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.

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

SAR Incremental Automatic Target Recognition Based on Mutual Information Maximization

Incremental SAR Automatic Target Recognition With Error Correction and High Plasticity

Dynamic Embedding Relation Distillation Network for Incremental SAR Automatic Target Recognition

Multilevel Adaptive Knowledge Distillation Network for Incremental SAR Target Recognition

SAR Target Incremental Recognition Based on Features With Strong Separability

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

Semi-Supervised SAR ATR Framework with Transductive Auxiliary Segmentation

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

SAR Target Recognition Based on Probabilistic Meta-Learning

Simulated Data Feature Guided Evolution and Distillation for Incremental SAR ATR

IncSAR: A Dual Fusion Incremental Learning Framework for SAR Target Recognition

Crucial Feature Capture and Discrimination for Limited Training Data SAR ATR

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

Decoupled Self-Supervised Subspace Classifier for Few-Shot Class-Incremental SAR Target Recognition

Open Set Recognition and Category Discovery Framework for SAR Target Classification Based on K-Contrast Loss and Deep Clustering

Exploring Reinforced Class Separability and Discriminative Representations for SAR Target Open Set Recognition

Optimizing Target Recognition in Synthetic Aperture Radar Imagery: A Hyperparameter- Tuned Approach With Iterative Transfer Learning and Branched-Convolutional Neural Network

Weighted residual network for SAR automatic target recognition with data augmentation

SAR Target Recognition via Meta-Learning and Amortized Variational Inference

Sar Target Recognition Via Supervised Discriminative Dictionary Learning and Sparse Representation of the Sar-Hog Feature