CN-122004762-A - Infant state monitoring method and infant monitoring equipment based on machine learning

CN122004762ACN 122004762 ACN122004762 ACN 122004762ACN-122004762-A

Abstract

The invention discloses a baby state monitoring method and baby monitoring equipment based on machine learning, wherein the method comprises the steps of collecting multi-frame echo signals in a target period of a baby through a millimeter wave radar, extracting first characteristic sequence data based on the multi-frame echo signals, determining whether the baby is in an awake state in the target period based on the first characteristic sequence data, acquiring second characteristic sequence data based on the multi-frame echo signals when the baby is determined to be in the awake state, forming input of a pre-trained baby awake state monitoring model based on the second characteristic sequence data, outputting an awake state monitoring result, acquiring third characteristic sequence data based on the multi-frame echo signals when the baby is determined to be in a sleep state, forming input of a pre-trained baby sleep state monitoring model based on the third characteristic sequence data, and outputting the sleep state monitoring result.

Inventors

ZHANG ZHENHUAN
CHEN JUNBIN

Assignees

深圳市圣洲智能设备研发有限公司

Dates

Publication Date: 20260512
Application Date: 20260130

Claims (10)

1. The infant state monitoring method based on machine learning is characterized by being applied to infant monitoring equipment, wherein millimeter wave radar is arranged in the infant monitoring equipment, and the method comprises the following steps: acquiring multi-frame echo signals in the target time length of the baby through the millimeter wave radar; Extracting first characteristic sequence data based on the echo signals of a plurality of frames; determining whether the infant is awake for the target period of time based on the first characteristic sequence data; when the infant is determined to be in an awake state, acquiring second characteristic sequence data based on a plurality of frames of echo signals, forming the input of a pre-trained infant awake state monitoring model based on the second characteristic sequence data, and outputting an awake state monitoring result; and when the infant is determined to be in the sleep state, acquiring third characteristic sequence data based on the echo signals of a plurality of frames, forming the input of a pre-trained infant sleep state monitoring model based on the third characteristic sequence data and the characteristic sequence data, and outputting a sleep state monitoring result.
2. The machine learning based baby status monitoring method of claim 1 wherein the acquiring second signature sequence data based on a plurality of frames of the echo signals when the baby is determined to be in an awake state comprises: based on the multi-frame echo signals, performing phase extraction and separation to obtain vital sign sequence data, wherein the vital sign sequence data comprises respiratory sequence data and heart rate sequence data; Based on the multi-frame echo signals, local inching signal separation and recognition are carried out, and multiple local action characteristic sequence data are obtained, wherein the local action characteristic sequence data comprise head action sequence data, hand area action sequence data and leg area action sequence data.
3. The machine learning based infant status monitoring method of claim 2 wherein the forming an input to a pre-trained infant wakefulness status monitoring model based on the second signature sequence data, outputting a wakefulness status monitoring result comprises: forming the input of a physiological state identification network in the infant awake state monitoring model based on the vital sign sequence data and the multiple local action feature sequence data, and outputting physiological state sequence data; Acquiring historical monitoring data in a target period of time earlier than the target time length, wherein the historical monitoring data comprises nursing data and sleep data, and the nursing data comprises feeding data, diaper changing data and disease data; based on the vital sign sequence data, the plurality of local action feature sequence data, the physiological state sequence data and the history monitoring data, outputting an awake state category by using the infant awake state monitoring model.
4. The machine learning based infant status monitoring method of claim 3 wherein the outputting an awake status category using the infant awake status monitoring model based on the vital sign sequence data, a plurality of the local motion feature sequence data, the physiological status sequence data, and the historical monitoring data comprises: forming a context vector based on the historical monitoring data; Forming input of a one-dimensional convolution network and an attention mechanism network in the baby awake state monitoring model based on the multiple local motion characteristic sequence data, and outputting a local motion coding vector through the attention mechanism network; Forming an input of an LSTM network in the infant awake state monitoring model based on the vital sign sequence data and the physiological state sequence data, and outputting a vital sign coding vector through the LSTM network; Performing fusion splicing on the context vector, the local motion coding vector and the vital sign coding vector to obtain a fusion feature vector; And taking the fusion feature vector as the input of a classification network in the infant awake state monitoring model, and outputting the awake state type.
5. The machine learning based infant status monitoring method of claim 4 wherein the awake status categories comprise demand categories including hunger category, drowsiness category, discomfort category, the method further comprising at least one of: different prompt messages are given according to the identified demand category; When at least one of the following conditions is identified, the sleep state is changed to the awake state, the crying state and the crawling state, and the corresponding alarm information under each condition is output.
6. A machine learning based infant status monitoring method as in claim 5, wherein the method further comprises: Training the infant wakefulness monitoring model; said training said infant wakefulness monitoring model includes: Acquiring a first training data set, wherein each training sample in the first training data set comprises sample data and a sample label, wherein the sample data comprises vital sign sequence data corresponding to the training sample, corresponding multiple local action characteristic sequence data and corresponding historical monitoring data, and the sample label comprises a physiological state sequence label and a demand category label; The infant wakefulness monitoring model is trained based on the first training data set.
7. The machine learning based infant status monitoring method of claim 6 wherein the training the infant wakefulness status monitoring model based on the first training data set comprises: In the current iteration, predicting current physiological state sequence data corresponding to the training sample through the physiological state recognition network, and predicting a current demand category label corresponding to the training sample through the classification network; Determining a first loss value based on a difference between the current physiological state sequence data and the physiological state sequence tag; Determining a second loss value based on the current demand category and the demand category label; and calculating a current loss value based on the first loss value and the second loss value, and executing back propagation based on the current loss value until a termination condition is met, wherein the infant wakefulness monitoring model after the termination condition is met is used as a pre-trained infant wakefulness monitoring model.
8. The machine learning based infant status monitoring method of claim 1 wherein the sleep status monitoring results comprise durations of various types of sleep status within the target duration, wherein various types of sleep status comprise REM sleep, light sleep, deep sleep, the method further comprising at least one of: Determining and displaying sleep quality grades and sleep quality scores in the target duration according to the sleep state monitoring result; And displaying the duration of each sleep state and the duration proportion of each sleep state.
9. A machine learning based infant status monitoring method as in claim 8, wherein the method further comprises: Acquiring a second training data set, wherein each sampling sample in the second training data set comprises third characteristic sequence data corresponding to the sampling sample and a sleep state sequence label corresponding to the sampling sample; in the current iteration, outputting current sleep state data of the sampling sample through the infant sleep state monitoring model; calculating a sleep loss value based on the difference of the current sleep state data and the sleep state sequence tag; And based on the sleep loss value, executing back propagation until the iteration termination condition is met, and taking the infant sleep state monitoring model meeting the iteration termination condition as a pre-trained infant sleep state monitoring model.
10. An infant care apparatus comprising a memory and a processor, the memory storing a computer program which, when executed by the processor, causes the processor to perform the machine learning based infant condition monitoring method of any one of claims 1 to 9.

