Teaching large language models to see in radar: aspect-distributed prototypes for few-shot HRRP ATR

TL;DR

This paper introduces an aspect-distributed prototype strategy to improve few-shot radar target recognition using large language models, addressing overfitting and generalization issues by considering aspect variance.

Contribution

It proposes a novel aspect-distributed prototype approach that enhances LLM-based ATR robustness in few-shot scenarios, outperforming existing benchmarks.

Findings

Significant performance improvement over benchmarks in simulated and measured datasets.

Enhanced robustness to aspect variance in few-shot HRRP ATR.

Effective utilization of aspect distribution for better generalization.

Abstract

High-resolution range profiles (HRRPs) play a critical role in automatic target recognition (ATR) due to their richinformationregarding target scattering centers (SCs), which encapsulate the geometric and electromagnetic characteristics of thetarget.Under few-shot circumstances, traditional learning-based methods often suffer from overfitting and struggle togeneralizeeffectively. The recently proposed HRRPLLM, which leverages the in-context learning (ICL) capabilities of largelanguagemodels (LLMs) for one-shot HRRP ATR, is limited in few-shot scenarios. This limitation arises because it primarilyutilizesthe distribution of SCs for recognition while neglecting the variance of the samples caused by aspect sensitivity. Thispaperproposes a straightforward yet effective Aspect-Distributed Prototype (ADP) strategy for LLM-based ATRunder few-shotconditions to enhance aspect robustness. Experiments conducted on both simulated and measured aircraft electromagneticdatasets demonstrate that the proposed method significantly outperforms current benchmarks.