A Two‐Stage SlowFast‐YE Network for Robust Recognition of Crew Irregularities on the Bridge

Read full text Summarise with VIRAT
Authors: Deshan Chen, Long Jiang, Liang Fan, Di Zhang
Cited by: 0
References: 35

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

ABSTRACT Crew irregularities during duty are critical human factors that significantly increase the risk of maritime accidents, particularly in complex ship bridge environments. Vision‐based behaviour recognition offers a practical and non‐contact solution for real‐time crew monitoring; however, accurate recognition remains challenging due to small‐scale targets, occlusions, illumination variations, and limited image resolution. To address these issues, this paper proposes a two‐stage real‐time crew irregularities recognition framework, termed SlowFast‐YE. In the first stage, a lightweight crew detection network, YOLOX‐SP, is designed by incorporating the Softpool mechanism to enhance feature preservation for small and occluded targets. In the second stage, an efficient pyramid squeeze attention (EPSA) module is integrated into the SlowFast network to improve multi‐scale feature extraction and discrimination among similar irregular behaviours. A self‐constructed dataset of crew irregularities on the bridge (DCIB) containing over 250,000 video frames across six typical crew irregularity categories is established, covering diverse conditions such as illumination changes, occlusion, blur, and both simulated and real ship bridge environments. Experimental results demonstrate that SlowFast‐YE achieves a Top‐1 accuracy of 87.4%, outperforming the original SlowFast network by 2.2%, while maintaining real‐time performance at 14.1 ms per frame. These results verify the effectiveness and practical applicability of the proposed framework for maritime safety monitoring.

Fields of Study & Subjects

Human Pose and Action Recognition Advanced Neural Network Applications Video Surveillance and Tracking Methods

Publication details

Journal / source
IET Intelligent Transport Systems
Volume
20
Issue
1
Publication date
1 January 2026
Language
EN