Refined Object Detection: Integrating C2f and SE Mechanisms in YOLOv5

Huanchang Tu

2024

Abstract

The precise detection of small objects against complex backgrounds is crucial for advancing computer vision technologies, with wide-ranging applications from autonomous navigation to surveillance. This paper presents a novel integration of the modified Cross-Stage Partial bottleneck structure (C2f) and the Squeeze-and-Excitation (SE) attention layer within the You Only Look Once version 5 (YOLOv5) framework. The primary objective is to enhance the model's sensitivity to subtle object features, thus improving detection accuracy in challenging environments. By leveraging the C2f module's effective feature integration and the SE layer's focus on essential feature recalibration, the model achieves a balanced representation of depth and detail in features. Experimental results on the COCO128 dataset reveal a notable improvement in detection accuracy, surpassing existing methods. This study underscores the efficacy of targeted neural network modifications in addressing specific detection challenges, providing valuable insights for the development of more adaptable detection systems. The success of this approach highlights the potential for sophisticated architectural enhancements to enhance the versatility and effectiveness of computer vision models across diverse real-world scenarios.

Paper Citation

in Harvard Style

Tu H. (2024). Refined Object Detection: Integrating C2f and SE Mechanisms in YOLOv5. In Proceedings of the 1st International Conference on Engineering Management, Information Technology and Intelligence - Volume 1: EMITI; ISBN 978-989-758-713-9, SciTePress, pages 730-735. DOI: 10.5220/0012969900004508

in Bibtex Style

@conference{emiti24,
author={Huanchang Tu},
title={Refined Object Detection: Integrating C2f and SE Mechanisms in YOLOv5},
booktitle={Proceedings of the 1st International Conference on Engineering Management, Information Technology and Intelligence - Volume 1: EMITI},
year={2024},
pages={730-735},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0012969900004508},
isbn={978-989-758-713-9},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 1st International Conference on Engineering Management, Information Technology and Intelligence - Volume 1: EMITI
TI - Refined Object Detection: Integrating C2f and SE Mechanisms in YOLOv5
SN - 978-989-758-713-9
AU - Tu H.
PY - 2024
SP - 730
EP - 735
DO - 10.5220/0012969900004508
PB - SciTePress