Remote sensing building change detection based on DeepLabV3+ twin network
To address the situation that small buildings are easily missed and the accuracy of building contour boundary detection is insufficient at multiple scales,a dual-channel twin network based on DeepLabV3+ is proposed in this paper.First,in order to improve the accuracy of segmentation results and avoid the overfitting problem of the model caused by the deepening of network layers,the improved ResNeXt50(32×4d)is used as the backbone network to extract features and optimize the computational efficiency of the model;second,to address the problem of inadequate feature fusion in the twin network,an attention-based dual-channel fusion module is designed;in addition,to improve the overall model Finally,feature alignment module and fully connected CRF are introduced in the feature recovery stage to further complement and refine the segmentation results.The precision,recall and F1 indices reach 0.923 3,0.899 4 and 0.911 2,respectively,on the LEVIR-CD dataset.
NETL
NSTL
万方数据
维普
