Scholarlink

<p>Face recognition in real-world scenarios presents significant challenges due to variations in lighting conditions, occlusions, pose changes, and low-resolution images. To address these challenges, this study proposes a Modified Residual YOLO Network-based Single Candidate Optimization algorithm, which integrates a Modified Residual Network with YOLOv2 to improve recognition accuracy while maintaining computational efficiency. Unlike conventional YOLO models, the proposed framework introduces bipolar feature extraction, enabling the network to capture both high-level global features and low-level local details, thus improving robustness against occlusions and illumination variations<b>.</b> Atailored cost function is introduced to minimize bounding box localization errors and classification inaccuracies<b>,</b> incorporating weighted penalties to enhance model precision. Additionally, a Single Candidate Optimization algorithm is employed to fine-tune hyperparameters, ensuring faster convergence and lower computational complexity<b>.</b> The performance of the proposed technique is evaluated using the benchmark datasets and the simulation results that the proposed framework outperforms existing YOLO-based models<b>, a</b>chieving an accuracy of 98.8%, a recall of 98.0%, and an F1-score of 98.1%while maintainingalower computational cost than YOLOv3, YOLOv4, and YOLOv5. The experimental results also indicate that the tailored cost function reduces misclassification errors by 6.1% and accelerates convergence by 20%<b>,</b> making the proposed framework highly suitable for real-time face recognition applications in resource-constrained environmentssuch asIoT-based surveillance and smart security systems<b>.</b></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

High-accuracy face recognition with the MRYOLO-SCO algorithm: combining residual networks and YOLOv2 for improved precision

  • 期刊论文

  • 发表日期：

    2025年5月12日

查看全文
  • Anitha Ettiyappan,
  • Hari kumar palani,
  • Kavitha Duraipandian,
  • Nandhini Selvaganapathy

摘要

Face recognition in real-world scenarios presents significant challenges due to variations in lighting conditions, occlusions, pose changes, and low-resolution images. To address these challenges, this study proposes a Modified Residual YOLO Network-based Single Candidate Optimization algorithm, which integrates a Modified Residual Network with YOLOv2 to improve recognition accuracy while maintaining computational efficiency. Unlike conventional YOLO models, the proposed framework introduces bipolar feature extraction, enabling the network to capture both high-level global features and low-level local details, thus improving robustness against occlusions and illumination variations. Atailored cost function is introduced to minimize bounding box localization errors and classification inaccuracies, incorporating weighted penalties to enhance model precision. Additionally, a Single Candidate Optimization algorithm is employed to fine-tune hyperparameters, ensuring faster convergence and lower computational complexity. The performance of the proposed technique is evaluated using the benchmark datasets and the simulation results that the proposed framework outperforms existing YOLO-based models, achieving an accuracy of 98.8%, a recall of 98.0%, and an F1-score of 98.1%while maintainingalower computational cost than YOLOv3, YOLOv4, and YOLOv5. The experimental results also indicate that the tailored cost function reduces misclassification errors by 6.1% and accelerates convergence by 20%, making the proposed framework highly suitable for real-time face recognition applications in resource-constrained environmentssuch asIoT-based surveillance and smart security systems.