<p>Fittings are crucial components in overhead transmission lines. However, due to their small size in drone images and the similarity in heights among some fittings, traditional detection algorithms often perform suboptimally. To address these challenges, Faster R-CNN is selected as the baseline, and a fittings detection network using enhanced feature fusion and attention mechanisms (FDNet) is proposed. First, a three-path feature fusion network is integrated into Faster R-CNN, enabling the model to better combine multi-scale features and enhance the representation of small targets. Second, to improve the network's sensitivity to subtle features, a decomposed Manhattan self-attention module is introduced after the feature fusion network. Additionally, a Bounded IoU Loss (B_IoU) is developed to effectively mitigate the challenges posed by dense fitting distributions. Experimental results on a dataset containing six types of fittings demonstrate that FDNet achieves a 4.1% improvement in detection accuracy compared to the baseline Faster R-CNN model.</p>