MSMT-TL-Based Source Localization Estimation in Uncertain Environments of Shallow Seas

As a crucial research direction in underwater acoustics, passive acoustic source localization is essential for naval defense systems. In recent years, localization estimation methods powered by deep learning have been very promising in this field. However, due to the lack of real oceanic data, researchers have to use the simulation data generated by the Kraken acoustic tool for model training. Although the shortcoming of the lack of real oceanic data is overcome, another challenge is raised: the performance of the model may deteriorate severely because of the difference between the marine environment and the simulation conditions. This article proposes a multiscale multitask transfer learning (MSMT-TL)-based source localization estimation method to solve this issue. The novel approach applies multiscale feature extraction techniques in a multitask learning framework, integrating transfer learning strategies to minimize the adverse effects of simulation-based training, thereby enhancing the generalization ability of the model. Experimental results indicate that the proposed method is effective, achieving average absolute errors of 0.03 km in range prediction and 0.14 m in depth estimation. These outcomes suggest superior performance in practical applications compared to conventional approaches that rely solely on the simulation data for training.