A Deep Convolutional Neural Network Inspired by Auditory Perception for Underwater Acoustic Target Recognition.

Area of Science:

• Signal Processing
• Machine Learning
• Bio-inspired Computing

Background:

• Underwater acoustic target recognition (UATR) is crucial but challenging due to complex marine environments.
• Existing methods struggle with the intricacies of ship-radiated noise.
• Neural mechanisms of auditory perception offer a promising avenue for improved UATR.

Purpose of the Study:

• To propose a novel end-to-end deep neural network, the Auditory Perception inspired Deep Convolutional Neural Network (ADCNN), for UATR.
• To leverage bio-inspired mechanisms for enhanced feature extraction and classification of underwater acoustic targets.
• To demonstrate the efficacy of the ADCNN model in handling ship-radiated noise.

Main Methods:

• Designed a bank of multi-scale deep convolution filters inspired by frequency component perception to decompose raw time-domain signals.
• Employed plasticity-inspired random initialization and learned optimization for filter parameters.
• Utilized max-pooling and fully connected layers for feature extraction from decomposed signals.
• Integrated features in fusion layers for deep representation and classification of underwater acoustic targets.

Main Results:

• The ADCNN model successfully decomposed, modeled, and classified ship-radiated noise signals.
• Achieved a classification accuracy of 81.96%, outperforming other methods in contrast experiments.
• Demonstrated efficient processing of complex underwater acoustic data.

Conclusions:

• The proposed ADCNN model effectively simulates deep acoustic information processing structures found in the auditory system.
• Auditory perception-inspired deep learning methods show significant potential for improving UATR performance.
• The ADCNN offers a robust solution for classifying underwater acoustic targets in challenging conditions.