Aerial images captured by drones at various altitudes often exhibit diverse target sizes and complex background noise, which present substantial challenges for small target detection. Traditional methods primarily focus on extracting global features while neglecting finer details and background noise, often resulting in suboptimal detection performance. To address these challenges, we propose a novel detection network named WMS-YOLO. Specifically, we propose a Wavelet-Enhanced Convolution (WEC), and apply it to the backbone network. By combining a discrete wavelet transform with parallel \(1 \times 1\) convolutions, assigning different weight values. WEC effectively captures both local details and global context, while also denoising and suppressing redundant features. Next, following the C2PSA module, we propose a Multi-Scale Fusion Attention (MSFA) module, which integrates channel attention, multi-scale convolutions, and spatial attention. This module adaptively captures rich features of targets across multiple scales, thereby enhancing the model’s sensitivity and capability in detecting multiscale objects. Finally, during the feature fusion stage, we propose a high-resolution Small Target Detection Layer that accurately reconstructs target details and applies the wavelet transform to eliminate background noise, thereby reducing the overall complexity of the model. Experimental results on the publicly available VisDrone2019 and DIOR datasets demonstrate that our method outperforms existing state-of-the-art approaches in both detection accuracy and robustness.