Abstract: In view of the problems of over-fitting and low prediction accuracy of chlorophyll inversion by the PRO-4SAIL radiation transfer model coupled with BP neural network, there are some problems.In this paper, measured hyperspectral data and simulated spectral data in the research area are used as data sources, and some measured sample data are added to the training set composed of simulated sample data to build the BP neural network chlorophyll inversion model, and the model verification and accuracy evaluation are carried out with additional measured data.The results show that the training concentration can solve the over-fitting problem of chlorophyll inversion model by adding a small amount of measured data, improve the accuracy of chlorophyll content prediction, and accurately invert the vegetation information of roadways, which can be applied to the remote sensing monitoring and evaluation of roadways environment.

Key words: PRO-4SAIL, back propagation neural networks, overfitting, chlorophyll, expressway vegetation