Application of improved U-Net model in building extraction from remote sensing images
This paper proposed an improved model based on the U-Net model to address the issues of low accuracy and poor image edge prediction performance of traditional building extraction methods for remote sensing images in extracting complex background images. Firstly,to prevent overfitting,a random dropout function was added to the U-Net shrinkage path. Secondly,to improve the training speed of the model,batch normalization layers were added to the expansion path. Finally,in order to improve the image edge prediction performance of the model,the joint loss function was selected as the model loss function. Through experiments on the Wuhan University(WHU) building dataset,the results show that the model presented in this paper performs well in building extraction integrity and boundary segmentation accuracy,especially for smaller buildings. The accuracy indicators UIo and AO,as well as the Kappa coefficient reach 76.876％,91.413％,and 81.225％,respectively,which are better than the accuracy indicators of the comparative model. This verifies the reliability of the method proposed in this paper.
remote sensing imagesimprove the U-Net modelbuilding extractionjoint loss functionrandom dropout function
万方数据
维普
