Abstract: Given that the pavement roughness is difficult to describe during the 3D pavement reconstruction,a method to reverse the pavement using the measured load spectrum is proposed.On the basis of the establishment of the acquisition method and test system,the actual driving load spectrum of a vehicle is collected,and the 3D pavement reconstructed by taking the straight-line constant speed driving conditions under three typical roads as an example.Based on the BP neural network,the pavement reverse model is established,and the BP neural network is trained and verified by using the output response of the vehicle seven-degree-of-freedom road input vibration model under the standard road input.With the measured and integrated load spectrum as the input,the 2D pavement spectrum is obtained and verified.On this basis,the 3D pavement is reconstructed using the triangular mesh method.The results show that when the inverse pavement is used as an excitation input to the seven-degree-of-freedom pavement input vibration model,the relative error between the output spring acceleration and the measured acceleration is less than 8%,which verifies the validity of the pavement reverse results.It provides a feasible method for 3D pavement reconstruction.