In order to realize the rapid and accurate detection of chlorophyll content in plants, an inversion method based on MMD migration was proposed. Taking Epipremnum aureum leaves as the research object, the motion trajectory of photons was simulated with the Monte Carlo method based single-layer flat plate model, totally 12000 simulated luminance distribution maps were obtained. The convolutional neural network was used to train the simulated spectral data to obtain the pre-training model. Then based on the pretraining model, the model was fine-tuning on the measured spectral data of a small amount of Epipremnum aureum leaves to realize the inversion of the optical parameters. The inversion results were as follows: absorption coefficient μa was 84.83% and scattering coefficient μs was 83.33%. On this basis, the maximum mean difference method was added to improve the migration effect. The results showed that the MMD migration method had a better inversion effect with absorption coefficient μa was 87.55% and scattering coefficient μs was 86.67% compared with the common model migration method. The chlorophyll regression model was established by using the full connection layer characteristics obtained from MMD migration, and the determination coefficient R2 of this method was 0.0468 and 0.0620 higher than that of the model established directly using optical parameters and spectral images, respectively. The experimental results showed that the inversion method based on optical characteristic parameters can provide important reference for the research of chlorophyll nondestructive detection.