Submit

Login Login

Register Register

Home / Journals / CMC / Online First / doi:10.32604/cmc.2026.083432
Journal Menu
Special Issues
Table of Content

All issues

Online First

2026

2025

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

2008

2007

2006

2005

2004

Open Access

ARTICLE

EMW-YOLO: A Detail-Preserving and Multi-Scale Fusion Detector for Remote Sensing Small Object Detection

Heng Wang1, Shichao Li1, Long Xu2,*, Chuqiao Wang1, Yanzhou Feng1, Zou Zhou1,3,4,*
1 School of Information and Communication, Guilin University of Electronic Technology, Guilin, China
2 School of Life Sciences and Medical Engineering, Guangxi Medical University, Nanning, China
3 Ministry of Education Key Laboratory of Cognitive Radio and Information Processing, Guilin, China
4 Guangxi Academy of Artificial Intelligence, Nanning, China
* Corresponding Authors: Long Xu. Email: hahamyt211@outlook.com; Zou Zhou. Email: zhouzou@guet.edu.cn
(This article belongs to the Special Issue: Advanced Object Detection and Visual Understanding in Intelligent Systems)

Computers, Materials & Continua https://doi.org/10.32604/cmc.2026.083432

Received 03 April 2026; Accepted 15 June 2026; Published online 06 August 2026

Abstract

The inherent challenges of small objects in remote sensing imagery encompass the degradation of fine-grained spatial details throughout the downsampling stages, semantic inconsistency during multi-level feature fusion, along with unreliable localization caused by noisy samples. To address these issues, this paper proposes an efficient small-object detector termed EMW-YOLO. An Efficient Down-sampling (EDS) module is introduced to preserve fine-grained spatial information and enhance feature representation during feature extraction through spatial rearrangement and cross-dimensional attention. A Multi-Scale Fusion and Enhancement (MSFE) architecture is further developed to improve semantic consistency across feature levels by combining local enhancement with global feature alignment. In addition, the WIoU v3 loss function is also incorporated to suppress outlier samples and enhance the robustness of regression for small objects. Experimental results on the VisDrone2021 dataset demonstrate that EMW-YOLO achieves absolute improvements of 6.2%, 6.9%, and 7.9% over YOLOv8n in precision, recall, and mAP0.5, respectively. Further evaluation on the NWPU VHR-10 and RSOD datasets confirms that the proposed method exhibits strong generalization capabilities.

Keywords

Small object detection; attention mechanism; multi-scale fusion; remote sensing images; feature representation

  • 79

    View

  • 17

    Download

  • 0

    Like

Related articles
Copyright© 2026 Tech Science Press
© 1997-2026 TSP (Henderson, USA) unless otherwise stated

Share Link