The special-shaped surface components are widely used in aviation, aerospace, shipbuilding and other fields, and their bonding quality has become an important factor affecting work safety. With its unique physical characteristics, terahertz technology is almost the only means to detect the bonding quality of Polymethaerylimide(PMI) materials. The curvatures of the upper and lower surfaces of the special-shaped surface components are inconsistent, which makes it impossible to accurately judge the bonding quality of the special-shaped surface components when only the upper surface is used as the reference surface for path planning. To tackle this problem, metal sheets of different shapes on the lower surface are fixed to simulate the defects in the bonding layer. Taking the upper and lower surfaces of the special-shaped curved surface components as the benchmark surfaces respectively, the robotic terahertz time-domain spectrum detection system is employed for detection. The terahertz imaging results of the two detection methods are compared and analyzed. The results show that when the lower surface is used as the benchmark surface for detection, the signal to noise ratios at area II and III are improved by 12.09 dB and 10.39 dB than that when the upper surface is the benchmark surface; the ratios of the actual detection area to the theoretical area of the metal sheet at area II and III are improved by 55.97% and 80.81% than that when the upper surface is used as the benchmark surface for detection. This better meets the detection requirements of special-shaped curved surface components. This work will promote the application of terahertz imaging in practice and provide a basis for further research in related fields.