Article: DAFPN: a lightweight multi-objective framework for small object detection in remote sensing images Journal: International Journal of Computer Applications in Technology (IJCAT) 2026 Vol.78 No.2 pp.144 - 154 Abstract: Remote sensing object detection faces persistent challenges in accurately identifying small-scale targets embedded in high-resolution, cluttered scenes. Conventional detectors often suffer from feature dilution, scale variance and high computational cost, limiting their applicability in real-time or edge-based remote sensing scenarios. To address these issues, we propose DAFPN, a lightweight Dynamic Attention-guided Feature Pyramid Network that integrates asymmetric multi-scale fusion and dual-branch attention, consisting of spatial and channel-wise attentions, into a unified architecture optimised via multi-objective constrained learning aimed at simultaneously maximising detection accuracy, attention alignment and architectural compactness. On DOTA-v2.0, our method improves mAP@0.75 by 4.5% and mAP@0.5 by 3.8% over YOLOv8, while achieving similar gains on FAIR1M, DIOR and RSSOD. The results confirm DAFPN's robustness under variable input resolutions and dense object distributions, highlighting its practical value for deployment in real-time and resource-constrained remote sensing applications. Inderscience Publishers - linking academia, business and industry through research

Title: DAFPN: a lightweight multi-objective framework for small object detection in remote sensing images

Authors: Wanqi Ren

Addresses: College of Economics, Shandong University of Finance and Economics, Jinan, Shandong, China

Abstract: Remote sensing object detection faces persistent challenges in accurately identifying small-scale targets embedded in high-resolution, cluttered scenes. Conventional detectors often suffer from feature dilution, scale variance and high computational cost, limiting their applicability in real-time or edge-based remote sensing scenarios. To address these issues, we propose DAFPN, a lightweight Dynamic Attention-guided Feature Pyramid Network that integrates asymmetric multi-scale fusion and dual-branch attention, consisting of spatial and channel-wise attentions, into a unified architecture optimised via multi-objective constrained learning aimed at simultaneously maximising detection accuracy, attention alignment and architectural compactness. On DOTA-v2.0, our method improves mAP@0.75 by 4.5% and mAP@0.5 by 3.8% over YOLOv8, while achieving similar gains on FAIR1M, DIOR and RSSOD. The results confirm DAFPN's robustness under variable input resolutions and dense object distributions, highlighting its practical value for deployment in real-time and resource-constrained remote sensing applications.

Keywords: remote sensing; small object detection; multi-objective optimisation; feature pyramid networks; dynamic attention; lightweight detection; aerial imagery; real-time inference.

DOI: 10.1504/IJCAT.2026.151721

International Journal of Computer Applications in Technology, 2026 Vol.78 No.2, pp.144 - 154

Received: 11 Jun 2025
Accepted: 20 Oct 2025
Published online: 17 Feb 2026 *

Title: DAFPN: a lightweight multi-objective framework for small object detection in remote sensing images

Keep up-to-date