Scholarlink

<p>Side-scan sonar (SSS) is widely utilized in fields such as marine ecology, underwater archaeology, and the identification and damage assessment of underwater structures. However, due to the complexity of the underwater environment and inherent limitations of sonar systems, raw SSS images typically exhibit complex and severe noise, particularly non-uniform speckle noise. To address the challenge of denoising SSS images exhibiting non-uniform speckle noise, this paper proposes a novel algorithm, SIDA, based on deep learning and compressed sensing (CS). The proposed algorithm consists of three main components: compressed sensing sampling, initialization, and network reconstruction. In the network reconstruction stage, CoordConv layers are first employed for feature extraction. By introducing absolute spatial position information as additional input channels, these layers help the network learn spatial relationships among features and effectively suppress non-uniform noise. Subsequently, a multi-scale residual module processes the extracted features, capturing more image details and structural information, thereby enhancing the model’s representational capacity. An edge-enhanced attention module is then incorporated, combined with an RNN network, to leverage spatial semantic information from multiple directions. This enhances object edge representation and enriches the overall feature representation. Experimental results on both real and simulated SSS images demonstrate that the SIDA algorithm effectively removes speckle noise while minimizing the loss of image quality and edge details.</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

Side-scan sonar image denoising algorithm based on deep learning and compressed sensing

  • 期刊论文

  • 发表日期：

    2025年7月3日

查看全文
  • Jingwen Li,
  • Xinghai Yang,
  • Hongxiu Yang,
  • Jingjing Wang

摘要

Side-scan sonar (SSS) is widely utilized in fields such as marine ecology, underwater archaeology, and the identification and damage assessment of underwater structures. However, due to the complexity of the underwater environment and inherent limitations of sonar systems, raw SSS images typically exhibit complex and severe noise, particularly non-uniform speckle noise. To address the challenge of denoising SSS images exhibiting non-uniform speckle noise, this paper proposes a novel algorithm, SIDA, based on deep learning and compressed sensing (CS). The proposed algorithm consists of three main components: compressed sensing sampling, initialization, and network reconstruction. In the network reconstruction stage, CoordConv layers are first employed for feature extraction. By introducing absolute spatial position information as additional input channels, these layers help the network learn spatial relationships among features and effectively suppress non-uniform noise. Subsequently, a multi-scale residual module processes the extracted features, capturing more image details and structural information, thereby enhancing the model’s representational capacity. An edge-enhanced attention module is then incorporated, combined with an RNN network, to leverage spatial semantic information from multiple directions. This enhances object edge representation and enriches the overall feature representation. Experimental results on both real and simulated SSS images demonstrate that the SIDA algorithm effectively removes speckle noise while minimizing the loss of image quality and edge details.