Aiming at the samples selected by the traditional acoustic emission location based on neural network is not widely representative, acoustic emission localization technologies based on fiber Bragg grating (FBG) network and BP neural network are proposed. Time difference between the extracted narrowband signals is used as training sample. According to the back propagation (BP) neural network, acoustic emission sources are located without a large number of samples of difference characteristics acoustic emission and wave velocity. To solve the problem of slow convergent rate in gradient-descent algorithm, Levenberg-Marquardt algorithm is applied to improve convergent rate. The algorithm is verified in 500 mm×500 mm aluminum alloy plate. The average localization accuracy of aluminum alloy plate is 6.65 mm, which verifies the validity and accuracy of the method.