In the processing of vertical seismic profile (VSP) data, coupled noise is the main interfering signal. Its complex propagation characteristics and frequency overlap make it difficult for traditional denoising methods to achieve precise suppression. Therefore, a hybrid wavelet-Transformer (HWT) framework was proposed for efficient suppression of coupled noise. Firstly, the multi-scale frequency decomposition of the signal was performed by discrete wavelet transform (DWT), and the original signal was decomposed into different frequency bands to reduce frequency aliasing. Then, the two-branch Transformer architecture was used to capture local and global features respectively, where the local branch extracted the high-frequency interference characteristics of the coupled noise, and the global branch models the time-frequency global dependence of the target signal. In addition, a multistage feature aggregation module (MFAM) was designed. By integrating the frequency domain features decomposed by DWT with the local and global features of the two-branch output, the features were deeply integrated. Based on the mask generation strategy in the time-frequency domain, the time-frequency representation was constrained, which further improved the separation quality. Experiments on synthetic data and real VSP data show that the proposed method can suppress coupling noise significantly while maintaining the integrity of target signal, which provides an efficient and robust solution for the pre-processing of VSP data.