Underwater Acoustic Signal Recognition Based on Salient Feature

TL;DR

This paper tackles underwater acoustic signal recognition in complex environments where traditional time-frequency analysis and rule-based systems struggle. It introduces a Demon Spectrum Line Feature Extraction that converts Demon spectrograms into a compact five-dimensional feature vector with components $f_s$, $f_b$, $A_{avg}$, $f_{s,max}$, and $f_{b,max}$, coupled with a cascaded neural network that first predicts broad ship categories and then refines to specific types. Experiments on a private 300-sample dataset demonstrate robust performance and coherent confusion-matrix results for 4-class and 10-class classifications, validating the approach’s practicality. The work advances automated spectrogram-based feature learning for ship-type recognition and suggests data augmentation and multi-level cascading as avenues to boost real-world applicability in underwater monitoring.

Abstract

With the rapid advancement of technology, the recognition of underwater acoustic signals in complex environments has become increasingly crucial. Currently, mainstream underwater acoustic signal recognition relies primarily on time-frequency analysis to extract spectral features, finding widespread applications in the field. However, existing recognition methods heavily depend on expert systems, facing limitations such as restricted knowledge bases and challenges in handling complex relationships. These limitations stem from the complexity and maintenance difficulties associated with rules or inference engines. Recognizing the potential advantages of deep learning in handling intricate relationships, this paper proposes a method utilizing neural networks for underwater acoustic signal recognition. The proposed approach involves continual learning of features extracted from spectra for the classification of underwater acoustic signals. Deep learning models can automatically learn abstract features from data and continually adjust weights during training to enhance classification performance.

Figures (5)

• Figure 1: Flowchart of the algorithm
• Figure 2: Demon spectrum
• Figure 3: structure of the module
• Figure 4: The results of the 4-class confusion matrix
• Figure 5: The results of the 10-class confusion matrix