CN-121971064-A - Intelligent vital sign monitoring method, system and device based on migration learning mechanism and UWB radar

CN121971064ACN 121971064 ACN121971064 ACN 121971064ACN-121971064-A

Abstract

The invention discloses an intelligent vital sign monitoring method, system and device based on a migration learning mechanism and UWB radar, and relates to the technical field of radar signal processing and medical health monitoring, and the technical key points are that the monitoring method and system of the invention pre-trains a deep neural network model on simulation radar data by generating the simulation radar data so as to enable the simulation radar data to learn the basic characteristics of vital sign signals; and then, the pre-training model is finely adjusted by using a small amount of acquired real UWB radar data, so that knowledge of the pre-training model is migrated and adapted to the current specific scene and individuals, and the effect of training a large amount of data is achieved by using a small amount of real data. The invention uses the simulation data to complement the deficiency of the real data, and enables the model to be rapidly adapted to the new monitoring environment and the target individual through the migration learning mechanism, thereby finally achieving the non-contact vital sign monitoring with high accuracy, high stability and high practicability.

Inventors

ZHANG FENGYUN
YE JINYUAN
WANG FUDI
WANG MENGKUI
WU JIENING
HE CHEN

Assignees

西南大学

Dates

Publication Date: 20260505
Application Date: 20251216

Claims (6)

1. The intelligent vital sign monitoring method based on the migration learning mechanism and the UWB radar is characterized by comprising the following steps of: s1, signal acquisition and preprocessing, namely acquiring a channel impulse response sequence of an environment by using a UWB radar sensor, and preprocessing the channel impulse response sequence; S2, selecting a key distance gate, namely calculating an autocorrelation function of each distance gate signal in the preprocessed channel impulse response sequence, selecting a distance gate with the largest autocorrelation value, wherein the corresponding signal sequence is the most obvious vital sign information and is used as the input of subsequent processing; S3, extracting time-frequency characteristics, namely performing discrete wavelet transformation on the selected signal sequence, decomposing to obtain sub-band signals with different scales, and extracting characteristics capable of simultaneously representing time domain information and frequency domain information; s4, vital sign estimation based on transfer learning: a) Model pre-training, namely simulating the acquisition flow of S1-S3 through a physical simulation model, constructing simulation radar data, inputting the simulation radar data into a deep neural network model for pre-training, adjusting parameters in the deep neural network model to enable the deep neural network model to primarily have the capability of returning respiratory rate and heart rate from radar signals, B) The model fine tuning, namely, the actual hardware is deployed on the monitoring site, the real data of the monitoring site is acquired according to the acquisition steps of S1-S3 and is input into the pre-trained deep neural network model to carry out fine tuning on the model, so that the deep neural network model is quickly adapted to the channel characteristics of the new environment and the physiological characteristics of the current user, C) On-line reasoning, namely reasoning radar signal characteristics acquired in real time by using the finely tuned deep neural network model, and outputting vital sign estimation results in real time; And S5, multi-unit cooperation and decision fusion, namely, under a complex scene, deploying a plurality of UWB radar sensing nodes, independently processing data through each node, uploading vital sign estimation results and signal quality indexes to a central node or a cloud for data fusion and decision, and outputting final vital sign estimation results.
2. The intelligent vital sign monitoring method based on the transfer learning mechanism and the UWB radar as recited in claim 1, wherein the preprocessing in S1 comprises the specific steps of performing mean filtering and low-pass filtering on the original signal of the channel impulse response sequence.
3. The intelligent vital sign monitoring method based on the transfer learning mechanism and the UWB radar of claim 1, wherein the simulation radar data in S4 is simulation radar data with respiration, heartbeat simulation signals and random environmental noise.
4. The intelligent vital sign monitoring system based on the transfer learning mechanism and the UWB radar of claim 1, wherein the system comprises a data acquisition module, a signal processing module, a transfer learning reasoning module, a communication and control module and a power management module; The data acquisition module is composed of a UWB radar chip, an antenna array and a driving circuit and is mainly responsible for transmitting and acquiring UWB signals and acquiring channel impulse response sequences in the environment; The signal processing receives the acquired channel impulse response sequence, performs preprocessing on the channel impulse response sequence, calculates the preprocessed channel impulse response sequence, and selects a distance gate with the largest autocorrelation value and the most obvious vital sign information as input of subsequent processing; the migration learning reasoning module generates simulation radar data by constructing a physical simulation model, and then inputs the simulation radar data into a deep neural network model for pre-training; finally, reasoning radar signal characteristics acquired in real time by utilizing the depth neural network model after fine tuning, and outputting vital sign estimation results in real time; The communication and control module is responsible for control of a system, data exchange and communication with an upper computer or a cloud platform; the power management module provides a stable power supply for the system, ensuring that each module can maintain efficient operation during long-term operation.
5. The intelligent vital sign monitoring device based on the transfer learning mechanism and the UWB radar according to claim 1, wherein the intelligent vital sign monitoring device comprises a power supply and a monitoring device, a special chip is arranged in the monitoring device and is used for executing the monitoring method according to claims 1-3, the special chip is connected with the power supply, an external antenna and a sensor are further connected with the special chip, and the vital sign signal acquisition, processing and transmission are realized through the cooperative work of the special chip, the power supply, the external antenna and the sensor.
6. The intelligent vital sign monitoring device based on the transfer learning mechanism and the UWB radar of claim 1, wherein the special chip comprises a UWB radio frequency unit, a main control unit, an AI computing unit, a storage unit, a communication interface and a shell.

