Terahertz frequency-modulated continuous-wave (FMCW) imaging inherently suffers from low spatial resolution, with edge and texture information often being obscured by diffraction noise. This paper proposes to employ the discrete wavelet transform (DWT) to decompose the original low-resolution image into multi-level subbands containing high-frequency detail features and low-frequency global structural features. Subsequently, a three-branch parallel architecture is used to extract shallow features, detail features, and global features, respectively. These features are then concatenated in the same spatial domain and adaptively fused through convolution to generate a high-resolution reconstructed image.