Abstract: In practical detection operations using ground-penetrating radar (GPR), factors such as environmental noise and instrument errors frequently cause signals to be mixed with substantial noise, seriously reducing signal quality and the reliability of analytical results. To address this issue, this study proposed a time-frequency peak filtering method combined with minimum mean cross-entropy (TFPF-MMCE) for denoising GPR signals. This method combined time-frequency peak filtering with the cross-entropy function, enabling effective noise suppression and precise preservation of valid signals through precise optimization of the time-frequency representation, thereby significantly improving the quality of GPR signals. Numerical simulation and field GPR experiments validated that the TFPF-MMCE method exhibited a high noise removal capability and, thus, can effectively eliminate random noise while significantly improving signal clarity and reliability. Compared to traditional denoising methods, TFPF-MMCE shows significant advantages in denoising effectiveness and noise resistance stability, suggesting promising application potential and practical value in the field of GPR signal processing.