An intelligent analysis system for infant needs

Abstract

The present invention relates to an intelligent analysis system for infant needs, comprising at least a first intelligent terminal, a second intelligent terminal, and a cloud server. The first intelligent terminal sends automatically collected physiological information, sound information, and / or behavioral information of an infant to the cloud server. The second intelligent terminal sends manually input life record information and / or actual emotional state information of the infant to the cloud server. The cloud server includes at least a behavioral feature extraction module and a deep learning training module. The feature extraction module extracts behavioral feature parameters of the infant based on a specific pattern established by the deep learning training module, and matches corresponding action training units, thinking training units, and emotional training units based on the behavioral feature parameters. The deep learning training module selects action training units, thinking training units, and / or emotional training units whose coordination is within a set threshold range based on the infant's specific pattern to form an infant training program. The present invention can accurately determine the impact of external changes on infants and improve the infant's living conditions.

Description

[0001] This invention is a divisional application of the invention patent with application number 201811306109.2, application date November 2, 2018, and invention name: An intelligent baby monitoring system. Technical Field

[0002] The present invention relates to the technical field of infant monitoring, and in particular to an intelligent analysis system for infant needs. Background Art

[0003] Currently, infants and young children in China primarily rely on formula milk in addition to breastfeeding. The amount of formula milk consumed directly determines the daily intake of various nutrients, including calories, protein, fat, carbohydrates, minerals, and vitamins. Infants and young children have relatively small reserves of nutrients and are less adaptable than adults. Insufficient intake of certain nutrients or digestive dysfunction can significantly impact developmental progress in a short period of time. Therefore, the amount and rhythm of formula milk intake directly impacts infant growth and development.

[0004] Because infants and young children of different ages have different needs for formula milk intake, and different brands of milk powder have different effects on a child's growth and development due to the different nutrients and trace elements they contain, choosing the right formula milk and measuring its intake is crucial. New parents often lack experience with daily milk intake and feeding schedules, and require guidance and assistance from professionals. When providing this guidance, professionals often need to fully understand the child's diet, sleep, and growth and development over a certain period of time to fully understand the child's individual differences and current situation and provide appropriate advice.

[0005] When recording their child's daily formula intake, parents often rely on the markings on the bottle. However, due to irregular shapes and inadequate quality control, some bottles have inaccurate markings, hindering parents' ability to accurately assess their child's formula intake. Furthermore, due to various practical constraints in daily life, it's difficult for parents to consistently and accurately record their child's milk intake, making it difficult to maintain a complete formula intake history. Furthermore, parents of newborns often forget when their child drinks, and keeping separate records is tedious, sometimes too busy to do so while caring for their child, or inconvenient when out and about.

[0006] Currently, there are some devices on the market for monitoring the care of infants. A Chinese patent (CN 105708468A) discloses a mother-child bracelet, which includes a child ring and a mother ring, wherein the child ring is used to detect the behavioral status information of the child ring wearer, generate abnormal prompt information based on the behavioral status information, and send the abnormal prompt information to the mother ring; the mother ring communicates with the child ring, and the mother ring prompts the mother ring wearer based on the abnormal prompt information. This patent allows the guardian to understand the abnormal situation of the cared person in a timely manner and better monitor the cared person. However, this patent is only used to monitor the abnormal state of the baby, and only informs the guardian at the mother ring end in advance when an abnormality occurs. It cannot record and analyze the baby's drinking time and bedwetting time. In particular, it is not possible to judge the baby's needs based on the sound of the baby's crying to facilitate the guardian's care for the baby.

[0007] Therefore, there is an urgent need in the market for an intelligent monitoring system that can both record infant care information and analyze infant needs. Summary of the Invention

[0008] In response to the deficiencies of the prior art, the present invention provides an intelligent infant monitoring system comprising at least a first intelligent terminal, a second intelligent terminal, and a cloud server. The first intelligent terminal is used to automatically collect physiological information, sound information, and / or behavioral information of an infant. The second intelligent terminal is used to manually input life record information and / or actual emotional state information of the infant. The cloud server comprises at least a teaching module and an emotional state analysis module that performs theoretical emotional state analysis based on the physiological information, sound information, and / or behavioral information of the infant using an emotional state analysis algorithm.

[0009] The cloud server also includes a behavior feature extraction module, which extracts the baby's behavior feature parameters based on the baby's physiological information, the sound information and / or the behavior information, and selects a matching physical exercise program based on the analysis of the behavior feature parameters.

[0010] According to a preferred embodiment, the cloud server also includes a deep learning training module, which learns and masters the specific patterns of infant limb behavior based on the infant's behavioral characteristic parameters, and the deep learning training module generates an infant training program based on the specific pattern that includes at least movement training, thinking training and / or emotion training.

[0011] According to a preferred embodiment, the feature extraction module extracts the behavioral characteristic parameters of the infant based on the specific pattern established by the deep learning training module, and matches the corresponding action training units, thinking training units and emotion training units based on the behavioral characteristic parameters. The deep learning training module selects the action training units, thinking training units and emotion training units whose coordination is within a set threshold range based on the infant's specific pattern to form an infant training plan.

