Scholarlink

<p>Corals have sensitive and medicinal properties, making it crucial for monitoring the underwater ecosystem. Accurate and rapid coral identification is the primary focus in coral classification within the Internet of Underwater Things (IoUT) while preserving coral information to maintain confidentiality. Autonomous Underwater Vehicles (AUVs) capture coral images autonomously from the targeted underwater area. The primary objectives in IoUT are to reduce energy consumption, minimize delay, and improve communication time, while also enhancing accuracy and preserving data integrity. We propose a Transfer Federated Learning (TFL)-based edge-fog-cloud (TFL-IoUT) paradigm as a solution for fast and accurate coral classification in an energy-efficient manner in the IoUT to improve network lifetime. The proposed TFL-IoUT system operates in three hierarchical tiers: edge, fog, and cloud. The AUVs present at the edge layer transfer the local model parameters to the upper layers. Fog then transmits the model parameters to the cloud for fast computation. The proposed TFL-IoUT performs local training at the edge layer using a local dataset and transmits model parameters only to the cloud through fog, rather than transmitting the entire dataset, thereby improving data transmission time and privacy, and making the system more efficient. We perform TFL-IoUT learning in the Python-based Flower framework. Simulation results demonstrate that the proposed TFL-IoUT system achieves an accuracy of 98.7% within just 20 global rounds, outperforming the existing centralized approach, which attains 97.65% accuracy over 1300 rounds using the RSMAS dataset. Because the proposed TFL-IoUT system involves edge trains, which means local models are trained on local data and only updated model parameters are transmitted to the cloud through fog, this approach reduces communication overhead while improving accuracy. The proposed system improves energy consumption by 56% and computation time by 38% over a centralized cloud-based system. Therefore, the proposed TFL-IoUT system improves computational efficiency by enabling faster learning and response times in an energy-efficient manner.</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

TFL-IoUT: Transfer federated learning for privacy preserving coral classification on the internet of underwater things

  • 期刊论文

  • 发表日期：

    2026年7月3日

查看全文
  • Kamalika Bhattacharjya,
  • Debashis De

摘要

Corals have sensitive and medicinal properties, making it crucial for monitoring the underwater ecosystem. Accurate and rapid coral identification is the primary focus in coral classification within the Internet of Underwater Things (IoUT) while preserving coral information to maintain confidentiality. Autonomous Underwater Vehicles (AUVs) capture coral images autonomously from the targeted underwater area. The primary objectives in IoUT are to reduce energy consumption, minimize delay, and improve communication time, while also enhancing accuracy and preserving data integrity. We propose a Transfer Federated Learning (TFL)-based edge-fog-cloud (TFL-IoUT) paradigm as a solution for fast and accurate coral classification in an energy-efficient manner in the IoUT to improve network lifetime. The proposed TFL-IoUT system operates in three hierarchical tiers: edge, fog, and cloud. The AUVs present at the edge layer transfer the local model parameters to the upper layers. Fog then transmits the model parameters to the cloud for fast computation. The proposed TFL-IoUT performs local training at the edge layer using a local dataset and transmits model parameters only to the cloud through fog, rather than transmitting the entire dataset, thereby improving data transmission time and privacy, and making the system more efficient. We perform TFL-IoUT learning in the Python-based Flower framework. Simulation results demonstrate that the proposed TFL-IoUT system achieves an accuracy of 98.7% within just 20 global rounds, outperforming the existing centralized approach, which attains 97.65% accuracy over 1300 rounds using the RSMAS dataset. Because the proposed TFL-IoUT system involves edge trains, which means local models are trained on local data and only updated model parameters are transmitted to the cloud through fog, this approach reduces communication overhead while improving accuracy. The proposed system improves energy consumption by 56% and computation time by 38% over a centralized cloud-based system. Therefore, the proposed TFL-IoUT system improves computational efficiency by enabling faster learning and response times in an energy-efficient manner.