RadarPLM: Adapting Pretrained Language Models for Marine Radar Target Detection with Preference-aware Loss

TL;DR

RadarPLM addresses marine radar target detection under sea clutter by fine-tuning pretrained language models with a lightweight adaptation module and a novel preference-aware loss that emphasizes informative, generalizable feature patches. The approach combines five sequence radar features into patches, LoRA-based parameter-efficient updates, and a reference-model–driven token reweighting scheme, followed by autoencoder-based head retraining. Empirical results on IPIX data show significant gains in detection rate, especially under low SCR and data-scarce settings, while maintaining practical latency. The work demonstrates the viability of PLMs as general-purpose solvers for radar signal processing and suggests directions for scaling and compression to enable real-time deployment.

Abstract

Recent advances in pre-trained language models (PLMs) have demonstrated their capabilities in capturing universal knowledge, making them promising applications for radar signal processing. Nevertheless, directly fine-tuning PLMs on radar signals is both computationally expensive and prone to overfitting, particularly in low signal-to-clutter ratio (SCR) environments. In this paper, we propose a novel fine-tuning framework for PLM-based marine radar target detection. First, we design a lightweight adaptation module, enabling parameter-efficient fine-tuning while preserving the pretrained model's general knowledge. Second, a novel preference-aware loss is developed to selectively optimize different feature patches based on their online evaluated learning values, guiding the model to concentrate on the most generalizable feature patterns during optimization. Extensive experiments on real-world marine radar datasets demonstrate that the proposed finetuning framework achieves an average performance improvement of 9.9% over the standard approach under low SCR conditions. Furthermore, the fine-tuned model, RadarPLM, consistently outperforms state-of-the-art detectors, particularly when training data are limited.

Figures (7)

• Figure 1: Overview of our RadarPLM framework.
• Figure 2: Overview of the reference model.
• Figure 3: An illustration of the patching operation.
• Figure 4: IP, DSE, SMS, Amp, and DP features for target and sea clutter echo signal on IPIX database.
• Figure 5: Illustration of the LoRA fine-tuning process.