An Efficient UAV Object Detection Model Using Multi-Scale Feature Fusion and Attention

View at publisher Summarise with VIRAT
Authors: Feng Wang, Dong Zhao, Jinquan Hou, Shanshan Wu, Jinwei Zhang
Cited by: 0
References: 21

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Considering the small size of targets and their weak visual cues, we develop MSFFA-YOLO, a detection framework built upon the YOLOv5 architecture for UAV-based object detection. During feature extraction, the Weight-Selective Channel Attention (WSCA) mechanism is introduced. This module assigns attention weights to multi-scale feature information across channels and spatial dimensions, thereby filtering out crucial feature information of targets in UAV aerial images. During feature fusion, a Multi-Scale Feature Pooling Network (MSFPN) is proposed. This network enhances information transfer between features of different scales by adding branching paths for multi-scale feature fusion. For classification and regression, a lightweight FB-decoupled head is introduced to decouple the two prediction branches, allowing the model to achieve more accurate object classification and localization. Experiments on the public VisDrone2019 dataset show that the proposed model improves mAP @ 0.5 by 14 % over YOLOv5s. Furthermore, it outperforms several representative state-of-the-art detectors, including YOLOX, YOLOv6, YOLOv7, YOLOv8s, YOLOv9-C, YOLOv10, YOLO11s, YOLOv12s, and RT-DETR-L.

Fields of Study & Subjects

Advanced Neural Network Applications Visual Attention and Saliency Detection Domain Adaptation and Few-Shot Learning

Publication details

Journal / source
International Conference on Electrical Automation and Artificial Intelligence (ICEAAI)
Pages
738–744
Publication date
16 January 2026
Language
EN