Scholarlink

<p>Underwater object detection (UOD) is affected due to low visibility, color distortion, and variations in object scale and contour affects the performance of UOD. Conventional segmentation methods lacks in leveraging multi-scale contextual information in dense underwater ecological scenes. This provides poor classification accuracy and imprecise object boundaries. To overcome these limitations, this work presents an enhanced semantic segmentation model on the basis of the DeepLabv3 + for efficient and accurate UOD. The standard DeepLabv3 + model is powerful, but it has high complexity and unsuitable for real-time deployment. Therefore, this work presents a MobileNetv2 as a lightweight backbone for reducing complexity and dual attention mechanism is combined in the DeepLabv3+. Then, the Convolutional Block Attention Module (CBAM) is integrated into the encoder stage for enhancing important spatial-channel-wise features and eliminating redundant information. Then, the Efficient Channel Attention (ECA) mechanism is incorporated in the decoder stage for enhancing high-level semantic feature representation. This improves object localization and classification. Thus, the suggested model is suitable for real-time applications in underwater ecological monitoring and object detection.</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

Dual-Attention DeepLabv3 + with MobileNetV2 for efficient underwater object detection

  • 期刊论文

  • 发表日期：

    2025年12月27日

查看全文
  • J. Jeno Jasmine,
  • R. M. R. Shamija Sherryl,
  • T. Jarin,
  • M. A. Mukunthan

摘要

Underwater object detection (UOD) is affected due to low visibility, color distortion, and variations in object scale and contour affects the performance of UOD. Conventional segmentation methods lacks in leveraging multi-scale contextual information in dense underwater ecological scenes. This provides poor classification accuracy and imprecise object boundaries. To overcome these limitations, this work presents an enhanced semantic segmentation model on the basis of the DeepLabv3 + for efficient and accurate UOD. The standard DeepLabv3 + model is powerful, but it has high complexity and unsuitable for real-time deployment. Therefore, this work presents a MobileNetv2 as a lightweight backbone for reducing complexity and dual attention mechanism is combined in the DeepLabv3+. Then, the Convolutional Block Attention Module (CBAM) is integrated into the encoder stage for enhancing important spatial-channel-wise features and eliminating redundant information. Then, the Efficient Channel Attention (ECA) mechanism is incorporated in the decoder stage for enhancing high-level semantic feature representation. This improves object localization and classification. Thus, the suggested model is suitable for real-time applications in underwater ecological monitoring and object detection.