Building extraction based on improved DeepLabV3+model
In respo-nse to the problem of low extraction accuracy in building extraction from re-mote sensing images,this paper constructs an M2-CA-DccpLabV3+model based on the Decp-LabV3+model.The model first uses MobileNetV2 as a feature extraction network,which greatly reduces the number of parameters in the model and improves its feature extraction abili-ty;Secondly,the ability of the network model to learn and extract architectural features is en-hanced through the coordinate attention mechanism.The results show that the model constructed in this article performs better than the original DeepLabV3+model in different regional back-grounds,with an intersection ratio and F1 score of 4.14％and 3.52％higher than the original model,respectively.
building extractionCoordinate attention mechanismSemantic segmentationDeep-LabV3+MobileNetV2
万方数据
维普
