Aiming at the challenges of complex background interference, random distribution and large differences in transmission line scales in UAV transmission line detection tasks, this paper proposes an algorithm framework based on multi-scale feature enhancement. The framework is based on multi-modal feature fusion input and integrates three core modules: the multi-scale feature enhancement module (multi-scale feature enhance, MSFE) is used to capture transmission line features of different scales; the dual-path interactive attention module (dual-path interactive attention, DPIA) focuses on key transmission line regions through a dual-path mechanism of channel and space; the adaptive feature fusion module (adaptive feature fusion, AFF) dynamically balances the semantic information of the encoder and the edge information of the decoder, aiming to improve the detection robustness in complex scenarios. Experimental results show that the proposed method performs excellently in various complex scenarios such as foggy days, snowy days, and low light. Compared with existing methods, all indicators have achieved significant improvements. Ablation experiments fully verify the effectiveness of each module: on the basis of the baseline model, IoU, Robj, Recall, Precision, and F1-Score are increased by 12.56%, 5.53%, 4.54%, 3.03%, and 3.78% respectively, indicating that they are crucial for enhancing the detection capability of the model. Qualitative analysis results further confirm that the algorithm can accurately locate the slender structure of transmission lines in complex scenarios and effectively suppress background noise. Therefore, the method in this paper has practical application value for UAV transmission line detection tasks.