双域差异和尺度选择增强的遥感影像变化检测

  • 打印
  • 收藏
收藏成功


打开文本图片集

关键词:遥感变化检测;双域联合差异增强;尺度选择增强;门控差异感知图分类号:TP751 文献标识码:A doi:10.37188/CJLCD.2025-0155 CSTR:32172.14.CJLCD.2025-0155

Abstract: In remote sensing image change detection,continuous downsampling often results in severe information loss and pseudo-change interference.To address these issues,we propose the dual-domain diference and scale selection enhance change detection network (DDSE-Net),which is built upon a dualbranch encoder and U-Net architecture. The proposed framework incorporates three key innovations: (1)A dual-domain joint difference enhancement module(DJEM),which first employs channel atntion to strengthen diference representations along the channel dimension,and further integrates wavelet transform with spatial attention to achieve dual-domain enhancement in both spatial and frequency domains. This design efctively emphasizes true change information while suppressing pseudo changes.(2)A scale selection enhancement downsampling module (SEDM),which captures multi-scale features through convolution and pooling operations of varying receptive fields during the downsampling process.The extracted features are subsequently refined with spatial and channel attention,thereby alleviating information loss. (3)A gated difference perception module (GDPM),which introduces a gating mechanism to adaptively weight and fuse multi-scale change features,enabling more comprehensive integration and enhancing the network's multi-scale representation capability. The proposed DDSE-Net achieves F1-score improvements of at least 4. 48% , 2.18% ,and 1.16% on the WHU,Google,and LEVIRdatasets,respectively,compared with eight mainstream change detection networks,including FC-EF,FC-Conc,IFN,SNUNet,BIT, MSCANet,LightCDNet, and STADE-CDNet, demonstrating the efectiveness of DDSE-Net.

Key words: remote sensing change detection; dual-domain joint diference enhancement; scale selection enhancement;gated difference perception

1引言

遥感变化检测通过比对不同时期同一地区的影像差异,揭示地表信息变化的过程[1]。(剩余18728字)

monitor
客服机器人