Multi-modal Underwater Acoustic Target Recognition

Underwater acoustic target recognition (UATR) has gained significant attention across various fields in recent years. However, traditional UATR methods, which primarily rely on data from a single sensor, have limitations in fully capturing target characteristics. This constraint ultimately impacts the accuracy and robustness of recognition. This paper presents a multi-modal UATR method that fuses data from two different sensors, thereby improving underwater target recognition accuracy. Additionally, a new training framework is proposed to facilitate the learning of consistent feature representations across diverse data sources, allowing for more effective capture of the intrinsic characteristics of acoustic data. Experimental results demonstrate that the method has positive effectiveness under various Signal-to-Noise Ratio (SNR) conditions and achieves an improvement in recognition accuracy ranging from 0.95% to 8.33%.