The dynamic modulus, an important viscoelastic parameter of asphalt pavements, plays an essential role in the performance evaluation of pavement structure throughout the whole life cycle from design to in-service. The bedrock beneath the pavement structure will significantly influence the results of the deflection response in falling weight deflectometer testing and dynamic modulus back-calculation. This effect has often been neglected in traditional analysis or back-analysis. Here, an intelligent back-calculation method for the dynamic modulus of asphalt pavement on bedrock is proposed. The dynamic modulus intelligent back-calculation model is established by a convolutional neural network rather than general numerical methods. The datasets containing the deflection time history curves were from the spectral element method with fixed-end boundary conditions of viscoelastic surface asphalt pavements on bedrock. The accuracy of the intelligent back-calculation model, the sensitivity of the parameters in the modified Havriliak–Negami (MHN) model to the dynamic modulus master curve, and the deviation degree from the back-calculated master curves of the dynamic modulus and phase angle with larger and smaller errors to the theoretical curves are evaluated. The results demonstrated the model’s good back-calculation effect without any overfitting. Parameter changes in the MHN model caused the dynamic modulus master curves to shift up and down, rotate, or shift left and right in each frequency band with a positive or negative correlation. Moreover, the back-calculated master curves of the dynamic modulus and phase angle exhibited good agreement with the theoretical curves. The intelligent back-calculation approach exhibited validity, reliability, and broad applicability.