Scholarlink

This study introduces YOLOv11-SOCeNNs, a novel integration of Second-Order Cellular Neural Networks (SOCeNNs) with YOLOv11, designed to enhance object detection under noisy conditions (turbidity < 50%, small objects < 32 × 32 pixels, illumination 50–500 lx). Evaluated on an 11,027-image dataset, the model achieved a mean Average Precision (mAP)@0.5 of 0.97587, surpassing YOLOv11 by 8.698%, with real-time performance of 87 frames per second (fps) on a GeForce RTX 3080 Ti (27.6B floating-point operations, FLOPs). A pruned version reduced computational cost to 22.0B FLOPs, achieving 92 fps while retaining at 0.97587. Robustness analysis showed a 5.3% mAP improvement in turbid conditions (Table 3), with statistical significance (p < 0.05), advancing Artificial Intelligence (AI) applications in real-time detection for aquaculture, underwater robotics, and wildlife monitoring.

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 Robust Hybrid Architecture for Fish Classification and Disease Detection in Turbid Aquaculture Environments

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Quang H. Doan,
  • Hoan Q. Nguyen,
  • Truyen T. Nguyen,
  • Ban V. Pham,
  • Duc Anh Duong

摘要

This study introduces YOLOv11-SOCeNNs, a novel integration of Second-Order Cellular Neural Networks (SOCeNNs) with YOLOv11, designed to enhance object detection under noisy conditions (turbidity < 50%, small objects < 32 × 32 pixels, illumination 50–500 lx). Evaluated on an 11,027-image dataset, the model achieved a mean Average Precision (mAP)@0.5 of 0.97587, surpassing YOLOv11 by 8.698%, with real-time performance of 87 frames per second (fps) on a GeForce RTX 3080 Ti (27.6B floating-point operations, FLOPs). A pruned version reduced computational cost to 22.0B FLOPs, achieving 92 fps while retaining at 0.97587. Robustness analysis showed a 5.3% mAP improvement in turbid conditions (Table 3), with statistical significance (p < 0.05), advancing Artificial Intelligence (AI) applications in real-time detection for aquaculture, underwater robotics, and wildlife monitoring.