Synthetic aperture radar (SAR) recognition systems often need to collect new data and update the network accordingly. However, the network faces the challenge of catastrophic forgetting, where previously learned knowledge might be lost during the incremental learning of new data. To improve the applicability and sustainability of SAR target classification methods, we propose a multi-stage regularization-based class-incremental learning (CIL) method for SAR targets, called SCF-CIL, which addresses catastrophic forgetting. This method offers three main contributions. First, for the feature extractor, we fuse the convolutional neural network features with the scattering center features using a cross-attention feature fusion structure, ensuring both the plasticity and stability of the extracted features. Next, an overfitting training strategy is applied to provide clustering space for unseen classes with an acceptable trade-off in the accuracy of the current classes. Finally, we analyze the influence of training with imbalanced data on the last fully connected layer and introduce a multi-stage regularization method by dividing the calculation of the fully connected layer into three parts and applying regularization to each. Our experiments on SAR datasets demonstrate the effectiveness of these improvements.
Reconstructing building models from urban SAR tomography (TomoSAR) point clouds is often constrained by limited resolution, low positioning accuracy in elevation, as well as data incompleteness and artifacts caused by microwave imaging mechanisms. These challenges seriously restrict the extraction of high-accuracy building models with structural details from TomoSAR point clouds. This paper proposes a refined urban building modeling method that effectively utilizes structural priors, including directionality, orthogonality, and potential symmetry. First, a piecewise fitting strategy integrated with density-based segmentation is employed to iteratively estimate the main directions of the buildings and capture finer geometric variations of complex façade footprints than simple-plane approximations. Second, a roof extraction algorithm combining an adaptive Doug-las–Peucker approach with symmetry evaluation and constraints is developed to regularize roof outlines and repair data defects. Crucially, to handle extreme cases where roof data are entirely missing, a novel building width estimation method based on building shadow analysis is proposed. Experiments conducted on the SARMV3D-1.0 and SARMV3D-3.0 point cloud datasets demonstrate that the proposed method significantly enhances reconstruction accuracy and geometric fidelity in urban regions compared to state-of-the-art approaches.