In order to address the challenges of difficult recognition and high miss-detection rates for dense screws and cables in electrical cabinet inspections, DPF-YOLO a lightweight detection algorithm based on an optimized YOLOv8 framework, was used to investigate the effectiveness of specialized feature fusion and dynamic sampling mechanisms. This approach integrates a Lightweight Cross-scale Feature Fusion (LCFF) module to enhance deep semantic and spatial representation, an improved SPPFL module for global multi-scale modeling, and a combination of DySample and C2f_DRM to focus on small and elongated targets. The results show that DPF-YOLO outperforms state-of-the-art methods, achieving an overall mAP@50 of 0.9566 and a recall of 0.937. Specifically, the model demonstrates high reliability in complex environments, with screw detection reaching a recall of 0.953 and cable detection maintaining a precision of 0.924. It is concluded that this methodology provides a robust vision solution for intelligent maintenance robots, significantly enhancing the efficiency and accuracy of automated electrical cabinet inspections.