Abstract:Recent years have witnessed a remarkable breakthrough in synthetic aperture radar automatic target recognition (SAR ATR) with the development of deep learning (DL). Nonetheless, once deployed, the DL-based methods' ability to incrementally learn new knowledge from few-shot samples without forgetting the old is fragile, hindering them from discriminating unseen targets in real-world situations. In this article, we propose a cosine prototype learning (CPL) framework to first unlock few-shot class-incremental learning (FSCIL) in the SAR ATR field inspired by the intrinsic relationships between target azimuth-aware knowledge and semantic features under the cosine criterion. By condensing class-specific characteristics into individual prototypes, stable profiles of targets are depicted without losing generalization. For the model's plasticity, a pairwise structure separation (PSS) loss is introduced to separate old and new classes and compact intraclass features. Meanwhile, the model's transferability on new classes is guaranteed by a prototype consistency (PC) loss. For the model's stability, we propose a prototype–exemplar distillation (PED) loss and a prototype recalibration (PR) strategy to penalize semantic drifts of old-class feature spaces and alleviate the misalignment of the learned prototypes successively. At inference, a nearest-class-mean (NCM) classifier is adopted for evaluation by comparing cosine similarity scores between testing samples and class-specific prototypes. In experiments, the proposed components of our method are explored by ablation studies. Strong baselines are established, and extensive experiments conducted on the moving and stationary target acquisition and recognition (MSTAR) dataset show that our method outperforms state-of-the-art methods under various FSCIL conditions, verifying its effectiveness for the FSCIL of SAR ATR.

Simulated Data Feature Guided Evolution and Distillation for Incremental SAR ATR

Dynamic Embedding Relation Distillation Network for Incremental SAR Automatic Target Recognition

An Evaluation of Representative Samples Replay and Knowledge Distillation Regularization for SAR ATR Continual Learning

SAR Target Incremental Recognition Based on Features With Strong Separability

Multilevel Adaptive Knowledge Distillation Network for Incremental SAR Target Recognition

Incremental SAR Automatic Target Recognition With Error Correction and High Plasticity

Improving SAR Automatic Target Recognition via Trusted Knowledge Distillation From Simulated Data

An Incremental SAR Target Recognition Framework via Memory-Augmented Weight Alignment and Enhancement Discrimination

SAR ATR Method with Limited Training Data via an Embedded Feature Augmenter and Dynamic Hierarchical-Feature Refiner

Incremental Learning Using Feature Labels for Synthetic Aperture Radar Automatic Target Recognition

Incremental Learning Based on Anchored Class Centers for SAR Automatic Target Recognition

Few-Shot Class-Incremental SAR Target Recognition Based on Hierarchical Embedding and Incremental Evolutionary Network.

Class-Incremental SAR Obeject Detection Via Adaptive Distributed Response Distillation

Random Interpolation Data Augmentation for Incremental Automatic Target Recognition

Few-Shot Class-Incremental SAR Target Recognition via Cosine Prototype Learning

Target Recognition in SAR Images by Deep Learning with Training Data Augmentation

Improving Pre-Training and Fine-Tuning for Few-Shot SAR Automatic Target Recognition

Weighted residual network for SAR automatic target recognition with data augmentation

Few-Shot Class-Incremental SAR Target Recognition Via Decoupled Scattering Augmentation Classifier

MFFA-SARNET: Deep Transferred Multi-Level Feature Fusion Attention Network with Dual Optimized Loss for Small-Sample SAR ATR

Layer-Wise Representative Exemplar Selection-Based Incremental Learning for SAR Target Recognition