中圖分類號： TP29
文獻標識碼： A
DOI： 10.20044/j.csdg.2097-1788.2022.03.012
引用格式： 陳澤輝，熊繼平，李金紅，等. 基于深度學習的有效iPPG信號識別研究[J].網絡安全與數據治理，2022，41(3)：74-80.

Research on effective iPPG signal recognition based on deep learning

Abstract： Imaging photoplethysmography(iPPG) is a technique that uses a camera to remotely measure blood volume pulses. It has become a promising technique for early screening of atherosclerosis and cardiovascular disease. However, the iPPG signal contains a large amount of noise, and the traditional method of selecting an effective signal has low accuracy and poor universality. Based on the above problems, this paper designs and develops an effective iPPG signal recognition software based on Python. The software collects face video and locates the face through the camera, extracts the iPPG signal in the dynamic region of interest, and preprocesses the signal. Using the deep learning-based iPPG signal discrimination model for validity discrimination, the average accuracy rate is as high as 95.1%, which has good accuracy and stability for iPPG signal discrimination in practical application scenarios. This paper also verifies the feasibility of using deep learning for non-contact feasibility signal validity discrimination.

Key words : imaging photoplethysmography; signal processing; facial video processing; deep learning