Open Access

ARTICLE

CP-YOLO: A Multi-Scale Fusion Method for Electric Vehicle Charging Port Identification

He Tian^1,2, Ziliang Zhu^1,2, Jiangping Li^1,2, Ziyun Li^1,2, Baofeng Tang^1,2, Pengfei Ju^1,2,*

1 Tianjin Key Laboratory for Advanced Mechatronic System Design and Intelligent Control, School of Mechanical Engineering, Tianjin University of Technology, Tianjin, China
2 National Demonstration Center for Experimental Mechanical and Electrical Engineering Education, Tianjin University of Technology, Tianjin, China

* Corresponding Author: Pengfei Ju. Email:

Computers, Materials & Continua 2026, 87(3), 92 https://doi.org/10.32604/cmc.2026.075309

Received 29 October 2025; Accepted 13 February 2026; Issue published 09 April 2026

Abstract

As the number of electric vehicles continues to rise, pressure on charging infrastructure grows increasingly intense. Mobile charging technology, with its flexibility and deployability, has emerged as an effective solution. Within this technology, charging robots or vehicles must autonomously locate and dock with charging ports. Consequently, precise and stable charging port recognition constitutes both a prerequisite and the core bottleneck for achieving automated operations in mobile charging systems. However, in practical scenarios, charging ports often prove difficult to detect reliably due to factors such as physical obstructions, variations in lighting, and long shooting distances. To address this, this paper proposes CP-YOLO, an improved YOLOv9-based detection model for electric vehicle charging ports. The method first incorporates Space-to-Depth Convolution (SPDConv) into the backbone network, reorganising spatial dimensions into channel dimensions to preserve fine-grained features, thereby providing high-fidelity detail for subsequent processing. Secondly, it constructs the C2f-MLKA module, which fuses multi-level features and enhances contextual modelling capabilities through large-kernel attention and gating mechanisms. Finally, introducing Spatially Adaptive Feature Modulation (SAFM) in the neck network enables multi-scale feature fusion and dynamic adjustment of feature weights in critical regions. This forms a progressive mechanism—detailed fidelity, enhanced contextual understanding, and spatial adaptability—to elevate the model’s recognition performance. Experimental results demonstrate that compared to the original YOLOv9 model, Mean Average Precision (mAP) improves by 3.8%, outperforming Faster Region-based Convolutional Neural Networks (R-CNN), EfficientDet and Real-Time Detection Transformer (RT-DETR) by 22.8%, 18.8% and 10%, respectively, and surpasses other YOLO variants by 12.6%, 8.0%, 3.8%, 5.5%, 7.3% and 9.4%, respectively. This effectively enhances detection accuracy for charging ports in complex environments, offering practical reference value for mobile charging technology applications.

---

Keywords

---