[0012] According to a preferred embodiment, the teaching module completes the teaching process of the theoretical emotional state stored in the preset database based on the actual emotional state information input by the second intelligent terminal, and the emotional state analysis module instructs the second intelligent terminal to issue a demand prompt and / or warning prompt to the current user based on the demand information corresponding to the actual emotional state of the infant.

[0013] According to a preferred embodiment, the teaching module in the cloud server analyzes and completes the teaching process based on the life record information and / or actual emotional state information input by the second smart terminal, wherein the teaching module pre-configures the parameters of the state analysis algorithm based on at least two abnormal emotional state information.

[0014] According to a preferred embodiment, the emotional state analysis module stores the infant's life record information and / or actual emotional state information input by the user to the second smart terminal in the form of text, voice, video and / or graphics and the external conditions, physiological information, sound information and / or behavioral information automatically collected by the first smart terminal in an associated manner to a preset database, or the first smart terminal records the external conditions that trigger the infant's actual emotional state and stores or provides them to the preset database in a form associated with the corresponding actual emotional state, the emotional state analysis module performs analysis based on the correlation between the infant's specific emotional state and the external conditions, and based on the correlation, issues an early warning to the second smart terminal indicating the triggering of the specific emotional state.

[0015] According to a preferred embodiment, the cloud server also includes a correction module, which corrects the theoretical emotional state information determined based on the analysis of external conditions, physiological information, sound information and / or behavioral information collected by the first smart terminal based on the actual emotional state information of the infant, and forms a preset database composed of the corrected theoretical emotional state information that can be retrieved based on the infant's life record information, physiological information and / or external conditions.

[0016] According to a preferred embodiment, the cloud server is provided with a nursing advice module for associating medical information with a third-party medical institution. The nursing advice module retrieves corresponding medical care information based on the actual emotional state of the infant and issues a prompt message through the second smart terminal, and / or the nursing advice module prompts the infant disease warning information of the nearby area and / or specific time period issued by the third-party medical institution through the second smart terminal based on the geographical location determined by the second smart terminal, and prompts the infant disease rate evaluated based on the infant physiological information collected by the first smart terminal through the second smart terminal.

[0017] According to a preferred embodiment, the nursing suggestion module issues infant nursing suggestions through the second smart terminal based on infant disease warning information in nearby areas and / or specific time periods provided by a third-party medical institution.

[0018] According to a preferred embodiment, the cloud server also includes a baby profile that matches the baby's personality. The baby profile is created based on the baby's life record information and actual emotional state change trend information and adaptively adjusts the initial configuration information to form the baby's emotional parameters. The baby profile selects a corresponding parenting plan based on the analysis of the baby's emotional parameters and sends it to the second smart terminal; wherein, the initial configuration information includes at least the baby's date of birth, blood type and / or gender.

[0019] The present invention also provides an intelligent analysis system for infant needs, which includes at least a first intelligent terminal, a second intelligent terminal and a cloud server. The first intelligent terminal sends automatically collected physiological information, sound information and / or behavioral information of the infant to the cloud server, and the second intelligent terminal sends manually input life record information and / or actual emotional state information of the infant to the cloud server. The cloud server includes at least a behavioral feature extraction module and a deep learning training module. The feature extraction module extracts the behavioral feature parameters of the infant based on the specific pattern established by the deep learning training module, and matches the corresponding action training unit, thinking training unit and emotion training unit based on the behavioral feature parameters. The deep learning training module selects the action training unit, thinking training unit and / or emotion training unit whose coordination is within a set threshold range based on the specific pattern of the infant to form an infant training plan.

[0020] Preferably, the feature extraction module extracts behavioral feature parameters of the infant based on the physiological information, the sound information and / or the behavioral information of the infant, and selects a matching physical exercise program based on analysis of the behavioral feature parameters.

[0021] Preferably, the feature extraction module extracts behavioral feature parameters based on the CNN model, and at least one behavioral feature parameter corresponds to a specific pattern of the infant; the feature extraction module uses cosine similarity to match the behavioral feature parameters with the behavioral model, and uses at least one action training unit, at least one thinking training unit and / or at least one emotion training unit corresponding to the maximum similarity value as a component element of the training program.

[0022] Preferably, the feature extraction module includes a matching recognition module, which sends the behavioral feature parameters to the matching recognition module. The matching recognition module uses cosine similarity to compare the behavioral feature parameters with the behavioral models in the database and outputs a set of similarity values, sorts the similarity values ​​from large to small, and outputs the physical exercise plan corresponding to the maximum similarity value in the database.

[0023] Preferably, the matching identification module in the feature extraction module matches the physiological information corresponding to the action training unit with the physiological information corresponding to the thinking training unit. If the matching degree is between 60% and 80%, the action training unit and the thinking training unit are coordinated. If the matching degree is less than 60%, the action training unit and the thinking training unit are not coordinated.

