ANALISIS METODE SUPPORT VECTOR MACHINE (SVM) UNTUK KLASIFIKASI PENGGUNAAN LAHAN BERBASIS PENUTUP LAHAN PADA CITRA ALOS AVNIR-2

The development of remote sensing technology developed rapidly, especially after the cold war. Remote sensing technology is very well used as the data of land use map-making, because of the higher mapping needs, especially to detect changes in land use. To obtain land use information from remote sensing image takes a special method, especially for remote sensing image processing digitally. One method of remote sensing image processing method is a method of Support Vector Machine (SVM). Methods Support Vector Machine (SVM) is a machine learning method of the class with a method of neural network that can recognize patterns of input or examples given and also belong to the supervised learning. This study aims to analyze the influence of each parameter on the method of SVM and most combinations yield the highest accuracy, and analyze the ability of the method Support Vector Machine (SVM) to manufacture land use map based on 1:100,000 scale land cover. Classification of land use land cover with a method based Support Vector Machine (SVM) using the spectral data, the slope of the spatial data in the form of data and data filters mean texture. Data are the mean texture filter used is the data of each band as well as a composite of all the bands with processing window 3x3, 5x5, 7x7, 9x9. Classification scheme used is a land use classification scheme according BPN 2012 with modifications adapted to the conditions on the ground. The results showed the results of the classification method Support Vector Machine (SVM) spectral data resulted in an overall accuracy and kappa 0.7524 78.8845%. In addition the slope of the data resulted in an overall accuracy of 80.7973% and 0.7755 for the kappa value. Merging data spectral and texture mean filter with 9x9 window processing on the combined bands 1, 2, 3 and 4 turned out to be more and raise the level of the overall accuracy of the classification results into 92.8619% and 0.9163 kappa. While the combination of simulated spectral data, the slope of the data and the data turns out to produce a texture filter higher accuracy, especially if the texture simulations use the mean of all the mean texture band with 9x9 processing window, obtained an overall accuracy up to 92.8951% and kappa reaches 0 , 9170