Abstract: Unmanned aerial inspection of power transmission lines imposes a triple constraint on insulator defect detection: high accuracy, low latency, and low power consumption. However, existing lightweight convolutional neural networks (CNNs) still suffer from missed and false detections, particularly under complex backgrounds and when detecting small-scale defects. Multi-scale wavelet decomposition, with its excellent time-frequency localization capability, enhances edge and texture information across different resolutions, providing fine-grained frequency-domain priors that are highly beneficial for small object detection. In this paper, we propose an end-to-end real-time detection framework that integrates multi-scale wavelet decomposition with a lightweight CNN. Specifically, a two-level discrete wavelet transform is first applied to the input image to explicitly inject low-frequency structural components and high-frequency details into the shallow feature maps. A wave-domain and spatial-domain channel alignment module is then designed, combining depthwise separable convolutions and attention mechanisms to enable efficient multi-domain feature fusion. Furthermore, a receptive field expansion module and an end-to-end training strategy are introduced to enhance robustness against cluttered backgrounds and multi-scale defect patterns. Experimental results on public aerial datasets and self-collected UAV imagery demonstrate that the proposed method achieves significant improvements in accuracy and recall compared to mainstream lightweight detectors, while maintaining real-time inference speed and an extremely compact model size. The approach also maintains stable performance under varying weather and lighting conditions.