Automatic building detection from remote sensing images plays an important role in a wide range of applications. In this paper, we apply improved U-NET and HF-FCN as main models to detect small building which is more difficult than big building. MUL-Pan Sharpen and PAN data used as the training data. Improved U-NET and HF-FCN were selected as main models. In order to detect small building, we oversample small building areas and under sample large building areas. We adapt morphological methods to dilate and erode output of the mod-el. With the optimization of model’s outputs, we can fill in the disconnected area, but also eliminates part of the false detection noise.
CITATION STYLE
Wang, H., Li, S., Sun, B., Du, R., Zhao, L., Li, W., & Chang, Y. (2019). A deep learning based objection detection method for high resolution remote sensing image. In Communications in Computer and Information Science (Vol. 913, pp. 50–56). Springer Verlag. https://doi.org/10.1007/978-981-32-9987-0_5
Mendeley helps you to discover research relevant for your work.