Individual Identification of Active Sonar Transmitter Based on Multi-Feature Fusion and Domain Adaptation

The individual identification of active sonar transmitters is engineered to differentiate individuals based on the variations in transmitted signal performance, which arise from hardware discrepancies among them. Subsequently, distinctive features are extracted from the received signals through advanced signal processing techniques. To enhance the individual identification performance of transmitters in complex underwater acoustic environments, a method for individual identification of active sonar transmitters based on multi-feature fusion and domain adaptation is proposed. Firstly, four transformation features are extracted from the received pulse signal. Secondly, a basic network framework for multi-feature fusion individual identification was designed, utilizing spatially separable convolution and center loss. Additionally, to address the degradation of identification performance due to changes in underwater acoustic channel distance, it is proposed to introduce a domain adaptive layer into the basic network framework for calculating the maximum mean discrepancy (MMD) distance between domains. By incorporating the minimization of MMD distance into the loss function, the distribution discrepancy between labeled data in the source domain and unlabeled data in the target domain at unknown distances can be narrowed, enabling the model to learn the “invariant” characteristics of individual transmitters under varying distance conditions. The experimental results demonstrate that this method can effectively be applied to individual identification across various channel distances, achieving an average identification accuracy exceeding 80% on the experimental dataset. The extracted features exhibit good stability and robustness.