METODE DEFUZZIFIKASI ARTIFICIAL NEURAL NETWORK UNTUK KLASIFIKASI PENGGUNAAN LAHAN

Abstract

Artificial neural network is a non-parametric approach which does not require class statistical distribution to be described normally, the advantages of non-parametric approach are its ability to combine septral data and non-spectral data, which is assumed by adding non-spectral data, the resulting accuracy also increased. The method used in this study is a fuzzy logic-based approach with defuzzification using an artificial neural network algorithm. The input spectral data were taken from the ALOS AVNIR-2 image and non-spectral data in the form of slope, soil solum and image texture. This study resulted in an overall accuracy and kappa index that was not good or proper, however the results were still acceptable. The results of the overall accuracy calculation for the 14 land use classes of the artificial neural network defuzzification process involving spectral data only (4 bands ALOS AVNIR-2) were 55% and the kappa index was 0.51. Meanwhile, when non-spectral data such as slope slope, soil solution and image texture are involved as input, the accuracy decreases in 14 land use classes, with an overall accuracy of 40% and a kappa index of 0.35 where the input slope and soil solum are interpolated in the software. ArcGIS first, while the one that was directly interpolated in the Idrisi Selva software, the overall accuracy was 49.75% and the kappa index was 0.45%.