An image compression model based on the FA-LMBP hybrid neural network algorithm is proposed by the combination of the firefly algorithm and LM (Levenberg-Marquardt) algorithm. First, a set of optimal approximate solutions of the backpropagation(BP)network weight threshold are obtained by the global search of the target function based on the firefly algorithm, then the approximate solution is used as the initial value of the BP model, and the LM algorithm is used to carry out the secondary optimization training for these weight thresholds, thereby obtaining the final image compression model. The experimental results show that the reconstruction quality of the compression image model based on the FA-LMBP hybrid neural network algorithm is obviously higher than that of the basic BP model and the LMBP model under the same training times and error accuracy.