improve the accuracy of classification significantly. The effect of the method based on prior probability is better. The test results prove that prior probability plays an important role in classification. Both normal distribution statistical classification and fuzzy adaptive resonance theory map improve the accuracy significantly, elevates 8.6% and 10.4% separately on overall accuracy, 0.106 and 0.129 separately on the Kappa coefficient. The method using prior knowledge as an additional band in classification improves the classification accuracy compared to spectral classification, it elevates 7.3% on overall accuracy, 0.095 on the Kappa coefficient. However, its effect is worse than that of the method based on prior probability. The test results prove that the fuzzy adaptive resonance theory map has priority over normal distribution, spectral classification, it elevates 1.3% on overall accuracy and 0.011 on the Kappa coefficient, using prior probability, it elevates 3.1% on overall accuracy and 0.034 on the Kappa coefficient.