What question did this study set out to answer?

This research aims to tackle challenges in few-shot object detection for remote sensing, focusing on enhancing feature representation and alignment.

March 8, 2026Open Access

DG-Net: Few-Shot Remote Sensing Detection with Dynamic Dual-Stream Collaboration and Generative Meta-Learning

View Full Paper

Ask AI

Bookmark

Authors

This framework enhances few-shot object detection in remote sensing by improving feature alignment and reducing bias.

This research aims to tackle challenges in few-shot object detection for remote sensing, focusing on enhancing feature representation and alignment.
Proposed a Dynamic Relation Dual-Stream Network for support-query feature alignment.
Developed an Optimal Transport-based Generative Meta-Learner for feature distribution bias mitigation.
Incorporated an Orthogonal Frequency Decomposition Head to separate query features and reduce background noise.
Achieved consistent performance gains over leading baseline methods in few-shot scenarios.
Validated effectiveness across various remote sensing datasets and backbone networks.

Liu et al. (2026) studied this question.

View Full Paper

Ask AI

Bookmark