Because the infrared focal plane detector is limited by manufacturing technology, the image is inevitably nonuniform. The traditional neural network algorithm solves the "ghost" problem using the guided filtering image as the expected template to prevent image edge smoothing by the filter. When the scene is moving, the nonuniformity correction parameters are continuously updated using the time-domain iteration strategy. To suppress the common ghosting phenomenon in the algorithm, an adaptive learning rate was designed based on a combination of the spatial local variance and the time-domain scene change rate, and the threshold was adjusted adaptively using the correction parameters before and after. Simulation results show that the root mean square error of the proposed algorithm is reduced by 45.45% compared with that of the traditional algorithm, and the proposed algorithm can suppress the "ghost" phenomenon well while correcting image nonuniformity.