[0024] Preferably, the matching recognition module in the feature extraction module matches the physiological information corresponding to the action training unit with the physiological information corresponding to the emotion training unit, and the matching degree of the physiological information corresponding to the thinking training unit with the physiological information corresponding to the emotion training unit is within the range of a preset threshold.

[0025] Preferably, the deep learning training module uses the CNN algorithm to train to obtain a CNN model; the deep learning training module transfers the trained CNN model to the feature extraction module; the feature extraction module uses the CNN model to extract the physiological information, the sound information and / or the behavioral information of the infant to obtain a feature vector or a behavioral model.

[0026] Preferably, the cloud server also includes a teaching module, which completes the teaching process of the theoretical emotional state stored in the preset database based on the actual emotional state information input by the second intelligent terminal. The cloud server also includes an emotional state analysis module, which uses an emotional state analysis algorithm to perform theoretical emotional state analysis based on the infant's physiological information, sound information and / or behavioral information. The emotional state analysis module instructs the second intelligent terminal to issue demand prompts and / or warning prompts to the current user based on the demand information corresponding to the actual emotional state of the infant.

[0027] Preferably, the teaching module in the cloud server analyzes and completes the teaching process based on the life record information and / or actual emotional state information input by the second smart terminal, wherein the teaching module pre-configures the parameters of the state analysis algorithm based on at least two types of abnormal emotional state information.

[0028] Preferably, the emotional state analysis module stores the infant's life record information and / or actual emotional state information input by the user to the second smart terminal in the form of text, voice, video and / or graphics and the external conditions, physiological information, sound information and / or behavioral information automatically collected by the first smart terminal in an associated manner into a preset database.

[0029] Beneficial technical effects of the present invention:

[0030] (1) The present invention can accurately analyze the baby's emotional expression and needs by monitoring the baby's physiological information, voice and behavioral information, which is conducive to the guardian's timely care of the baby;

[0031] (2) Different infants have different ways of expressing their emotions. The present invention can form an analysis process that conforms to the individual personality of each infant through a teaching process, thereby more accurately analyzing the specific emotions of each infant.

[0032] (3) The present invention uses external conditions as analysis factors and can accurately determine the impact of external changes on infants, so that guardians can promptly improve the infant's living conditions in response to changes in external conditions and prevent the infant from crying due to external conditions;

[0033] (4) The present invention provides reasonable care suggestions to guardians based on the medical information of infants in nearby areas, preventing guardians from taking their infants to areas with high incidence of diseases, thereby ensuring the health of the infants. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 is a schematic diagram of a logic module of the present invention; and

[0035] Figure 2 It is a schematic diagram of the method steps of the emotional state analysis module of the present invention.

[0036] Reference Signs List

[0037] 10: First smart terminal 20: Second smart terminal

[0038] 30: Cloud Server 31: Database

[0039] 32: Teaching module 33: Emotional state analysis module

[0040] 34: Correction module 331: Sound analysis module

[0041] 332: Sentiment Analysis Module DETAILED DESCRIPTION

[0042] The following is a detailed description with reference to the accompanying drawings.

[0043] The present invention provides an intelligent baby monitoring system, which comprises at least a first intelligent terminal 10, a second intelligent terminal 20 and a cloud server 30. Figure 1 As shown, the first smart terminal 10 and the second smart terminal 20 are connected to the cloud server 30 via a wired or wireless manner. The first smart terminal 10 and the second smart terminal 20 are connected via a wired or wireless manner.

[0044] The first smart terminal 10 is set on the baby's body. The first smart terminal is a wearable device, such as a smart ring, smart clothing and smart sticker. The first smart terminal is used to automatically collect the baby's physiological information, sound information and / or behavioral information. The baby's physiological information includes information such as body temperature, pulse, heartbeat, sweat humidity and so on. Sweat humidity refers to the degree of sweating of the baby. Sound information includes the rhythm, frequency, brightness and / or loudness of the sound made by the baby, specifically including the rhythm, frequency, brightness and / or loudness of laughter, and the rhythm, frequency, brightness and / or loudness of crying. The baby's behavioral information includes at least behavioral information such as the arm swing frequency and swing amplitude, body twisting, and leg swing frequency and swing amplitude.

[0045] Preferably, the first smart terminal 10 also collects information about the external conditions of the baby's environment. The external condition information includes at least information such as external temperature, humidity, noise, oxygen content, light brightness, and weather conditions.

[0046] The second intelligent terminal 20 is used to manually input the baby's life record information and / or actual emotional state information. The life record information at least includes the baby's feeding time, milk consumption, diaper changing time, defecation time, awake time and / or sleeping time.

[0047] The actual emotional state information is the actual emotional state of the infant. Due to differences in individual personalities, even with the same behavioral movements and crying rhythm, the emotions expressed can vary. Therefore, in the early stages of using the system of the present invention, a teaching process is required to correct the error between the theoretical emotional state and the actual emotional state in the system. The guardian is the most direct observer of the infant's actual emotional state and can manually input the infant's actual emotional state to reflect the actual situation. Alternatively, the guardian can determine the actual emotional state expressed by the infant's crying based on the infant's response to feeding, diaper changing, etc. Through multiple teaching processes, the parameters of the emotional state analysis algorithm within the system will be corrected, forming a monitoring system that matches the infant's personality and provides accurate analysis. Therefore, the infant's actual emotional state information needs to be manually input.

