A study on the weakly supervised deep learning for active sonar target classification algorithm using Meta Pseudo Labels

Current target classification using active sonar relies heavily on the expertise and sustained concentration of sonar operators over extended periods. Recently, neural network-based target classification methods for the active sonar system have been researched, but they suffer from difficulty in obtaining a sufficient amount of labeled data. To address this problem, we developed a weakly supervised deep learning algorithm for active sonar target classification in this paper. The proposed system consists of a Convolutional Recurrent Neural Network (CRNN) with Meta Pseudo Labels to handle both labeled and unlabeled active sonar data. To evaluate our method, a simulation for target classification was performed, and the results show that the proposed method achieves significantly improved accuracy compared to the baseline model.