Scholarlink

<p>The classification and analysis of underwater ecological objects remain challenging due to varying lighting conditions, turbidity, and complex color patterns inherent in marine environments. To address these challenges, this study proposes a novel Edge Region Segmenting Scheme (ERSS) for identifying and classifying underwater objects as living or non-living. The scheme begins with edge extraction based on color saturation variations and employs a Lateral Neural Network (LNN) that captures feature differences and structural similarities between training and testing samples. A fuzzy model is integrated into the segmentation process to handle uncertainties in edge patterns and variations in illumination. The edge saturation is evaluated using structural index values, enabling the system to effectively differentiate overlapping marine entities, such as plants, animals, and static objects. Experimental evaluations were conducted using a large-scale underwater image dataset containing 4279 annotated images. The proposed ERSS achieved a classification accuracy of 96.4%, outperforming benchmark methods, such as CDD (89.9%), IECP (91.0%), and UISS-Net (94.2%). Additionally, ERSS improved precision (97.6%), structural index factor (0.987), and reduced segmentation difference (0.0763) and processing time (0.87s). These results demonstrate that ERSS offers a significant performance boost, particularly in terms of precision and reliability under variable underwater conditions. The proposed approach is well suited for real-time marine ecology applications, including marine species monitoring, submerged object detection, and habitat mapping, where accurate classification under challenging imaging conditions is critical.</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

Edge region segmenting scheme for underwater ecology object classification using lateral neural network

  • 期刊论文

  • 发表日期：

    2025年7月2日

查看全文
  • Chinnathambi Kamatchi,
  • Selvarani Nachimuthu,
  • Sandeep Kumar Mathivanan,
  • Sangeetha Ramaswamy,
  • Sukumar Rajendran,
  • Mohd Asif Shah

摘要

The classification and analysis of underwater ecological objects remain challenging due to varying lighting conditions, turbidity, and complex color patterns inherent in marine environments. To address these challenges, this study proposes a novel Edge Region Segmenting Scheme (ERSS) for identifying and classifying underwater objects as living or non-living. The scheme begins with edge extraction based on color saturation variations and employs a Lateral Neural Network (LNN) that captures feature differences and structural similarities between training and testing samples. A fuzzy model is integrated into the segmentation process to handle uncertainties in edge patterns and variations in illumination. The edge saturation is evaluated using structural index values, enabling the system to effectively differentiate overlapping marine entities, such as plants, animals, and static objects. Experimental evaluations were conducted using a large-scale underwater image dataset containing 4279 annotated images. The proposed ERSS achieved a classification accuracy of 96.4%, outperforming benchmark methods, such as CDD (89.9%), IECP (91.0%), and UISS-Net (94.2%). Additionally, ERSS improved precision (97.6%), structural index factor (0.987), and reduced segmentation difference (0.0763) and processing time (0.87s). These results demonstrate that ERSS offers a significant performance boost, particularly in terms of precision and reliability under variable underwater conditions. The proposed approach is well suited for real-time marine ecology applications, including marine species monitoring, submerged object detection, and habitat mapping, where accurate classification under challenging imaging conditions is critical.