The invention provides a ballistic target identification method and
system based on a long short-
term memory neural network, and the method comprises the steps: obtaining to-be-classified
radar sample data, recording the echo data of a shell attribute to-be-classified target, carrying out the amplitude normalization, carrying out the RCS calibration of the current echo data through the RCS value under a
reference distance and a pulse width, obtaining the preprocessed data, and carrying out the classification of the target. Dividing into a
training set and a
test set, and training the LSTM network by using the
training set; and carrying out pre-
processing on to-be-classified
radar sample data which are newly obtained subsequently, inputting the pre-processed data which are obtained subsequently as
test data into the trained LSTM network for prediction, and obtaining a
projectile attribute prediction value. According to the method, the
algorithm implementation complexity is reduced, the method is suitable for distinguishing the
rocket projectile and the grenade of the emplacement investigation
radar, the robustness is high, the recognition is accurate, our radar is assisted to master the military condition of the enemy in time, convenience is provided for confirming the firepower configuration of the enemy, and the correct recognition rate of the
rocket projectile and the grenade reaches 90% or above.