Abstract:

The pixel- and object-level classification methods are investigated separately, while the hybrid of them has not been explored. This paper tries new exploration on the hybrid of pixel- and object-level classification, and proposes synthesis classification method for remote sensing image based on improved pixel- and object-level classification. Firstly, an improved RBF neural network classifier is proposed to obtain the pixel-level classification result, and a hierarchy classification model based on improved fuzzy support vector machines and decision tree is utilized to obtain the object-level classification result. Then a specific synthesis algorithm of pixel- and object-level classification is proposed to obtain the synthesis classification result. The experiments show the synthesis classification method can improve the accuracy of classification result effectively and provide more accurate classification result than single pixel- or object-level method.

Key words: pixel-level classification, object-level classification, synthesis classification, hierarchy classification