Scholarlink

<p>Sea cucumber, as an aquatic product of significant economic and ecological value, accurate population statistics for sea cucumbers are critical for achieving sustainable aquaculture. However, traditional manual sampling methods suffer from low efficiency, high cost, and significant errors due to sparse sampling and low coverage. Image-based approaches also struggle with efficient and accurate multi-object counting underwater, challenged by complex backgrounds, variable lighting, and target occlusion. To address these issues, this study proposes a ROV-assisted in situ density estimation for sea cucumbers via lightweight YOLOv8-FA and enhanced ByteTrack. First, the YOLOv8-FA algorithm was designed by replacing C2F modules with C3FA modules to enhance detection efficiency. Second, improvements were made to the ByteTrack framework through optimized target association and re-identification mechanisms, complemented by line-crossing counting to reduce missed and false detections. Finally, precise calculation of scanned areas via underwater camera geometric modeling enabled accurate sea cucumber density estimation. Experimental results demonstrate the outstanding performance of the proposed framework for sea cucumber density estimation. By integrating the optimized detection and tracking algorithms, the model achieves an average counting accuracy of 87.5% (corresponding to a low normalized mean absolute error of 12.5%), a decisive improvement over the baseline method. This achievement is supported by the lightweight YOLOv8-FA detector. More importantly, the enhanced ByteTrack with a line-crossing strategy effectively overcame issues such as ID switches and trajectory fragmentation, ensuring the reliability of the final count. All key metrics significantly outperform comparative methods, validating the effectiveness of this study. Furthermore, this method is not only applicable to sea cucumber farming but can also be extended to other marine organisms, providing critical references for precision aquaculture and ecological monitoring technology advancement.</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

ROV-assisted in situ density estimation for sea cucumbers via lightweight YOLOv8-FA and enhanced ByteTrack

  • 期刊论文

  • 发表日期：

    2025年8月23日

查看全文
  • Yanqiang Yang,
  • Haolong Ban,
  • Junyi Wang,
  • Zejin Liu,
  • Fangqun Niu,
  • Qijun Chen,
  • Jiaxu Zhang,
  • Wei Wang,
  • Zhijun Li,
  • Yuanshan Lin

摘要

Sea cucumber, as an aquatic product of significant economic and ecological value, accurate population statistics for sea cucumbers are critical for achieving sustainable aquaculture. However, traditional manual sampling methods suffer from low efficiency, high cost, and significant errors due to sparse sampling and low coverage. Image-based approaches also struggle with efficient and accurate multi-object counting underwater, challenged by complex backgrounds, variable lighting, and target occlusion. To address these issues, this study proposes a ROV-assisted in situ density estimation for sea cucumbers via lightweight YOLOv8-FA and enhanced ByteTrack. First, the YOLOv8-FA algorithm was designed by replacing C2F modules with C3FA modules to enhance detection efficiency. Second, improvements were made to the ByteTrack framework through optimized target association and re-identification mechanisms, complemented by line-crossing counting to reduce missed and false detections. Finally, precise calculation of scanned areas via underwater camera geometric modeling enabled accurate sea cucumber density estimation. Experimental results demonstrate the outstanding performance of the proposed framework for sea cucumber density estimation. By integrating the optimized detection and tracking algorithms, the model achieves an average counting accuracy of 87.5% (corresponding to a low normalized mean absolute error of 12.5%), a decisive improvement over the baseline method. This achievement is supported by the lightweight YOLOv8-FA detector. More importantly, the enhanced ByteTrack with a line-crossing strategy effectively overcame issues such as ID switches and trajectory fragmentation, ensuring the reliability of the final count. All key metrics significantly outperform comparative methods, validating the effectiveness of this study. Furthermore, this method is not only applicable to sea cucumber farming but can also be extended to other marine organisms, providing critical references for precision aquaculture and ecological monitoring technology advancement.