Study on extraction land use/cover information from Landsat ETM+ images by combining classifier

Remotely sensed images are principle data source in researches of land use and land cover. Despite many methods are developed and the classification accuracy is steadily enhanced, they cannot fast classify images and are unsuitable for regular operation to process mass data. We develop a combining classifier based on unsupervised classifier and decision tree classifier to process an image of Landsat ETM+ under ERDAS IMAGINE 8.7. Then we compare the classification results from the combining classifier, unsupervised classifier and decision-tree classifier in two aspects: classification accuracy and time efficiency. The comparison shows that this combining classifier is not only optimal with a total accuracy 87.6%, kappa coefficient 0.853, but also encouraging in time efficiency. It is concluded that this method is simple and applicable, suitable for regular operations of mass images classification.