High-Resolution Range Profile (HRRP) radar recognition suffers from data scarcity challenges in real-world applications. We present HRRPGraphNet++, a framework combining dynamic graph neural networks with meta-learning for few-shot HRRP recognition. Our approach generates graph representations dynamically through multi-head self attention (MSA) mechanisms that adapt to target-specific scattering characteristics, integrated with a specialized meta-learning framework employing layer-wise learning rates. Experiments demonstrate state-of-the-art performance in 1-shot (82.3%), 5-shot (91.8%), and 20-shot (94.7%) settings, with enhanced noise robustness (68.7% accuracy at 0 dB SNR). Our hybrid graph mechanism combines physical priors with learned relationships, significantly outperforming conventional methods in challenging scenarios.
High Resolution Range Profiles (HRRPs) have become a key area of focus in the domain of Radar Automatic Target Recognition (RATR). Despite the success of deep learning based HRRP recognition, these methods needs a large amount of training samples to generate good performance, which could be a severe challenge under non‐cooperative circumstances. Currently, deep learning based models treat HRRPs as sequences, which may lead to ignorance of the internal relationship of range cells. This letter proposes HRRPGraphNet, a novel graph‐theoretic approach, whose primary innovation is the use of the graph‐theory of HRRP which models the spatial relationships among range cells through a range cell amplitude‐based node vector and a range‐relative adjacency matrix, enabling efficient extraction of both local and global features in noneuclidean space. Experiments on the aircraft electromagnetic simulation dataset confirmed HRRPGraphNet's superior accuracy and robustness compared with existing methods, particularly in limited training sample condition. This underscores the potential of graph‐driven innovations in enhancing HRRP‐based RATR, offering a significant advancement over sequence‐based methods.
This letter introduces a pioneering, training-free and explainable framework for High-Resolution Range Profile (HRRP) automatic target recognition (ATR) utilizing large-scale pre-trained Large Language Models (LLMs). Diverging from conventional methods requiring extensive task-specific training or fine-tuning, our approach converts one-dimensional HRRP signals into textual scattering center representations. Prompts are designed to align LLMs’ semantic space for ATR via few-shot in-context learning, effectively leveraging its vast pre-existing knowledge without any parameter update.
26 May 2026 in IEEE Transactions on Aerospace and Electronic Systems
High-resolution range profile (HRRP) provides compact one-dimensional electromagnetic scattering representations but faces challenges due to signal ambiguity and the scarcity of labeled data for unseen target classes. Recently, Large Language Models (LLMs) have demonstrated potential by leveraging pre-trained semantic knowledge via In-Context Learning (ICL). However, a fundamental modality gap persists: the semantic space of frozen LLMs, evolved from natural language, is inherently misaligned with the non-Euclidean manifolds of electromagnetic scattering, resulting in a lack of sensitivity to fine-grained signal discriminators. To bridge this gap via manifold alignment, we propose Scattering-aware Episodic Adaptation via Semantics (SEAS), a unified framework that enhances the reasoning capabilities of LLMs with domain-specific signal alignment. Specifically, we conceptualize a hybrid episodic adaptation mechanism that combines the implicit, attention-based reasoning of frozen LLMs with an explicit, gradient-based outer loop facilitated by Low-Rank Adaptation (LoRA). To bridge the gap between radar signals and textual semantics, we introduce the Scattering-Aware Episodic Construction (SEAC) pipeline, which transforms HRRP measurements into fidelity-validated instruction episodes. Extensive empirical validation demonstrates that SEAS effectively aligns internal semantic representations with radar signal manifolds. Notably, SEAS enables Qwen3-8B to outperform current LLM-based HRRP ATR baselines including DeepSeekV3.2-Exp and Qwen3-max, while our scaled-up variant achieves a 31.6% relative improvement over the strongest baseline.
11 February 2026 in IEEE Transactions on Signal and Information Processing over Networks
In recent years, radar automatic target recognition (RATR) methods employing deep-learning models with High Resolution Range Profiles (HRRPs) have demonstrated significant advancements over traditional approaches. Nevertheless, these deep architectures exhibit critical performance degradation in few-shot scenarios due to severe overfitting caused by limited training samples. To address this challenge, we propose Gramian Angular Field-based Meta-Learning Graph Neural Network (GAF-MLGNN) - an efficient few-shot HRRP recognition framework that makes full use of both inter-sample information of HRRP sequences and the intra-sample information between HRRP range cells. To fully leverage inter-sample information of HRRPs for higher recognition accuracy under few-shot circumstances, this paper introduces Graph Neural Network (GNN) and proposes a novel task set-based meta-learning framework MLGNN (Meta-Learning for Graph Neural Network) that fits the characteristic of GNN to obtain optimal parameters. Besides, we incorporate the technique of Gramian Angular Field (GAF) for efficient utilization of HRRP intrinsic intra-sample information. Experiments conducted on simulation datasets demonstrate the superior performance of the proposed method compared to current state-of-the-art meta-learning and few-shot HRRP RATR approaches.
20 March 2026 in IEEE Transactions on Neural Networks and Learning Systems
Few-shot learning (FSL) enables deep learning models to generalize to unseen categories with minimal labeled data, making it crucial for data-constrained domains such as healthcare. However, existing FSL models often lack explainability, obscuring the reasoning behind their predictions and limiting trust in their deployment. In this work, we introduce contrastive cross-class attribution (C3A) to enhance explainability in FSL. Specifically, we present C3A for explaining query predictions and its variant (C3A-E) for episodic learning. Unlike prior explainability methods designed for specific architectures, our approach is classifier-agnostic, leveraging the backbone network to ensure broad applicability across diverse FSL paradigms. In addition, we propose contrastive local descriptors that integrate naive-Bayes nearest-neighbor (NBNN) and Fisher score-based attribution to preserve fine-grained details and enhance interclass discrimination. Finally, we comprehensively evaluate our methods across four FSL benchmarks, demonstrating consistent improvements over state-of-the-art approaches, with absolute gains of 19.72% in insertion-area-under-curve (iAUC) and 25.88% in deletion-area-above-curve (dAAC). Our posthoc design ensures seamless integration with various architectures, including Conv64F, ResNet12, ResNet18, and ViT. The experimental results confirm that C3A and its variant (C3A-E) provide more stable and interpretable explanations compared to existing FSL explainability methods, offering deeper insights into the episodic learning process. We have made our code publicly available at https://github.com/MountainChenCad/C3A