<p>With the continuous development and intelligent transformation of power systems, the safety monitoring and maintenance of high-voltage line equipment has become a key aspect in ensuring the stable operation of the power grid. Currently, the identification of high-voltage line equipment relies on image acquisition from drones; however, image blurriness poses challenges for small target recognition, leading to false detections, missed detection, inaccurate localization, and slow detection speeds. To address this issue, this paper proposes a lightweight feature extraction model, LFE-YOLO, based on YOLOv8-N. The model uses C2BCE modules to replace some C2f modules in the backbone network to enhance feature capture capabilities and introduces ADown modules to process feature maps, reducing the number of parameters to improve operational efficiency and avoid false detections or missed detections. In the neck network, ADown and MSC modules further enhance the model’s feature capture ability, while the introduction of the narrow-neck structure VOVGLFC module reduces computational burden and parameter count, thereby improving detection accuracy for small targets while meeting lightweight requirements. In the head network, to address the issue of excessive parameters in the YOLOv8-N detection head that cannot handle multi-scale targets, the LPIH module is designed to ensure model accuracy while reducing computational load and parameter count. Experiments on public and private datasets have shown that compared to the baseline model, the mAP50 of the LFE-YOLO model has increased by 4.1% and 0.9%, respectively, while the number of parameters has decreased by 43.33% and 28.57%, fully verifying the feasibility and effectiveness of the model. Importantly, this article tested the inference speed using the JOINT framework on the MAIX-III AXera Pi, reaching 26 frames/s, demonstrating efficient detection output.</p>