An attention mechanism-based multi-domain feature fusion approach for active sonar target recognition

Abstract

Due to the complex and changeable marine environment, the active sonar target recognition problem has always been difficult in the field of underwater acoustics. Deep learning-based fusion recognition technology provides an effective way to solve this problem, but relying on simple concatenation strategies to fuse multi-domain features can cause information redundancy, and it is not easy to effectively mine correlation information between domains. Therefore, this paper proposes an attention mechanism-based multi-domain feature fusion approach for active sonar target recognition. By preprocessing active sonar echo signals and constructing a multi-domain feature extraction and fusion network, this method uses a one-dimensional convolutional neural network with long short-term memory (1DCNN-LSTM) and a two-dimensional convolutional neural network (2DCNN) with channel attention introduced to extract deep features from different domains. Subsequently, combining feature concatenation and constructing multi-domain cross-attention, intra- and cross-domain feature fusion is performed, which can effectively eliminate redundant information and promote inter-domain information interaction, while maximizing the retention of target features. Experimental results show that compared with single-domain methods, the network using an attention mechanism for multi-domain feature fusion strengthens cross-domain information interaction and significantly improves feature representation capability. Compared with other methods, the proposed method has obvious advantages in performance and maintains stable generalization ability in scenarios with low signal-clutter ratios.

1. Introduction

With the development of marine resource exploitation and national defense technology, the demand for underwater acoustic target recognition using sonar continues to grow (Li et al., 2024). Active sonar can achieve high-precision localization and recognition of targets in various underwater detection scenarios, holding significant importance in applications such as underwater search and rescue, fishing, seabed environment detection, and frogman/submarine detection (Huang and Li, 2019). However, the complex and variable marine environment poses numerous challenges for underwater acoustic target recognition (Domingos et al., 2022). First, during sound wave propagation, the multipath effect occurs due to reflections from the sea surface and seabed (Fang et al., 2019), while moving targets such as submarines, torpedoes, and underwater vehicles generate Doppler effects (Khan et al., 2024), leading to distortion of echo signals. Second, various types of clutter interference and noise exist in underwater environments, which may cover or confuse target signals, increasing recognition difficulty. Additionally, the high complexity and difficulty of underwater exploration result in a scarcity of samples, constraining the feature extraction capability of models and making it challenging to fully extract target feature information. Facing these problems, researchers have been continuously committed to research in this field.

Early underwater acoustic target recognition relied on manual listening to analyze the characteristics of acoustic signals (Arrabito et al., 2005), but this approach is susceptible to environmental and personnel factors, greatly limiting its recognition rate. With the development of computer science and the proposal and validation of signal processing techniques, the advancement of underwater target recognition technology has been promoted.

Researchers have adopted machine learning methods to classify targets based on feature extraction. Shin et al. (1997) observed and analyzed the relationship between target echo signals and target shape/internal structure, extracting multiple features including time-domain waveforms, frequency spectra, time–frequency spectra, and auditory perceptual characteristics. By fusing the feature information and combining various machine learning algorithms such as neural networks, they achieved correct classification of target echoes and clutter.