Inspections of power transmission systems are essential as they directly impact the safety and stability of the entire power transmission infrastructure. With the advancement of technology, the utilization of unmanned aerial vehicles (UAVs) for power line inspections has progressively gained widespread acceptance. Accurately assessing the operational status of critical components such as insulators holds paramount importance in ensuring the security of power transmission systems. However, employing UAVs for power line inspections results in the generation of large volumes of image data, while manual inspection methods prove inefficient and error-prone. To address these challenges, this study proposes an advanced target recognition algorithm. This algorithm effectively tackles the considerable parameter and computational costs associated with the CBS downsampling network by adopting the ADown downsampling network, simultaneously enhancing the propagation of gradient information. Furthermore, an innovative redesign of the generalized efficient layer aggregation network (GELAN) incorporates the integration of the Dilated Reparam Block (DRB) network, ultimately yielding the DRNCSPELAN4 module. This module amplifies the receptive field while reducing the burden of parameter and computational costs. The concatenated design of these two structures facilitates efficient extraction of feature information and optimization of gradient utilization, thereby achieving superior target detection outcomes. Experimental findings reveal not only an improvement of 1.6% and 1.0% in respective mAP50 and mAP50–95 values when compared to the original YOLOv8 model, but also varying degrees of reductions in model parameters and computational costs.