Open Access

ARTICLE

DSGF-Net: A Dense-SE Gated-Fusion Architecture for High-Accuracy Small Object Detection in UAV Imagery

Changzhu Shi, Hongmei Liu^*

School of Mathematical Sciences, Dalian Minzu University, Dalian, China

* Corresponding Author: Hongmei Liu. Email:

Computers, Materials & Continua 2026, 88(2), 26 https://doi.org/10.32604/cmc.2026.074281

Received 07 October 2025; Accepted 22 December 2025; Issue published 15 June 2026

Abstract

To address the critical challenges of small object detection in UAV imagery, this paper proposes DSGF-Net (Dense-SE Gated-Fusion Network), an enhanced architecture built upon YOLOv10. It integrates a Dense SE Network (DSENet) backbone, an Adaptive Gated Fusion (AGF) module, and a Channel-Spatial Attention (CSA) mechanism. Extensive experiments on VisDrone2019-DET and CODrone demonstrate that DSGF-Net achieves substantial mAP@0.5 improvements of 5.12% and 2.36% over the YOLOv10n baseline.

---

Keywords

---