Semantic Segmentation Based Building Extraction Method Using Multi-source GIS Map Datasets and Satellite Imagery

摘要

This paper describes our proposed building extraction method in DeepGlobe - CVPR 2018 Satellite Challenge. We proposed a semantic segmentation and ensemble learning based building extraction method for high resolution satellite images. Several public GIS map datasets were utilized through combining with the multispectral WorldView- 3 satellite image datasets for improving the building extraction results. Our proposed method achieves the overall prediction score of 0.701 on the test dataset in DeepGlobe Building Extraction Challenge.