[0048] Preferably, the actual emotional state information of the infant is different from the actual emotional state information of the adult.

[0049] Information about an infant's actual emotional state is relatively simple, encompassing not only real emotions like happiness, discomfort, irritability, sleepiness, fear, and nervousness, but also needs-based emotions like hunger, thirst, pain, diaper discomfort, overfullness, defecation, and requests for a child to be held. Therefore, analyzing an infant's emotional state helps guardians take appropriate care measures, meeting the infant's needs while also avoiding the fatigue of guessing what the infant wants.

[0050] Preferably, the cloud server 30 includes at least a teaching module 32 and an emotional state analysis module 33 that analyzes a theoretical emotional state using an emotional state analysis algorithm based on the infant's physiological information, sound information, and / or behavioral information. The teaching module completes the teaching process of the theoretical emotional state stored in the preset database 31 based on the actual emotional state information input by the second intelligent terminal.

[0051] The teaching process in the present invention refers to correcting the parameters in the emotional state analysis algorithm based on the deviation between the actual emotional state and the theoretical emotional state obtained by calculation and analysis, so that the theoretical emotional state obtained by calculation and analysis is consistent with the actual emotional state of the infant.

[0052] Preferably, the emotional state analysis module 33 instructs the second intelligent terminal to issue a demand prompt and / or a warning prompt to the current user based on the demand information corresponding to the actual emotional state of the infant.

[0053] After the emotional state analysis module 33 obtains the actual emotional state of the baby, it will send the baby's needs or corresponding measures to the guardian's second smart terminal. The second smart terminal will issue a demand prompt or warning prompt to the current user.

[0054] Preferably, the emotional state analysis module analyzes the changing trend of the infant's actual emotional state based on the change, and when the changing trend of the actual emotional state exceeds a critical value at a foreseeable time point, instructs the second smart terminal to issue an early warning prompt to the guardian.

[0055] Preferably, the teaching module in the cloud server analyzes and completes the teaching process based on the life record information and / or actual emotional state information input by the second intelligent terminal. Teaching in the present invention refers to exemplary artificial intelligence programming. The teaching module pre-configures the parameters of the emotional state analysis algorithm based on at least two types of abnormal emotional state information. Preferably, the emotional state analysis algorithm includes machine learning algorithms such as Bayesian classification algorithms, neural networks, support vector machines, decision trees, case-based reasoning learning, and association rule learning.

[0056] Preferably, the emotional state analysis module includes at least a sound analysis module 331 and an emotion analysis module 332. The sound analysis module 331 performs a first emotional state analysis based on the rhythm, frequency, brightness, and / or loudness of the infant's voice. The emotion analysis module 332 performs an analysis based on physiological information, behavioral information, life record information, and / or the first analysis information to obtain a second emotional state analysis information.

[0057] For example, the emotional state analysis algorithm includes the emotional state analysis of a crying baby. One of the emotional state analysis algorithms pre-stored in the preset database and its teaching process are shown below.

[0058] S1: Obtain multiple training data segments of baby crying audio, each training data segment corresponds to two known crying causes;

[0059] S2: performing feature extraction on each segment of training data to obtain a feature parameter vector for each segment of training data; preferably, after feature extraction, performing noise reduction on the crying signal in the training data to detect and remove data segments with noise greater than a predetermined threshold;

[0060] S3: Perform principal component analysis on the characteristic parameter vectors of multiple training data segments to obtain multiple principal components;

[0061] S4: Calculate the mean and variance of the projected scores of the training data corresponding to each crying reason on each principal component, and select P principal components from the plurality of principal components based on the variance, where P is an integer greater than 1;

[0062] S5: Obtain the data to be identified of the baby crying audio, and calculate the projection scores of the data to be identified on P principal components;

[0063] S6: Calculating the probability that the data to be identified corresponds to each theoretical emotional state according to the projection score, mean and variance of the data to be identified;

[0064] S7: Correct the projection score based on the infant's actual emotional state as reported by the second intelligent terminal. If the actual emotional state reported by the second intelligent terminal matches one of the theoretical emotional states, the forward projection score is corrected to further increase the probability of the theoretical emotional state. After repeated teaching, the theoretical emotional state will become more consistent with the actual emotional state.

[0065] Preferably, the multiple training data include N crying signal samples, and K feature parameters are extracted from the N crying signal samples respectively, wherein the K feature parameters extracted from the nth crying signal are recorded as the feature parameter vector Sn=[Sn l ,Sn2,…,sn k ] TFor N crying signal samples, calculate the covariance matrix corresponding to K characteristic parameters, denoted as C. Where C is a K-by-K matrix. Perform eigenvalue decomposition on the covariance matrix to obtain K eigenvalues ​​and eigenvectors corresponding to the K eigenvalues.

