Scholarlink

<p>Monopterus albus is a widely cultivated aquaculture species in Asia, particularly in China, where it plays a crucial role in freshwater fish farming. Accurate detection and behavioral tracking of Monopterus albus during farming operations can lead to increased yield, minimized disease outbreaks, and improved product quality. To overcome the limitations of current intelligent detection and tracking methods for Monopterus albus, this paper propose an efficient detection and tracking method combining the improved YOLOv8 (LS-YOLOv8) model and DeepSort algorithm, named LSYOLO-Tracker. First, the Large Separable Kernel Attention (LSKA) module is integrated into the SPPF layer of YOLOv8, thereby optimizing multi-scale feature representation and reducing computational resource requirements. Second, the Shape-IoU loss function was adopted to optimize bounding box regression, significantly improving detection accuracy by ensuring better alignment between predicted and ground-truth shapes. Finally, ResNet18 was utilized to replace DeepSort’s original neural network, while ECA-Net is integrated to strengthen the model’s capacity for extracting fine-grained information and accurately capturing the appearance characteristics of Monopterus albus. Experiments were performed on the Monopterus albus image dataset under different background conditions. The results confirmed that the improved YOLOv8 model achieves a Precision of 97.6%, Recall of 92.6%, mAP@0.5 of 95.8%, mAP@0.5:0.95 of 46.2%, and F1-score of 95.1%. Compared to the original YOLOv8 network, these metrics improved by 1.2%, 2.7%, 2.0%, 0.5%, and 2.1%, respectively. In the target tracking part, the improved DeepSort algorithm demonstrated significant performance gains, with a MOTA of 84.3% and a MOTP of 74.2%, reflecting increases of 3.6% and 1.8%, respectively, compared to the original version. Additionally, the frequency of ID switching during the swimming process of Monopterus albus is significantly reduced, reflecting enhanced tracking stability. The experimental results confirm that the proposed method exhibits high accuracy in target detection and tracking, offering robust technical support for the aquaculture of Monopterus albus.</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

LSYOLO-Tracker: A vision algorithm for efficient Monopterus albus detection and tracking

  • 期刊论文

  • 发表日期：

    2025年12月5日

查看全文
  • Fei Zhou,
  • Wei Liao,
  • Quan Yuan,
  • Zhen Xu,
  • Yifan Zhao,
  • Wenzong Zhou,
  • Wentao Liu

摘要

Monopterus albus is a widely cultivated aquaculture species in Asia, particularly in China, where it plays a crucial role in freshwater fish farming. Accurate detection and behavioral tracking of Monopterus albus during farming operations can lead to increased yield, minimized disease outbreaks, and improved product quality. To overcome the limitations of current intelligent detection and tracking methods for Monopterus albus, this paper propose an efficient detection and tracking method combining the improved YOLOv8 (LS-YOLOv8) model and DeepSort algorithm, named LSYOLO-Tracker. First, the Large Separable Kernel Attention (LSKA) module is integrated into the SPPF layer of YOLOv8, thereby optimizing multi-scale feature representation and reducing computational resource requirements. Second, the Shape-IoU loss function was adopted to optimize bounding box regression, significantly improving detection accuracy by ensuring better alignment between predicted and ground-truth shapes. Finally, ResNet18 was utilized to replace DeepSort’s original neural network, while ECA-Net is integrated to strengthen the model’s capacity for extracting fine-grained information and accurately capturing the appearance characteristics of Monopterus albus. Experiments were performed on the Monopterus albus image dataset under different background conditions. The results confirmed that the improved YOLOv8 model achieves a Precision of 97.6%, Recall of 92.6%, mAP@0.5 of 95.8%, mAP@0.5:0.95 of 46.2%, and F1-score of 95.1%. Compared to the original YOLOv8 network, these metrics improved by 1.2%, 2.7%, 2.0%, 0.5%, and 2.1%, respectively. In the target tracking part, the improved DeepSort algorithm demonstrated significant performance gains, with a MOTA of 84.3% and a MOTP of 74.2%, reflecting increases of 3.6% and 1.8%, respectively, compared to the original version. Additionally, the frequency of ID switching during the swimming process of Monopterus albus is significantly reduced, reflecting enhanced tracking stability. The experimental results confirm that the proposed method exhibits high accuracy in target detection and tracking, offering robust technical support for the aquaculture of Monopterus albus.