Skip to content
Open access

Region-Enhanced Gaussian Receptive Field Network for Few-Shot Object Detection in Remote Sensing Images

2026 · IEEE Geoscience and Remote Sensing Letters · Vol 23, pp. 6022805-6022805 · 0 citations · 24 references

Abstract

Few-shot object detection (FSOD) in remote sensing is challenged by discriminative feature dilution and label assignment imbalance due to tiny object sizes and scarce annotations. We propose a region-enhanced Gaussian receptive field (RGRF) network, a unified framework that jointly optimizes feature enhancement and sample assignment. A region-enhanced multilevel criss-cross attention (RE-MLCCA) module adaptively aggregates multiscale features and generates a discriminative spatial mask to guide long-range context modeling, suppressing background clutter. A Gaussian receptive field-guided dynamic label assignment (GRF-DLA) strategy models effective receptive fields (ERFs) and ground-truth boxes as Gaussian distributions, with a few-shot-aware hierarchical scheme that dynamically adjusts positive selection per class. These two modules are tightly coupled for joint optimization. Experiments on DIOR show that RGRF consistently outperforms state-of-the-art methods, achieving gains up to 9.7% in mAP, validating its effectiveness under extreme annotation scarcity.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.