Insulators play a vital role in transmission lines, and their defect detection using drone inspection technology can be adversely affected by complex environments, including influence of complex lighting, complex background and complex defects. To address the issue of complex lighting first, this paper innovatively designs an adversarial network based on the decomposition and fusion of structure and brightness layers. This network is further enhanced with a brightness balance loss strategy to improve imaging of insulators and their defects in overexposed and underexposed scenarios, resulting in an average PSNR improvement of 0.4 and an SSIM increase of 0.29.Next, to tackle the problem of complex backgrounds, an axial attention feature extraction backbone is designed, considering the distribution characteristics of insulators and their defects, to extract targets from complex backgrounds. Finally, for the multi-scale defect target in complex defects, we propose the Res-PANet, a feature fusion structure based on multi-scale residual connections, which enhances the network’s detection accuracy in scenarios with multiple targets. Experimental results demonstrate that the detection model, after preprocessing for complex lighting conditions, achieves higher detection accuracy, with an average precision of 89.93%. This underscores the model’s strong practical value for engineering applications.