To address the issues of high computational redundancy and difficult real‑time deployment of deep learning models in UAV transmission lines inspection

a lightweight detection method based on improved YOLOv11

termed DGCA‑YOLOv11 (dynamic gated and channel attention YOLOv11)

is proposed. First

a DG‑PCSA module combining dual‑gated partial convolution (DG‑PConv) and shuffle attention (SA) is designed for efficient multi‑scale feature fusion in the neck stage. Secondly

this method constructs a dynamic channel attention (DCA) mechanism and embeds it between the last layer of the backbone and the neck‑to‑detection‑head connection

where its adaptive convolution kernel enhances channel responses and suppresses redundant features. Further

DCA is integrated with re‑parameterized convolution (RepConv) to form the RD‑C3K2 module

which strengthens backbone feature extraction while reducing computational costs. In addition

kernel warehouse convolution (KWConv) is introduced in the network’s convolutional layers to replace standard convolution

significantly improving the feature reuse efficiency from the model’s foundation and effectively lowering the computational redundancy. Experimental results on the inspection dataset show that the model reduces parameter count and floating‑point operations by 34.2% and 51.5%

respectively

achieves a mean average precision of 83.0%

and outperforms the baseline YOLOv11 as well as other mainstream models

confirming its effectiveness and practical value for lightweight

high‑precision object detection scenarios.