The selection of the initial value of neural network for nonuniformity correction for IRFPA has great effect to constringency ratio.Aiming at this problem,three methods of selection of the initial value are put forward,which are value given directly,black body trained method and two point method.The initial value of neural network is optimized by these methods.The result of simulation shows that these three kinds of method have much more quick constringency ratio and high precision than traditional method,w...