Study on decision tree land cover classification based on MODIS data

Abstract

There are two popular decision tree calculations in the international world - CART and C4.5, and boosting and bagging technology, which are new classification technology in mechanical study field. To study the decision tree and new technologypsilas use in remote sensing classification, we use 250 m resolution data of northeast China to do land cover and classification study. The result shows that a decision tree can improve classification accuracy to better than MLC when there is a large enough training sample, but when there is not enough sample, its performance is worse than MLC. It is also found that, in production of a decision tree, CART is better than C4.5 in classification accuracy and tree structure, while improvement of classification accuracy is up to the construction of tree structure and trimming. When boosting is introduced to CART, the classification accuracy is improved to 25.6% from 18.5%.