Abstract: SEaTH algorithm is an object-oriented feature selection algorithm based on inter-class distance.We present an improved SEaTH—ISEaTH.It overcomes some limitations of existing SEaTH algorithms.The ISEaTH algorithm evaluates the features according to the relation between features,inter-class distance and intra-class distance respectively,and integrates multiple estimation results to obtain the best feature subset.Compared SEaTH and ISEaTH with QuickBird data in Kashi area of Xinjiang province,the experimental results show that the ISEaTH algorithm not only reduces feature dimensions,but also improves the classification accuracy,thus is a much more efficient object-oriented feature selection algorithm.