DiffoScope fusion and cooperative perception for detection of building changes
Given that the current building change detection methods have problems in edge detection blurring and capturing global information processing,and small target leakage detection,a building change detection method with diffoscope fusion and cooperative perception was proposed in this paper.The twin encoder was used to extract the image features and the image was restored by the differential decoder.On the one hand,a diffoscope fusion module was proposed,combining the differential and multi-scale technology conjunction,effectively extracting the change objects of different scales in the diachronic feature maps,and using the jump connection to connect them with the decoder;on the other hand,a cooperative perception module was designed,using the abstract features in the deep level to adjust the shallow feature maps in the decoder;And after the decoder outputs the feature maps,the effective channel attention module was used to further improve the detection performance of the model.To evaluate the effectiveness of the method,experiments were conducted on the LEVIR-CD and WUH-CD datasets,respectively,and the experimental results showed that it performed optimally on F1 and IoU compared to other building change detection methods,reaching 91.34％,84.06％,and 91.43％ and 84.21％,respectively,which indicated that the method achieved a significant performance in the building change detection task improvement.
万方数据
维普
