An improved shuffled frog leaping algorithm（ SFLA） is proposed. Beetle antennae search（ BAS）algorithm is used to optimize the search ability of the sub-population of the SFLA. Using the fitness function as the evaluation criteria

by constantly updating the worst individuals to improve the overall level of the subpopulation

so as to optimize the optimal solution of the entire population

that is

the optimal initial weight and threshold of the back propagation（ BP） neural network

and train the BP neural network to obtain the optimal network model. Using the BP neural network

SFLA-BP neural network and modified SFLA-BP neural network to classify the radar active covering jamming signals. The experimental results show that the average correct recognition rate of 0. 91 of modified SFLA-BP neural network is prior to that of BP neural network of 0. 853 5 and SFLA-BP neural network of 0. 891 7.