Mined-out areas, caves, voids and cavities usually appear as diffracted waves on ground penetrating radar (GPR) profiles. Therefore, the complete extraction of diffracted waves forms the foundation of the efficient usage of the GPR technique in geological surveys. We propose a method of enhancing GPR diffracted waves via singular value decomposition (SVD) filtering and establish an effective GPR data processing flowchart. First, the shallow and deep signal energies were controlled within a certain dynamic range by energy scaling in the traces. Next, the SVD filtering process was employed to suppress air waves and multiples with better transverse coherence and to extract GPR diffracted waves. Third, background noise was suppressed via band-pass filtering to further improve the signal-to-noise ratio (SNR) of the GPR data. Finally, fitting a diffraction time-distance hyperbola allow us to obtain a diffraction velocity. Constant velocity migration processing for the diffracted waves was based on the Kirchhoff migration technique. The feasibility and effectiveness of this GPR processing technique were verified with the discovery of geological flaws beneath the Mengshan Giant Buddha in China during a cavity survey. Our proposed flowchart efficiently extracts GPR diffracted waves and increases the data SNR. The resulting images are more readily interpreted within the local geological context.