Small object detection plays a critical role in applications such as security surveillance, autonomous driving, and remote sensing. However, conventional detection methods often struggle with high annotation costs, low resolution, and heavy computational requirements. To address these challenges, we propose CGF-YOLOv11n, which is the abbreviation of C2 block + parallel spatial attention module (C2PLUS)-Ghost Convolution (GhostConv)-Feature Diffusion Pyramid Network (FDPN)-You Only Look Once (YOLO)v11n, an efficient and real-time small object detection algorithm built upon the YOLOv11n framework. First, we introduce the C2PLUS module, which effectively enhances fine-grained feature extraction for small targets. Second, we design a plug-and-play Ghost-Residual Field-Aware Convolution module to strengthen the feature extraction capability of the backbone network. Finally, the FDPN module is incorporated to promote the balanced fusion between semantic features and spatial information. Experimental results on the VisDrone2019 dataset demonstrate that the proposed method achieves improvements of 3.5 and 3.1% in mAP@0.5 on the validation and test sets, respectively, outperforming the baseline YOLOv11n model. In addition, CGF-YOLOv11n achieves 34 frames per second on the Orange Pi 5 platform, confirming its suitability for real-time deployment and advancing the performance of small object detection systems. The related implementation details, including code and datasets, are available through the authors’ public project repository. In this study, we primarily contribute an efficient modular enhancement strategy for real-time small object detection by integrating C2PLUS, Ghost-based convolution, and FDPN into a lightweight YOLOv11n framework. While the proposed CGF-YOLOv11n demonstrates notable accuracy gains and real-time performance on an embedded platform, the current evaluation is limited to a single aerial benchmark dataset and does not fully explore robustness under extremely dense scenes or severe resolution degradation. Future work will focus on extending validation to more diverse datasets, improving generalization in complex real-world environments, and further optimizing the model for ultralow-power edge devices.