CFCW-YOLO: Collaborative Optimization of Feature Enhancement and Lightweight Design for Pedestrian Object Detection Models
IEEE Access · 2026 · 14 · 118769-118778
DOI: 10.1109/access.2026.3718690 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Min Zhu und 4 weiteren Autor:innen. Die Arbeit nennt 31 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Nantong Municipal Science and Technology Plan Project, National Natural Science Foundation of China.
Abstract (Originalsprache)
This study presents CFCW-YOLO, a lightweight pedestrian detection model designed for mobile and embedded platforms with strict memory and computational constraints. Built on the YOLOv8 framework, the model incorporates three key innovations. First, the cf-CGLU module combines low-redundancy feature extraction from C2f-FasterNet with the CGLU gated activation mechanism, reducing parameters by 41.6% while avoiding gradient vanishing. Second, WaveletPooling replaces conventional convolution subsampling, using Haar wavelet decomposition to preserve high-frequency details at only 18.2 GFLOPs and effectively suppressing aliasing in small targets. Third, SPPF-based multi-scale feature reconstruction contributes to further reducing computational overhead by aggregating features at different scales through parallel pooling operations in a compact manner, forming a lightweight computational graph with 6.48M parameters. Experiments confirm that CFCW-YOLO achieves effective Pareto optimization: under an accuracy loss of less than 0.2% at mAP@0.5, the model reduces parameters from 11.1M to 6.48M and computational cost to 18.2 GFLOPs. These results demonstrate its potential as an accurate yet efficient solution for embedded pedestrian detection in complex environments.
Förderung
- Nantong Municipal Science and Technology Plan Project
- National Natural Science Foundation of China
Zitieren
APA
Zhu, M., Shi, F., He, T., Shen, Z., & Li, J. (2026). CFCW-YOLO: Collaborative Optimization of Feature Enhancement and Lightweight Design for Pedestrian Object Detection Models. IEEE Access, 14, 118769-118778. https://doi.org/10.1109/access.2026.3718690
Vancouver
Zhu M, Shi F, He T, Shen Z, Li J. CFCW-YOLO: Collaborative Optimization of Feature Enhancement and Lightweight Design for Pedestrian Object Detection Models. IEEE Access. 2026;14:118769-118778. doi:10.1109/access.2026.3718690
BibTeX
@article{Zhu2026,
author = {Zhu, Min and Shi, Fengyuan and He, Tingting and Shen, Zhihua and Li, Ji},
title = {CFCW-YOLO: Collaborative Optimization of Feature Enhancement and Lightweight Design for Pedestrian Object Detection Models},
journal = {IEEE Access},
year = {2026},
volume = {14},
pages = {118769--118778},
doi = {10.1109/access.2026.3718690}
}
EUVIMED