[0066] Perform eigenvalue decomposition on the covariance matrix C and arrange the eigenvalues ​​from large to small to obtain {λ1,λ2,…,λ k}, take the Q eigenvalues ​​with the largest eigenvalues ​​and their corresponding eigenvectors to form two Q-dimensional feature subspaces, where the value of Q is The solution of g is any preset value between 0.9 and 0.99. Among these Q principal components, take the kth principal component, and the eigenvalue of this principal component is recorded as λ k ,λ k The corresponding eigenvector is denoted as u k , calculate the characteristic parameter vector S of the nth crying signal n The projection score on the kth eigenvector. Find the crying signal that belongs to the jth type of crying reason among the N crying signals, denoted as N j , then the mean of the projected scores of the K feature parameter vectors on the kth feature vector is and variance σ jk , then calculate

[0067] in,

[0068] In the above formulas, J represents the total number of types of crying reasons, X k Y represents the separation of the projection scores of the characteristic parameter vector of the crying signal on the kth principal component. k L represents the concentration of the projection score of the characteristic parameter vector of the crying signal on the kth principal component. k Represents the ability of each principal component to discriminate the cause of crying. k The larger the value, the stronger its discrimination ability.

[0069] Arrange the Q principal components in order and select L k The P principal components with the largest values ​​are used for subsequent crying cause identification, where P takes the smaller value of Q and M.

[0070]

[0071] h is a preset value between 2% and 0.5%.

[0072] In this emotional state analysis algorithm, the probability of the jth cause is

[0073]

[0074] Based on the obtained P principal components, steps S5 and S6 are performed for each training data point in the training data set to determine the probability of each cause associated with the training data point. The most probable cause associated with the training data point is then calculated and compared with the known crying cause associated with the training data point. Training data points with a most probable cause that differs from a known crying cause are removed from the two selected training data sets. The remaining training data points are used as the newly selected training data set and the above steps are repeated. This cycle continues until a predetermined exit condition is met.

[0075] Preferably, in the present emotional state analysis algorithm, the extracted features include any two or more of the following features, such as: average crying duration, crying duration variance, average crying energy, crying energy variance, fundamental frequency, average value of fundamental frequency, maximum value of fundamental frequency, minimum value of fundamental frequency, dynamic range of fundamental frequency, average rate of change of fundamental frequency, second formant frequency, average rate of change of second formant frequency, average value of first formant frequency, maximum value of first formant frequency, minimum value of first formant frequency, dynamic range of first formant frequency, second formant frequency, average rate of change of second formant frequency, average value of second formant frequency, maximum value of second formant frequency, minimum value of second formant frequency, dynamic range of second formant frequency, Mel frequency cepstrum parameter, and flipped Mel frequency cepstrum parameter.

[0076] Preferably, the emotional state analysis module determines that the baby's crying is due to hunger based on the baby's most recent feeding time, the frequency of the baby's limbs swinging, and the baby's voice characteristics.

[0077] Preferably, the emotional state analysis module analyzes the changing trend of the actual emotional state based on the infant's physiological information, sound information, and behavioral information, and determines the probability of the infant crying due to hunger. When the probability of crying due to hunger exceeds a preset threshold, the module instructs the second intelligent terminal to issue a feeding reminder message.

[0078] Preferably, the emotional state analysis module stores the infant's life record information and / or actual emotional state information input by the user to the second smart terminal in the form of text, voice, video and / or graphics and the external conditions, physiological information, sound information and / or behavioral information automatically collected by the first smart terminal in an associated manner into a preset database.

[0079] Preferably, the step in which the user inputs life record information or actual emotional state information on the second smart terminal includes: the user selects the current baby's life record information, emotional type and level by clicking, and / or the user inputs the baby's life record information and / or actual emotional state in the form of text, voice, video or graphics.

[0080] Specifically, the second intelligent terminal inputs the infant's life record information and / or actual emotional state information in the form of text, voice, video and / or graphics. For example, the second intelligent terminal is provided with a selection key that displays text and / or graphics. The guardian only needs to trigger the selection key to manually input the corresponding information. Alternatively, the second intelligent terminal is provided with a microphone and camera connected to the voice recognition module. The guardian records the infant's life information by voice input or by taking a video and inputting it.

[0081] For example, the second smart terminal is equipped with a touch screen that displays modules with various functions. The guardian can select the current baby's life record information, emotional type, and level by clicking. The guardian can also record the baby's life information and their own feelings by editing text messages, inputting voice records, taking pictures, or recording videos.

[0082] The first intelligent terminal and the second intelligent terminal respectively send the collected information to the cloud server. Preferably, the first intelligent terminal and the second intelligent terminal are respectively provided with a flash memory. When connected to the cloud server, the first intelligent terminal and the second intelligent terminal send the collected data information to the preset database of the cloud server in real time. In the case of an obstacle in connecting to the cloud server, the first intelligent terminal and the second intelligent terminal store the collected data information in the flash memory in the form of a flash memory. In the case of a good subsequent connection with the cloud server, the first intelligent terminal and the second intelligent terminal then send the flash memory data information to the preset database of the cloud server. Such a setting can ensure that the data of this system will not be lost due to signal transmission obstacles, and ensure the authenticity and validity of the data.

