SAR-RAG: Enhancing SAR Target Recognition with Retrieval-Augmented MLLMs

Researchers have introduced SAR-RAG, a novel AI agent designed to improve Automatic Target Recognition (ATR) in Synthetic Aperture Radar (SAR) imagery. SAR is critical for defense and security applications, particularly for detecting military vehicles that often appear indistinguishable in radar images. The proposed method integrates a Multimodal Large Language Model (MLLM) with a vector database of semantic embeddings, creating an Image Retrieval-Augmented Generation (ImageRAG) system. This architecture allows the AI to perform contextual searches through a library of known image exemplars, comparing test examples against verified target types to enhance differentiation and identification accuracy. The study evaluates the system using search metrics, categorical classification accuracy, and numeric regression of vehicle dimensions. Results indicate significant improvements over baseline MLLM methods when SAR-RAG is employed as an attached memory bank. This advancement leverages agentic AI tools to refine neural network capabilities, offering a robust solution for complex remote sensing tasks in military monitoring and security operations.