Abstract: In order to improve the classification accuracy of wetland classification through satellite remote sensing, it is necessary to overcome the “same object with different spectrum, different object with the same spectrum” problem which exists in the wetland classification of high spatial resolution satellite image. The research explores the combination of the Chinese sub-meter-level GF-2 image, which has a broad application prospect, and the object-based image analysis classification algorithm based on classification and regression tree. Then, the classification and extraction of wetlands in the Dongting Lake district were carried out by using the Yuanjiang city in Hunan as an example. Multi-dimensional object features including spectral information, geometric features, terrain features and texture features were selected to train the classifier. Constructed a set of wetland classification methods of classification and regression tree based on GF-2 image. Accuracy evaluation data shows that the overall classification accuracy of the method reaches higher value. The results show that the method can provide ideas for the classification of Dongting Lake district wetland based on GF-2 image.

Key words: wetland, GF-2, classification and regression tree, object-based, classification features, Dongting Lake district