Building Extraction from SAR Images Based on Encoder-decoder Network
There are always speckle noise and geometric distortion appearing in synthetic aperture radar(SAR)remote sensing images,which interferes with the building extraction process and results in unclear building boundaries.In view of the above problems,the paper proposes a multi-scale feature attention fusion(MSFAF)network.Firstly,combining the advantages of deep neural networks and SAR images,a spatial attention fusion(SAF)module is designed in the deep layer,where it integrates different levels of features and focuses on important spatial information.Moreover,by applying convolution kernels of different scales and the conversion of channel information,a multi-scale detail extraction(MSDE)module is given to extract feature information of different scales and redistribute channel information,which is beneficial to alleviate the interference problem of speckle noise.The experimental results show that the proposed method has better performance than other existing methods in SAR image building extraction.
SAR remote sensing imagebuilding extractionmulti-scale featureattention fusionneural network
NETL
NSTL
万方数据
