The existing methods for inverting the optical characteristic parameter of fruit tissue are time-consuming and labor-intensive, and poor in generality. Aiming at these problems, an optical characteristic parameter inversion method based on model migration was proposed. Taking apple as an example, a simulated double-layer tissue model based on the Monte Carlo method was used to generate 1.5 million light distribution maps. The light distribution map was input as a data set to the 8-layer convolutional neural net (CNN)) for training, to obtain a pre-trained model. The trained model finally was transferred to the actual measured data set containing 4000 apple hyperspectral point light source images, and fine-tuned to complete the inversion of optical parameters.The method was compared with the inversion results of several other algorithms. The results showed that when the measured data set was small, the inversion results of this method on apple optical characteristic were the peel absorption coefficient μa1 was 87.26%, the pulp absorption coefficient μa2 was 90.53%, the peel scattering coefficient μs1 was 86.66%, and the pulp scattering coefficient μs2 was 87.57%. The accuracy of the inversion was higher than other inversion methods. The pre-trained model was obtained by training a large number of light distribution maps based on the simulation model. The model was universal, and it can provide a solution to the problem of insufficient data amount in the inversion of fruit optical characteristic parameters.