Scholarlink

<p>Intelligent surveillance in IoT environments demands accurate and real-time tracking of multiple objects under varying conditions, especially when using thermal imaging data from sensors like the FLIR A65 IR Temperature Sensor. Traditional tracking models often face challenges in handling heterogeneous data representations, low-contrast thermal inputs, and dynamic object behaviors, leading to suboptimal accuracy and processing delays. To address these limitations, a novel method called the Adaptive Multi-Level Thermal Surveillance Network is introduced. This approach combines a Lightweight Single Shot Multi-Box Detector with a Dynamic Heterogeneous Graph Attention Mechanism, where network parameters are optimized using an Improved Sea Lion Optimization algorithm. The input dataset includes newly annotated thermal frames with varying dog instances, incorporating both long-fur and short-fur breeds exhibiting distinct thermal signatures. Pre-processing is performed using the Adaptive Morphological Wavelet Perona–Malik Filter Algorithm to enhance edge details and suppress noise, while feature extraction is carried out using an Elastic Decision Transformer to capture complex object dynamics. The proposed system achieves a notable tracking accuracy of 97.8%, offering fast execution suitable for edge-based IoT deployment and robust adaptability to complex thermal environments.</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

Adaptive smart surveillance for IoT using improved dynamic lightweight sea lion heterogeneous graph-based multi-box attention detector

  • 期刊论文

  • 发表日期：

    2026年7月16日

查看全文
  • M. Pala Prasad Reddy,
  • A. Sathya,
  • J. Jeyalakshmi,
  • Shree Ramalingam Balasubramanian,
  • I. Kumaraswamy

摘要

Intelligent surveillance in IoT environments demands accurate and real-time tracking of multiple objects under varying conditions, especially when using thermal imaging data from sensors like the FLIR A65 IR Temperature Sensor. Traditional tracking models often face challenges in handling heterogeneous data representations, low-contrast thermal inputs, and dynamic object behaviors, leading to suboptimal accuracy and processing delays. To address these limitations, a novel method called the Adaptive Multi-Level Thermal Surveillance Network is introduced. This approach combines a Lightweight Single Shot Multi-Box Detector with a Dynamic Heterogeneous Graph Attention Mechanism, where network parameters are optimized using an Improved Sea Lion Optimization algorithm. The input dataset includes newly annotated thermal frames with varying dog instances, incorporating both long-fur and short-fur breeds exhibiting distinct thermal signatures. Pre-processing is performed using the Adaptive Morphological Wavelet Perona–Malik Filter Algorithm to enhance edge details and suppress noise, while feature extraction is carried out using an Elastic Decision Transformer to capture complex object dynamics. The proposed system achieves a notable tracking accuracy of 97.8%, offering fast execution suitable for edge-based IoT deployment and robust adaptability to complex thermal environments.