A Survey of Underwater Acoustic Data Classification Methods Using Deep Learning for Shoreline Surveillance.

Area of Science:

• Marine technology
• Artificial intelligence
• Signal processing

Background:

• Underwater object classification is vital for shoreline surveillance.
• Deep learning offers advanced solutions for analyzing complex sonar data.
• Existing methods face challenges with limited annotated sonar datasets.

Purpose of the Study:

• To provide a systematic overview of deep learning methods for underwater sonar object classification.
• To identify key components driving advancements in this field.
• To discuss datasets and data augmentation techniques for improving model performance.

Main Methods:

• Deep learning with convolutional layers.
• Biologically inspired feature extraction filters.
• Frequency and time-frequency analysis for data classification.
• Machine learning for signal feature extraction.
• Transfer learning approaches.

Main Results:

• Identified five core components in current deep learning for sonar classification.
• Highlighted the importance of data augmentation for scarce datasets.
• Systematically described state-of-the-art techniques for vessel and object detection.

Conclusions:

• Deep learning methods are rapidly advancing underwater sonar data analysis.
• Addressing data scarcity through augmentation is critical for real-world applications.
• This work provides a foundational understanding for future research in sonar intelligence.