Scholarlink

<p>This study investigates and optimizes a deep learning framework for Direction of Arrival (DoA) estimation of marine acoustic sources using a Convolutional Neural Network (CNN) architecture with channel attention. While the model’s components such as multichannel filtering and attention mechanisms are built on established designs, the novelty lies in its rigorous application and evaluation on real world marine acoustic signals, including ship, whale, and shrimp sounds, across a broad range of Signal to Noise Ratio (SNR) conditions. DoA estimation is formulated as a classification task, targeting azimuth angles from 0° to 180° in 15° intervals. By processing raw multichannel input signals, the model captures relevant spatiotemporal patterns through initial convolutional layers and enhances feature representations using a channel attention module. These refined features are then passed to dense layers for robust classification. Experimental evaluation demonstrates that the model achieves average accuracy, 99.90% for ship noise, 99.37% for shrimp signals, and 95.52% for whale signals, outperforming traditional algorithms like Delay and Sum Beamforming (DSB) and MUltiple SIgnal Classification (MUSIC), particularly under low SNR conditions. The model also demonstrates strong robustness across a wide range of SNR conditions, achieving an average accuracy across tested signals of 98.69% at +10 dB and maintaining high performance even under noisy conditions, with 96.68% accuracy at -10 dB. This research offers a robust, data driven approach to underwater signal processing, contributing to marine habitat monitoring, acoustic noise management, and navigation systems. The proposed method presents a scalable solution for real time applications in complex oceanographic environments.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Direction of Arrival Estimation using Deep Learning: A Novel CNN-Based Approach for Marine Acoustic Monitoring under Varying Noise Conditions

  • 期刊论文

  • 发表日期：

    2025年8月20日

查看全文
  • Rashida K,
  • Minu A Pillai,
  • Supriya M H

摘要

This study investigates and optimizes a deep learning framework for Direction of Arrival (DoA) estimation of marine acoustic sources using a Convolutional Neural Network (CNN) architecture with channel attention. While the model’s components such as multichannel filtering and attention mechanisms are built on established designs, the novelty lies in its rigorous application and evaluation on real world marine acoustic signals, including ship, whale, and shrimp sounds, across a broad range of Signal to Noise Ratio (SNR) conditions. DoA estimation is formulated as a classification task, targeting azimuth angles from 0° to 180° in 15° intervals. By processing raw multichannel input signals, the model captures relevant spatiotemporal patterns through initial convolutional layers and enhances feature representations using a channel attention module. These refined features are then passed to dense layers for robust classification. Experimental evaluation demonstrates that the model achieves average accuracy, 99.90% for ship noise, 99.37% for shrimp signals, and 95.52% for whale signals, outperforming traditional algorithms like Delay and Sum Beamforming (DSB) and MUltiple SIgnal Classification (MUSIC), particularly under low SNR conditions. The model also demonstrates strong robustness across a wide range of SNR conditions, achieving an average accuracy across tested signals of 98.69% at +10 dB and maintaining high performance even under noisy conditions, with 96.68% accuracy at -10 dB. This research offers a robust, data driven approach to underwater signal processing, contributing to marine habitat monitoring, acoustic noise management, and navigation systems. The proposed method presents a scalable solution for real time applications in complex oceanographic environments.