Description

Infant state monitoring method and infant monitoring equipment based on machine learning Technical Field The invention relates to the technical field of artificial intelligence, in particular to a machine learning-based infant state monitoring method and infant monitoring equipment. Background With the popularization of intelligent child care concepts and the rapid development of sensing technologies, infant monitoring has been brought into an intelligent and data new stage from traditional manual care and basic audio-visual monitoring. The core driving force of the infant care pillow is to meet two primary needs, namely, the infant care pillow can relieve anxiety and overdraft of vigor of new generation parents (especially dual-worker families) and scientifically prevent accidental risks possibly occurring in infant sleeping, such as apnea, sudden death syndrome and the like. The traditional camera and audio baby monitor realizes remote nursing, but is essentially only the extension transmission of audio-visual signals, and lacks active analysis and early warning capability. Disclosure of Invention In order to solve the existing technical problems, the invention provides a machine learning-based infant state monitoring method and infant monitoring equipment, which can realize intelligent monitoring of infants. The infant state monitoring method based on machine learning comprises the steps of collecting multi-frame echo signals in infant target time through a millimeter wave radar, extracting first characteristic sequence data based on the multi-frame echo signals, determining whether an infant is in an awake state in the target time based on the first characteristic sequence data, obtaining second characteristic sequence data based on the multi-frame echo signals when the infant is determined to be in the awake state, forming input of a pre-trained infant awake state monitoring model based on the second characteristic sequence data, outputting an awake state monitoring result, obtaining third characteristic sequence data based on the multi-frame echo signals when the infant is determined to be in a sleep state, forming input of the pre-trained infant sleep state monitoring model based on the third characteristic sequence data, and outputting the sleep state monitoring result. In a second aspect, there is provided an infant monitoring device comprising a memory and a processor, the memory storing a computer program which, when executed by the processor, causes the processor to perform the machine learning based infant status monitoring method as provided in the first aspect of the application. In the real-time monitoring, the state of the target duration is firstly judged, and then only the corresponding state monitoring model is operated, so that two complex models are prevented from being operated simultaneously, the consumption of computing resources can be reduced, and the response speed is improved. Through the infant awake state monitoring model and the infant sleep state monitoring model, the infant awake state monitoring model and the infant sleep state monitoring model are processed in stages, the state is firstly judged, then the infant awake state monitoring model and the infant sleep state monitoring model are respectively processed, each model only needs to pay attention to the mode under the specific state, the learning task of the model is simplified, and therefore the accuracy of overall monitoring is improved. Drawings FIG. 1 is a diagram showing an application environment of a machine learning-based infant status monitoring method according to an embodiment; FIG. 2 is a flow chart of a method for monitoring infant status based on machine learning according to an embodiment; FIG. 3 is a schematic diagram of a network structure of an infant wakefulness monitoring model according to an embodiment; FIG. 4 is a schematic diagram of an infant status monitoring device based on machine learning according to an embodiment; Fig. 5 is a schematic diagram of an infant monitoring device according to an embodiment. Detailed Description The technical scheme of the invention is further elaborated below by referring to the drawings in the specification and the specific embodiments. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the scope of the invention. The term "and/or" as used herein includes any and all combinations of one or more of the associated listed items. In the following description, reference is made to the expression "some embodiments" which describe a subset of all possible embodiments, but it should be understood that "some embodiments" may be the same subset or a different subset of all pos