Terahertz time-domain spectroscopy (THz-TDS) imaging technology has demonstrated significant potential in the field of nondestructive testing. However, the reconstruction quality of terahertz images is often limited by the diffraction effects of terahertz beams. This study introduced a novel image reconstruction method for THz-TDS that combined deconvolution and compressed sensing techniques. Initially, THz-TDS equipment was utilized to collect time-domain spectroscopy data of the sample. Subsequently, fast Fourier transform (FFT) and spectral preprocessing were performed to determine the imaging frequency, 0.80 THz. Following this, point spread function (PSF) were constructed for various penetration depths, with the optimal reconstruction performance achieved at a penetration depth of 2.0 mm, effectively mitigating the adverse impact of terahertz beam diffraction on imaging quality. Four compressed sensing algorithms, namely orthogonal matching pursuit (OMP), stagewise orthogonal matching pursuit (StOMP), iterative hard thresholding (IHT), and total variation augmented Lagrangian alternating direction (TVAL3), were employed to reconstruct deconvoluted THz images at sampling rates of 70%, 55%, 40%, 25%, and 10%. Comparative analysis revealed that the TVAL3 reconstruction algorithm outperformed the other three algorithms. The integration of deconvolution with the TVAL3 compressive sensing algorithm, compared to TVAL3 reconstruction of non-deconvolved terahertz images, effectively enhanced the reconstruction details and resolution of the terahertz images, achieving efficient and high-fidelity reconstruction under undersampled data conditions. This study is expected to promote the development of terahertz high-resolution rapid reconstruction.