Multi Scale Electrical Component Identification and Localization Method Based on YOLO11 and FPN Improvement

Abstract

Accurate identification and localization of electrical components are essential for unmanned inspection, real-time fault warning, and safety control in intelligent power systems. In complex power scenarios, component detection faces large target-scale differences, high missed detection rates for small targets, dense overlap interference, and strong background noise. To address these challenges, this paper proposes a high-precision multi-scale electrical component recognition and localization method based on YOLO11 and an improved bidirectional feature pyramid network. The method introduces an enhanced backbone with adaptive convolution and lightweight channel attention, designs a BiFPN+ feature fusion structure to strengthen multi-scale information transmission, and optimizes bounding box regression and post-processing to improve dense target discrimination. The framework is suitable for identifying components such as circuit breakers, isolating switches, insulators, transformers, fuses, wiring terminals, indicators, and clamps in real inspection images. The study provides a practical visual perception method for power infrastructure monitoring and is closely related to electromagnetic engineering applications, including antenna-based inspection platforms, electromagnetic-wave sensing environments, and smart-grid equipment supervision.