Underwater acoustic target recognition under working conditions mismatch

The working conditions of the ship will have a great impact on the radiated noise of the ship. Even if the same ship is traveling in the same sea area, different working conditions will produce different radiated noise, thus affecting the accuracy of target recognition. Especially in the case of working condition mismatch, the correct rate of the recognition results will be greatly reduced. To address this problem, an intelligent underwater acoustic target recognition method based on knowledge distillation is proposed to improve the recognition accuracy. Auditory features are used as inputs to the system, and knowledge distillation is utilized to learn the intrinsic connection of target features under different working conditions. The teacher network, trained from a large amount of existing working condition data, is used to assist the student network (trained from a small amount of working condition data) to solve the working condition mismatch problem under different conditions. Tests were conducted using ship radiated noise datasets under four working conditions. The results show that the proposed method outperforms the other methods in all kinds of working condition mismatch problems, which demonstrates its intelligence and practicality in engineering problems.