Underdeveloped village extraction from high spatial resolution optical image based on GLCM textures and fuzzy classification

In this paper, we proposed a new strategy to extract underdeveloped villages from high spatial resolution optical image based on Grey Level Concurrence Matrix (GLCM) textures and fuzzy classification. Experiments were carried out using 126 samples with size of 300m*300m extracted from 1m spatial resolution optical image in Beijing and Tianjin in China. Results showed that our method can largely extract underdeveloped villages with the highest overall accuracy of 93.3% in our experiments. Also homogeneity is the most effective feature in all four selected GLCM textures in our approach. Multi-scale features are more stable under various training samples. The main difficulty lays in discriminating underdeveloped villages from other residential areas and also boundary areas connecting different types of land cover/use. However, our approach is very flexible, in which each class can only use features characterizing it well, so more scales and textures may be included to improve the result in the future.