[0083] Preferably, the first intelligent terminal records the external conditions that trigger the infant's actual emotional state and stores or provides them to a preset database in a form associated with the corresponding actual emotional state. The emotional state analysis module analyzes the correlation between the infant's specific emotional state and the external conditions, and based on the correlation, issues an early warning to the second intelligent terminal if the specific emotional state is triggered.

[0084] For example, for sensitive babies, when they encounter external conditions they don't like, they will cry or resist behaviorally, thereby triggering actual emotional information such as crying. Therefore, associating external condition information with actual emotional state information can effectively enable guardians to change bad external conditions, or take the baby away from bad external conditions, so that the baby stops crying and has a comfortable experience. When the correlation between the baby's specific emotional state and a certain external condition, such as noise intensity, is very high, the noise intensity value is listed as a warning external condition. When the emotional state analysis module detects that the noise intensity value collected by the first intelligent terminal is within the warning external condition range, it prompts the second intelligent terminal to change the noise environment in which the baby is located, or take the baby away from the noise environment.

[0085] Preferably, the second smart terminal is configured to allow the user to retrieve the actual emotional state information stored in the preset database and / or the second smart terminal in a manner related to the baby's life record information and / or external conditions.

[0086] For example, the second smart terminal is provided with a retrieval function. It is used for the guardian to retrieve the actual emotional state information stored in the preset database and / or the second smart terminal in a manner related to the baby's life record information and / or external conditions. The guardian enters the date or care behavior of the life record information of a certain day, and the life record of the baby on that day can be retrieved. Alternatively, the guardian enters external condition information, such as the outside temperature, and all the actual emotional state information of the baby at that temperature can be retrieved. The setting of the retrieval function makes it easy for the guardian to retrieve the baby's life record information and the actual emotional change information of a certain time period and display it to medical staff or third-party institutions, which is conducive to the study of the baby's condition, and is also conducive to the guardian's further understanding of the baby's living habits, and to avoid adverse events such as the baby developing allergies in the outside world.

[0087] Preferably, the cloud server further includes a correction module 34. The correction module can be one or more of a data analysis module, a data verification module, and a server.

[0088] The correction module corrects the theoretical emotional state information determined based on analysis of external conditions, physiological information, sound information, and / or behavioral information collected by the first intelligent terminal based on the infant's actual emotional state information. The corrected theoretical emotional state information is used to construct a preset database that can be searched based on the infant's life records, physiological information, and / or external conditions.

[0089] Although the present invention incorporates a teaching module to ensure that theoretical emotional state information matches actual emotional state information, a guardian's lack of experience in caring for infants can lead to incorrect judgments about the infant's actual emotional state information. This can lead to incorrect input of actual emotional state information via the second intelligent terminal, further misadjusting the parameters of the emotional state analysis algorithm. Therefore, the provision of a correction module is also crucial. Alternatively, the infant's behavioral information collected by the first intelligent terminal may contain errors, resulting in large errors in the recorded behavioral information.

[0090] For example, the correction module uses the actual emotional state to calibrate the theoretical emotional state information corresponding to the infant's physiological information, vocal information, and behavioral information. When the difference between the theoretical emotional state analyzed by the emotional state analysis module and the actual emotional state gradually increases, it indicates that the infant has experienced changes in expression due to learning and growth, with behavioral changes being the most significant. For example, early infants primarily express their desire for a hug by crying, with no clear pattern in their arm or leg movements. At five months, infants begin to express their desire for a hug more frequently by swinging their arms. If the guardian detects a discrepancy between the theoretical emotional state information received by the second intelligent terminal and the infant's actual emotional state, the system can be triggered to perform a correction by repeatedly triggering the correction module and the actual emotional state information. In this case, the correction module will correct the theoretical emotional state information corresponding to the behavioral information based on the actual emotional state. Because infants' lives are relatively simple, they experience repeated emotional states multiple times a day. Therefore, repeated corrections can correct any analysis deviations caused by growth.

[0091] Preferably, the cloud server is provided with a nursing suggestion module for associating medical information with a third-party medical institution. The nursing suggestion module retrieves corresponding medical care information based on the actual emotional state of the infant and issues a prompt message through the second smart terminal.

[0092] And / or, the nursing advice module prompts through the second smart terminal the infant disease warning information in the nearby area and / or specific time period issued by the third-party medical institution based on the geographic location determined by the second smart terminal, and prompts through the second smart terminal the infant disease rate evaluated based on the infant physiological information collected by the first smart terminal.

[0093] For example, if the emotional state analysis module determines that the infant is distressed and has an abnormal body temperature, it will send this information to the care advice module. Based on the infant's actual emotional state, the care advice module retrieves corresponding medical care information and issues a prompt via the second intelligent terminal, prompting the guardian to provide scientific care for the infant and preventing the guardian from making incorrect care decisions due to panic or lack of professional knowledge.

