Interferometric Synthetic Aperture Radar (InSAR) enables effective monitoring of volcanic deformation; however, the observed signals are often corrupted by atmospheric phase delays, seasonal surface changes, and decorrelation effects. Existing atmospheric correction methods, such as numerical weather model-based methods, can reduce these effects but do not consistently remove atmospheric artefacts and may introduce residual biases. To address these limitations, we propose a novel learning-based method for denoising unwrapped InSAR interferograms, using a hybrid training strategy that combines physically motivated synthetic deformation with real atmospheric noise. Specifically, we introduce WaveDINO, a wavelet-based multi-scale denoising framework conditioned on frozen DINOv3 foundation-model features and terrain information. Training uses synthetic magma-source deformation superimposed on short-term interferograms to expose the network to realistic atmospheric statistics while retaining known ground truth. Performance is evaluated on both controlled synthetic data and long-term real interferograms from Laguna del Maule (Chile) and Campi Flegrei (Italy), with independent GNSS measurements used for validation. WaveDINO consistently outperforms competing models, improving agreement with GNSS measurements, and reducing mean GNSS misfit by approximately 3% and 19% at two sites, respectively, while surpassing weather-model-based corrections.

翻译：摘要：干涉合成孔径雷达（InSAR）能够有效监测火山形变，但观测信号常受大气相位延迟、季节性地表变化及去相干效应干扰。现有大气校正方法（如基于数值天气模型的方法）虽可减弱这些影响，但无法始终消除大气伪影，并可能引入残余偏差。为解决上述局限，我们提出一种基于学习的解缠InSAR干涉图去噪新方法，采用融合物理驱动合成形变与真实大气噪声的混合训练策略。具体而言，我们引入WaveDINO——一种基于小波的多尺度去噪框架，其以冻结的DINOv3基础模型特征和地形信息为条件输入。训练过程将合成岩浆源形变叠加至短周期干涉图，使网络在保留已知真值的同时暴露于真实大气统计特征。方法性能通过受控合成数据及Laguna del Maule（智利）与Campi Flegrei（意大利）火山的长周期真实干涉图进行评估，并采用独立GNSS测量数据进行验证。实验表明，WaveDINO持续优于对比模型：在两个研究站点分别将GNSS观测一致性提升约3%和19%，并将平均GNSS残差降低至低于天气模型校正效果。