Abstract: In this study,four kinds of object-oriented methods,including nearest neighbor method,a member function method,support vector machine and decision tree,are used for forest classification with SPOT5 image in Huitong county of Hunan Province.As the actual forest classes in Huitong county,6 forest classes and 6 non-forest classes were extracted in this study,and the classification hierarchy is also constructed.By comparing the forest classification results of the four object-oriented methods,it is found that the nearest neighbor method performed the best for forest classification,especially for those forest classes with similar object features,and it is more suitable for extracting forest classes in hilly area,its classification accuracy can reach 76.12%(12 classes),its kappa coefficient can reach 0.73(12 classes)in the mountainous and hilly areas of southern China,which are obviously higher than those of other methods.

Key words: object-oriented method, hilly mountain area, forest classification