Research on Remote sensing satellite image segmentation based on Swin-Unet
In the past few years,convolutional neural network(CNN)has made great progress in the di-rection of image segmentation.However,due to the limitations of convolution operation,it can not deal with the global and long-distance dependence well.An image segmentation model based on Swin-Unet is proposed.By introducing the Transformer Block module into the Encoder and De-coder stages of the U-Net network model,and using the Dice＿loss loss function,which is more suitable for binary classification,feature extraction and learning are carried out.The Inria Aerial Image Labeling data set for remote sensing satellite image research of urban buildings was used for experiments.Experiments show that the Swin-Unet model can extract more semantic informa-tion from remote sensing satellite images,so as to achieve better recognition effect,and the IoU score is 0.70.
NETL
NSTL
万方数据
