Scholarlink

<p>Underwater image processing has lots of potential for exploring the depths. It is being used in a diversity of applications, comprising scanning underwater landscapes and powering autonomous underwater vehicles. Marine ecosystems have a lot of issues like pollution, climate change, and habitat loss, which threaten biodiversity and food security worldwide. This can significantly reduce the ability to identify objects. A lightweight framework for deep and machine learning-based underwater object detection is presented to overcome these problems. Initially, data are collected from the underwater object detection dataset. Then, a pre-processing system is used to upsurge underwater visibility by merging the Max-RGB method with shades of gray. Next, object detection is performed based on YOLOv8 with squeeze excitation and Re-attention vision transformer (YOLOv8SE-ReViT). After that, features are extracted using a dilated convolution-based Dense Net (DC-DN). Finally, for object classification, a Radial Basis Function Support Vector Machine (RBF-SVM) is utilized. The hyperparameters in the approach are classified by the chaotic puma optimization (CPO) approach. The experimental results achieve an accuracy of 99.15%, precision of 98.92%, recall of 98.89%, F1-score of 99.1%, and IoU of 98.52%, which shows that detection and classification are fast and accurate in assisting the robot to reach underwater working operations.</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

A lightweight transformer-enhanced YOLOv8 framework for robust underwater object detection and classification

  • 期刊论文

  • 发表日期：

    2026年7月28日

查看全文
  • Dhyanendra Jain,
  • Minakshi Tomer,
  • Shyla,
  • Tripti Rathee,
  • Shweta Sharma,
  • Mamta Gautam

摘要

Underwater image processing has lots of potential for exploring the depths. It is being used in a diversity of applications, comprising scanning underwater landscapes and powering autonomous underwater vehicles. Marine ecosystems have a lot of issues like pollution, climate change, and habitat loss, which threaten biodiversity and food security worldwide. This can significantly reduce the ability to identify objects. A lightweight framework for deep and machine learning-based underwater object detection is presented to overcome these problems. Initially, data are collected from the underwater object detection dataset. Then, a pre-processing system is used to upsurge underwater visibility by merging the Max-RGB method with shades of gray. Next, object detection is performed based on YOLOv8 with squeeze excitation and Re-attention vision transformer (YOLOv8SE-ReViT). After that, features are extracted using a dilated convolution-based Dense Net (DC-DN). Finally, for object classification, a Radial Basis Function Support Vector Machine (RBF-SVM) is utilized. The hyperparameters in the approach are classified by the chaotic puma optimization (CPO) approach. The experimental results achieve an accuracy of 99.15%, precision of 98.92%, recall of 98.89%, F1-score of 99.1%, and IoU of 98.52%, which shows that detection and classification are fast and accurate in assisting the robot to reach underwater working operations.