Study on remote sensing classification of Ejina oasis landscape

The accuracy improvement of remote sensing images has been always one of preceding issues in the filed of remote sensing. maximum likelihood to new decision tree, neural networks, wavelet transform and fuzzy recognition system, the classification accuracy has been improved. In this work, based on 3S technology and Landsat ETM data, using 3 classification methods (K-Mean, maximum likelihood and decision tree), Ejina oasis landscape is classified into 6 types: farmland, forestland, grassland, waters, saline land and desert; and the accuracies of the classification results are analyzed. The results indicate that the built decision tree model through NDVI, K-T transformation and principal component analysis is reasonable; the overall accuracy is up to 93.28%, higher than that of K-Mean (86.32%) and maximum likelihood (85.82%) obviously. It can provide scientific basis for the ecological health dynamic monitoring and regional sustainable development of Ejina oasis.