Comparison of Buildings Extraction Algorithms Based on Small UAV Aerial Images

摘要

Using small rotor UAV, this paper acquired the images of qionghalajun village of the artushi city in Xinjiang autonomous region. Then, two image analysis methods of object-oriented and pixel-oriented methods, respectively, were used to extract the building of the study area. It results show that the object-oriented method can effectively remove the impact of salt and pepper noise on classification and ensure the integrity of the shapes of the buildings. But the impact of different objects with similar spectral and texture information will decrease the classification accuracy. In the pixel-oriented extraction method, the improved mathematical morphology algorithm is added, which can effectively improve salt and pepper noise, maintain the continuity and integrity of the edge of the building, and solve the problem of partial farmland and houses in the object-oriented method.