Scholarlink

<p>Face recognition within telemedical and IoT-based applications plays a critical role in areas such as security, healthcare, and surveillance, where real time and reliable performance is essential. However, in uncontrolled environments such as crowded public spaces or outdoor settings with variable lighting, accuracy often deteriorates due to poor illumination, indirect viewing angles, camera face height mismatches, and partial occlusions. To address these limitations, this study introduces a deep learning based algorithm with a tree structured architecture designed to improve robustness under such challenging conditions. The proposed method partitions input data into smaller segments, with the partitioning guided by intermediate outputs of the network, thereby reducing error rates across feedback iterations and accelerating the learning cycle. Experimental results on the ORL dataset show that the proposed model achieves up to 98.4% recognition accuracy, with performance fluctuating between 96% and 98.4% across cycles, demonstrating effective precision, recall, and F1-score improvements. These results highlight its potential for enhancing face recognition performance in telemedical systems and other Internet of Things enabled applications operating in non ideal 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

Face Recognition Based on Deep Learning Convolutional Neural Network in Cloud Internet of Things Environment

  • 期刊论文

  • 发表日期：

    2025年12月26日

查看全文
  • Yaser Ghahremani,
  • Vafa Maihami

摘要

Face recognition within telemedical and IoT-based applications plays a critical role in areas such as security, healthcare, and surveillance, where real time and reliable performance is essential. However, in uncontrolled environments such as crowded public spaces or outdoor settings with variable lighting, accuracy often deteriorates due to poor illumination, indirect viewing angles, camera face height mismatches, and partial occlusions. To address these limitations, this study introduces a deep learning based algorithm with a tree structured architecture designed to improve robustness under such challenging conditions. The proposed method partitions input data into smaller segments, with the partitioning guided by intermediate outputs of the network, thereby reducing error rates across feedback iterations and accelerating the learning cycle. Experimental results on the ORL dataset show that the proposed model achieves up to 98.4% recognition accuracy, with performance fluctuating between 96% and 98.4% across cycles, demonstrating effective precision, recall, and F1-score improvements. These results highlight its potential for enhancing face recognition performance in telemedical systems and other Internet of Things enabled applications operating in non ideal environments.