[0094] For example, the haze weather worsens in spring, and the incidence of pneumonia in the community where the baby lives increases. The first smart terminal or the second smart terminal has the function of collecting geographic location information. Based on the geographic location determined by the first smart terminal and / or the second smart terminal, the nursing advice module retrieves the infant morbidity rate in nearby areas and / or specific time periods published by third-party medical institutions, and finds that the incidence of pneumonia is increasing. The nursing advice module prompts infant disease warning information such as pneumonia incidence through the second smart terminal. At the same time, the nursing advice module will independently evaluate the pneumonia prevalence based on the physiological information of the baby in the recent period, or send the physiological information of the baby in the recent period to a third-party medical institution for professional medical personnel to conduct a disease assessment. The infant morbidity rate of the nursing advice module is prompted through the second smart terminal to remind guardians to pay attention to the baby's recent life care to prevent the baby from contracting pneumonia. In particular, for infant diseases with infectious properties, the nursing advice module is particularly important. It can notify guardians in the first time during the infectious risk period to protect the baby, stay away from infectious disease areas or isolate infectious conditions.

[0095] Preferably, the nursing suggestion module issues infant nursing suggestions through the second smart terminal based on infant disease warning information in nearby areas and / or specific time periods provided by a third-party medical institution.

[0096] For example, the care advice module, based on infectious diseases in the infant's community provided by a third-party medical institution, sends care recommendations to the second smart terminal, such as frequent changes of clothing, reduced outings, and increased water feedings. This allows guardians to scientifically improve their infant's immunity, avoiding panic caused by infectious diseases and unnecessary care behaviors due to lack of professional knowledge, which can increase discomfort for the infant. These care recommendations enable guardians to care for their infants in an organized and scientific manner, protecting them from disease.

[0097] Preferably, the cloud server also includes an infant profile. An infant's personality is difficult to accurately determine in the early stages of development. It's difficult for guardians to confirm an infant's personality through observation, and therefore, they can't adopt appropriate parenting methods to support the infant's intellectual development. The earlier an infant's personality is identified, the more appropriate parenting methods can be used, which will aid the infant's intellectual development.

[0098] The infant profile of the present invention is individually configured for each infant. When the infant profile is created, it includes initial configuration information such as the infant's date of birth, blood type, and gender. Preferably, the initial configuration information also includes initial infant emotional parameters analyzed based on the date of birth, blood type, and gender. When the monitoring system of the present invention is in operation, the infant profile extracts the infant's life record information and actual emotional state information and its changing trends from a pre-stored database to adaptively adjust the initial infant emotional parameters in the infant profile to form the infant emotional parameters. The infant profile selects a more suitable parenting method based on the analysis of the infant's emotional parameters.

[0099] For example, if a baby develops faster than other children of the same age, learning to call out and babble in the second month and being able to make different sounds and express different emotions towards their loved ones, the baby profile will send corresponding parenting plans to the second smart terminal to promote interaction between different guardians and babies. For example, if the baby profile analyzes the baby's emotional parameters and finds a high frequency of anger and that the baby is irritable but has a clear response to music, the baby profile will send a more gentle parenting plan to the second smart terminal, including listening to soothing songs frequently to ease the baby's emotions, reduce the baby's behavior of deliberately crying due to anger, and reduce the guardian's fatigue.

[0100] Preferably, the first smart terminal includes at least one or more of a physiological information sensor, a sound collector, an acceleration sensor, a balance sensor, and a flash memory storage device. For example, the first smart terminal is a smart bracelet.

[0101] The second smart terminal includes at least one or more of an information input device, a display device, a flash memory storage device, and an analysis module. The second smart terminal can be a smart bracelet or a smart device, such as a smartphone, a smart bracelet, smart glasses, a notebook, or a computer.

[0102] Preferably, the system of the present invention may further include a first smart terminal, a second smart terminal, and a third smart terminal. The first smart terminal and the second smart terminal are both portable, lightweight smart bracelets that are easy to move, and the third smart terminal is a smartphone or computer. The second smart terminal is used to receive, display, and input the baby's life record information. The third smart terminal can perform all the functions of the second smart terminal while retrieving all information records about the baby, including the baby's actual emotional state information and its changing trends, the probability of illness, and nursing recommendations. It can even directly connect with third-party medical institutions to obtain the most direct and effective medical advice.

[0103] Example 2

[0104] This embodiment is a further improvement of embodiment 1, and repeated contents will not be repeated here.

[0105] The intelligent infant monitoring system of the present invention comprises at least a first intelligent terminal 10, a second intelligent terminal 20 and a cloud server 30. The cloud server of the present invention further comprises a behavior feature extraction module.

[0106] Preferably, the behavior feature extraction module extracts behavior feature parameters from the infant's physiological information, the sound information, and / or the behavior information based on a preset behavior model, such as the frequency of the infant's arm swinging, the frequency of the infant's leg shaking, the acceleration of the infant's kicking, heart rate parameters, and body temperature parameters when the infant laughs loudly.

