Skip to main navigation Skip to search Skip to main content

Research on Improvement Methods for YOLOv5s Based on Gaussian Decay

  • Zhiqi Liu
  • , Xiaopu Zhang
  • , Yuanrui Liang
  • , Yong Xia
  • China Aviation Industry Corporation
  • Northwestern Polytechnical University Xian

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

This study proposes an enhanced YOLOv5s algorithm that improves detection accuracy by integrating Gaussian decay into the non-maximum suppression (NMS) process. This modification optimizes the conventional greedy suppression strategy through a soft attenuation mechanism based on overlap degree. Comprehensive experiments on the TT100K and MS COCO datasets demonstrate that the Gaussian-weighted NMS achieves state-of-the-art performance, with a 95.48 % mAP@ 0.75 on TT100K and 39.7% mAP{@}0.5:0.95 on MS COCO, while maintaining real-time inference at 139 FPS. The method exhibits strong generalization capability and provides an optimal balance between precision and computational efficiency.

Original languageEnglish
Title of host publication2025 5th International Conference on Electronic Information Engineering and Computer Technology, EIECT 2025
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages109-113
Number of pages5
ISBN (Electronic)9798331575243
DOIs
StatePublished - 2025
Event2025 5th International Conference on Electronic Information Engineering and Computer Technology, EIECT 2025 - Jiangmen, China
Duration: 24 Oct 202526 Oct 2025

Publication series

Name2025 5th International Conference on Electronic Information Engineering and Computer Technology, EIECT 2025

Conference

Conference2025 5th International Conference on Electronic Information Engineering and Computer Technology, EIECT 2025
Country/TerritoryChina
CityJiangmen
Period24/10/2526/10/25

Keywords

  • Computational efficiency
  • Gaussian decay
  • Non-Maximum Suppression
  • Real-time detection
  • YOLOv5s

Fingerprint

Dive into the research topics of 'Research on Improvement Methods for YOLOv5s Based on Gaussian Decay'. Together they form a unique fingerprint.

Cite this