This study proposes the enhanced line-detection adaptive you only look once (ELA-YOLO), an enhanced YOLOv8-based object detection algorithm, to improve the identification and classification of critical power components. By integrating efficient multi-scale attention (EMA) into redesigned cross stage partial feature fusion (C2f) modules (C2f_EMA), the backbone network achieves dynamic multi-scale feature fusion. The neck network is further optimized through asymmetric padding convolution (APConv) in C2f_AP modules, enhancing spatial feature integration. Additionally, the large selective kernel (LSK) attention mechanism strengthens context-aware feature extraction capabilities. Experimental results demonstrate that ELA-YOLO outperforms YOLOv8s with a 2.8% improvement in mean average precision at intersection over union threshold 0.50 (mAP₅₀) while incurring only a 4.5% computational overhead, establishing an optimal balance between detection accuracy and operational efficiency for real-world power equipment inspection scenarios.