Scholarlink

Abstract <p>Slope stability is one of the key issues for the safe operation of muddy submarine channels and a crucial factor in ensuring smooth maritime traffic. The stability identification of muddy submarine channels relies on acoustic images of the channel, but it often faces challenges such as limited acoustic image data, complex features, and insufficient accuracy. To address these issues, this study proposes a deep learning recognition model based on an improved VGG-16 network. By constructing a multimodal feature enhancement method using bidimensional empirical mode decomposition and discrete wavelet transform, the channel dataset can be effectively augmented and the expression of related features reinforced. On this basis, the VGG-16 network structure was optimized by replacing the fully connected layers with lightweight convolutional modules, resulting in a feature extraction network composed of three convolutional layers. This significantly enhanced the model’s ability to analyze complex acoustic features, effectively addressing the challenges of limited data and feature complexity. Moreover, the use of transfer learning based on laboratory data further improved the accuracy of model recognition. Experimental results indicate that after 1000 iterations, the model achieves a classification accuracy of 98.5% on the test set, outperforming traditional convolutional neural network algorithms. This method can provide reliable technical support for the evaluation of submarine channel stability.</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

Deep Learning-Based Stability Identification of Acoustic Images for Muddy Submarine Channels

  • 期刊论文

  • 发表日期：

    2025年12月1日

查看全文
  • Yuchen Zhang,
  • Cunyong Zhang

摘要

Abstract

Slope stability is one of the key issues for the safe operation of muddy submarine channels and a crucial factor in ensuring smooth maritime traffic. The stability identification of muddy submarine channels relies on acoustic images of the channel, but it often faces challenges such as limited acoustic image data, complex features, and insufficient accuracy. To address these issues, this study proposes a deep learning recognition model based on an improved VGG-16 network. By constructing a multimodal feature enhancement method using bidimensional empirical mode decomposition and discrete wavelet transform, the channel dataset can be effectively augmented and the expression of related features reinforced. On this basis, the VGG-16 network structure was optimized by replacing the fully connected layers with lightweight convolutional modules, resulting in a feature extraction network composed of three convolutional layers. This significantly enhanced the model’s ability to analyze complex acoustic features, effectively addressing the challenges of limited data and feature complexity. Moreover, the use of transfer learning based on laboratory data further improved the accuracy of model recognition. Experimental results indicate that after 1000 iterations, the model achieves a classification accuracy of 98.5% on the test set, outperforming traditional convolutional neural network algorithms. This method can provide reliable technical support for the evaluation of submarine channel stability.