Scholarlink

<p>Heart rate estimation from videos is crucial and facilitates non-contact health monitoring with applications in patient care, human interaction, and sports. Specifically, the variations of heart rate are estimated by photoplethysmography (PPG), which measures the subtle changes of skin color based on the disparity of the heart pump. However, the existing methods of PPG signal extraction are affected by the motion artifacts in long-range facial videos, and the extraction process is prone to interference from illumination variation resulting in limited performance. To tackle these limitations, this research proposes the Hybrid Bhattacharya and Euclidean distance-based Equidae Fox Convolution Neural Network (HBE2F-CNN) framework for assessing the heart rate estimation. In the proposed approach, the Spatio-SINet-based Feature Extraction effectively captures the spatial and temporal patterns to attain robust prediction results. Furthermore, the effectiveness of the proposed approach is enhanced by the integration of the Equidae Cochleoid Fox Optimization (ECHO) Algorithm, which reduces the gradient disappearance problem and generates effective convergence. From this perspective, the ability of the prediction task is improved by the incorporation of the distance model, which reduces the error rate and achieves significant results appropriately. Experimental results demonstrate that the HBE2F-CNN model attains superior performance in terms of error metrics under a higher training percentage of 90%, achieving the value of 4.99 Mean Square Error, 2.23 Root Mean Square Error, 0.02 Root Mean Square Logarithmic Error (RMSLE), 1.43 Symmetric Mean Absolute Percentage Error (SMAPE), and 0.95 R<sup>2</sup> values.</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

HBE2F-CNN: Hybrid Bhattacharya and Euclidean Distance-Based Deep Learning Model for Heart Rate Estimation

  • 期刊论文

  • 发表日期：

    2025年6月19日

查看全文
  • Minal Chandrakant Toley,
  • Vishal Shirsath,
  • Ajitkumar Pundge

摘要

Heart rate estimation from videos is crucial and facilitates non-contact health monitoring with applications in patient care, human interaction, and sports. Specifically, the variations of heart rate are estimated by photoplethysmography (PPG), which measures the subtle changes of skin color based on the disparity of the heart pump. However, the existing methods of PPG signal extraction are affected by the motion artifacts in long-range facial videos, and the extraction process is prone to interference from illumination variation resulting in limited performance. To tackle these limitations, this research proposes the Hybrid Bhattacharya and Euclidean distance-based Equidae Fox Convolution Neural Network (HBE2F-CNN) framework for assessing the heart rate estimation. In the proposed approach, the Spatio-SINet-based Feature Extraction effectively captures the spatial and temporal patterns to attain robust prediction results. Furthermore, the effectiveness of the proposed approach is enhanced by the integration of the Equidae Cochleoid Fox Optimization (ECHO) Algorithm, which reduces the gradient disappearance problem and generates effective convergence. From this perspective, the ability of the prediction task is improved by the incorporation of the distance model, which reduces the error rate and achieves significant results appropriately. Experimental results demonstrate that the HBE2F-CNN model attains superior performance in terms of error metrics under a higher training percentage of 90%, achieving the value of 4.99 Mean Square Error, 2.23 Root Mean Square Error, 0.02 Root Mean Square Logarithmic Error (RMSLE), 1.43 Symmetric Mean Absolute Percentage Error (SMAPE), and 0.95 R2 values.