Description

Intelligent vital sign monitoring method, system and device based on migration learning mechanism and UWB radar Technical Field The invention relates to the technical field of radar signal processing and medical health monitoring, in particular to an intelligent vital sign monitoring method, system and device based on a migration learning mechanism and UWB radar. Background The non-contact vital sign monitoring technology becomes a research hot spot by virtue of the advantages of no interference to users, privacy protection and the like, wherein the ultra-wideband radar technology is regarded as an ideal sensing mode due to the characteristics of high time resolution, strong penetrating power, low energy consumption and the like. UWB radar obtains vital sign information by detecting electromagnetic wave phase change caused by chest and heart micro motion. However, the current UWB radar monitoring technology suffers from significant bottlenecks that firstly, data dependence and labeling cost are too high, a deep learning method is required to train by means of a large amount of accurate labeled real radar data, but acquisition of medical grade labeling data is not only high in cost and complicated in flow, so that data resources are deficient, secondly, environment and individual generalization capability are weak, performance of a model trained in a laboratory environment is greatly reduced when the model is deployed in a new scene or faces a new monitored object, self-adaptive adjustment capability is not required, thirdly, robustness in a complex environment is insufficient, radar signals are easily interfered by multipath reflection and other object micro-motion in practical application, and a traditional algorithm is difficult to stably extract weak vital sign signals. In the existing solutions, for example, although the signal processing method based on discrete wavelet transformation proposed by the university of Harbin industry improves the precision, a large amount of data still needs to be remarked under a new scene, and other methods based on traditional signal processing or single model have the problem of poor adaptability. The signal processing method based on discrete wavelet transformation, such as that proposed by Harbin university of industry, improves the accuracy, but needs a great amount of re-labeling data when facing a new scene. Other methods based on traditional signal processing or single models have the common problem of poor adaptability. To this end, the present invention aims to provide intelligent vital sign monitoring methods, systems and devices based on a transfer learning mechanism and UWB radar to solve the above-mentioned problems. Disclosure of Invention The invention aims to solve the problems, and provides an intelligent vital sign monitoring method, system and device based on a transfer learning mechanism and UWB radar, which utilize simulation data to complement the deficiency of real data, and enable a model to be quickly adapted to a new monitoring environment and a target individual through the transfer learning mechanism, so as to finally achieve non-contact vital sign monitoring with high accuracy, high stability and high practicability In order to achieve the above purpose, the technical scheme of the invention is as follows: The invention provides an intelligent vital sign monitoring method based on a migration learning mechanism and UWB radar, which comprises the following steps: S1, signal acquisition and preprocessing, namely acquiring a channel impulse response sequence of the environment by using a UWB radar sensor. The original signal is subjected to preprocessing such as mean filtering (DC offset removal) and low-pass filtering (high-frequency noise removal). S2, selecting a key range gate, namely calculating an autocorrelation function of each range gate signal after preprocessing, selecting a range gate with the largest autocorrelation value, and taking a corresponding signal sequence of the range gate as an input of subsequent processing, wherein the signal sequence is considered to contain the most remarkable vital sign information. S3, extracting time-frequency characteristics, namely performing discrete wavelet transformation on the selected signal sequence, decomposing to obtain sub-band signals with different scales (frequencies), and extracting characteristics capable of simultaneously representing time domain information and frequency domain information. S4, vital sign estimation based on transfer learning, wherein the step comprises two stages of offline training and online reasoning: Model pre-training (source domain) by simulating the acquisition process of S1-S3 by using a physical simulation model (such as electromagnetic simulation software and human chest motion model) to generate a large amount of simulation radar data with respiration and heartbeat simulation signals and various random environmental noises. A deep neural network (e.g.