3D Pulse Image Detection and Pulse Pattern Recognition Based on Subtle Motion Magnification Technology.
10.12455/j.issn.1671-7104.240567
- Author:
Chongyang YAO
1
;
Yongxin CHOU
2
;
Zhiwei LIANG
1
;
Haiping YANG
2
;
Jicheng LIU
2
;
Dongmei LIN
3
Author Information
1. Yancheng Institute of Technology, Yancheng,
2. Suzhou University of Technology, Suzhou,
3. Lanzhou University of Technology, Lanzhou,
- Publication Type:Journal Article
- Keywords:
binocular vision reconstruction;
machine learning;
pulse detection;
pulse pattern recognition
- MeSH:
Algorithms;
Imaging, Three-Dimensional/methods*;
Pattern Recognition, Automated;
Medicine, Chinese Traditional;
Motion;
Humans;
Pulse;
Signal Processing, Computer-Assisted;
Machine Learning
- From:
Chinese Journal of Medical Instrumentation
2025;49(3):255-262
- CountryChina
- Language:Chinese
-
Abstract:
To address the problem of large reconstruction errors in 3D pulse signals caused by excessively small out-of-plane displacement of the contact membrane in the existing traditional Chinese medicine fingertip tactile binocular vision detection technology, this study proposes a 3D pulse image detection method based on subtle motion magnification technology and explores its application in pulse pattern recognition. Firstly, a 3D pulse image detection system based on binocular vision to obtain pulse image signals is developed as experimental data. Then, the phase motion video magnification algorithm is used to amplify the original signals, and the amplified signals are reconstructed in three dimensions to obtain 3D pulse signals. On this basis, nine features are extracted from the 3D pulse signals and features selection is performed using a two-sample Kolmogorov-Smirnov test. Finally, machine learning algorithms such as decision trees and random forests are used to identify the five types of pulse conditions: deep pulse, intermittent pulse, flooding pulse, slippery pulse, and rapid pulse. The experimental results show that compared to the methods without subtle motion magnification technology, the proposed method significantly improves waveform clarity, amplitude stability, and periodic regularity. Meanwhile, the average accuracy in pulse pattern recognition reaches 96.29%±0.26%.
