A drone aerial tiny-object detection algorithm with improved detection accuracy but halved parameter usage

doi:10.1016/j.patrec.2026.09.001

A drone aerial tiny-object detection algorithm with improved detection accuracy but halved parameter usage

2026-09-04

0

PRE

AI

DOI：10.1016/j.patrec.2026.09.001

Original

Original request for help

Save

Abstract

En 中文

Key findings and contributions of this work:. • SAM-DETR: a lightweight UAV aerial tiny-object detector built on RT-DETR. • SCFNet fuses CSPDarknet with MambaVision for global context at linear cost. • TSS-FFM preserves spatial details via 3D scale-sequence feature fusion. • Masked Generative Distillation blocks background noise with zero inference cost. • 50.4% and 97.6% mAP on VisDrone2019 and RSOD with only 11.0 M parameters.

Keywords:

UAV aerial imagery

tiny object detection

RT-DETR

MambaVision

feature fusion

knowledge distillation

lightweight model

A drone aerial tiny-object detection algorithm with improved detection accuracy but halved parameter usage

Abstract

Journal

Organization

Cited Papers