[0107] Preferably, the behavior feature extraction module includes a matching recognition module. The behavior feature extraction module sends the behavior feature parameters to the matching recognition module. The matching recognition module analyzes the behavior feature parameters. Preferably, the matching recognition module compares the behavior feature parameters with the behavior models in a database using cosine similarity and outputs a set of similarity values. The similarity values ​​are sorted from largest to smallest, and the physical exercise program corresponding to the highest similarity in the database is output.

[0108] Preferably, since there are thousands of individual personalities of infants, the preset behavior model may not be able to fully match and extract the behavioral characteristics of infants. Therefore, it is particularly important to be able to establish a behavior model based on deep learning of infants' behaviors. Preferably, the cloud server also includes a deep learning training module. The deep learning training module uses a CNN algorithm to train a CNN model. The CNN model includes an input layer, a convolutional layer, a fully connected layer, and an output layer; a hidden layer is provided between the input layer and the convolutional layer. A hidden layer is provided between the convolutional layer and the fully connected layer. The deep learning training module transfers the trained CNN model to the feature extraction module.

[0109] The feature extraction module uses a CNN model to extract the infant's physiological information, acoustic information, and / or behavioral information to obtain a feature vector or a behavior model. The feature extraction module transmits the feature vector to the matching recognition module. The feature extraction module transmits the behavior model to a behavior model library in a database.

[0110] Preferably, the process of extracting and processing the feature vector or behavior model by the CNN model is as follows:

[0111] A. Inputting the infant's physiological information, the sound information, and / or the behavioral information into a cloud server through a first intelligent terminal and storing the information in an input layer of a CNN model;

[0112] B. Physiological information, sound information and / or behavioral information are convolved on the volume base layer to extract feature vectors; the calculation formula on the volume base layer is: conv = σ(imgMat°W+b).

[0113] Among them, conv represents the input parameters of the convolution layer, σ represents the activation function ReLU, imgMat represents the grayscale image matrix, W represents the convolution kernel, represents the convolution operation, and b represents the bias value.

[0114] C. The input layer is forward propagated to the convolutional layer. The forward propagation process is expressed as:

[0115] a2=σ(z2)=σ(a1*W2+b2).

[0116] Among them, a2 is the input parameter of the convolution layer, the subscript represents the number of layers, the asterisk represents convolution, b represents bias, and σ is the activation function ReLU.

[0117] D. The hidden layer is forward propagated to the convolutional layer. The forward propagation process is expressed as:

[0118] a1=σ(z1)=σ(a1-1*W1+b1)

[0119] Among them, a l is the input parameter of the convolution layer, the subscript represents the number of layers, the asterisk represents convolution, b represents bias, and σ is the activation function ReLU.

[0120] E. The hidden layer is forward propagated to the fully connected layer. The forward propagation process is:

[0121] A1=σ(z1)=σ(W1a1-1+b1).

[0122] Among them, A l is the input parameter of the fully connected layer, the subscript represents the number of layers, the asterisk represents convolution, b represents bias, and σ is the activation function ReLU.

[0123] Preferably, the deep learning training module learns and masters the specific pattern of the baby's limb behavior based on the baby's behavioral characteristic parameters. The specific pattern includes at least different emotional patterns. For similar frequencies of crying, arm swinging frequency and leg swinging frequency, the emotions expressed by each baby are different. Some express emotions as hunger, some express emotions as discomfort, and some express emotions as wanting to be hugged. Therefore, the deep learning training module learns and masters the specific pattern of the baby's limb behavior based on the baby's behavioral characteristic parameters, and can more personally know the baby's specific emotional expression pattern. The deep learning training module generates an infant training program that at least includes action training, thinking training and / or emotional training based on the specific pattern.

[0124] Preferably, the feature extraction module extracts the infant's behavioral characteristic parameters based on the behavioral model established by the deep learning training module corresponding to the specific pattern, and matches the corresponding action training unit, thinking training unit, and emotion training unit based on the behavioral characteristic parameters. The deep learning training module selects the action training unit, thinking training unit, and emotion training unit whose coordination is within a set threshold range based on the infant's specific pattern to form an infant training plan.

[0125] For example, the feature extraction module extracts behavioral feature parameters based on the CNN model. And at least one behavioral feature parameter corresponds to the specific pattern of the baby. The feature extraction module uses cosine similarity to match the behavioral feature parameters with the behavioral model, and uses at least one action training unit, at least one thinking training unit and / or at least one emotion training unit corresponding to the maximum similarity value as the training program constituent elements. Although the action training unit, the thinking training unit and the emotion training unit are extracted based on the behavioral feature parameters, it is also crucial to conform to the coordination of infant human body science. If the action training unit and the thinking training unit are contradictory, the baby will feel uncomfortable and have the opposite effect. Preferably, the coordination value between each action training unit, between the thinking training unit and / or between the emotion training unit is calculated.

[0126] For example, the physiological information corresponding to the action training unit is matched with the physiological information corresponding to the thinking training unit. If the matching degree is between 60% and 80%, the action training unit is coordinated with the thinking training unit. If the matching degree is less than 60%, the action training unit is not coordinated with the thinking training unit. For example, the action training unit slows down the heart rate and breathing rate, and the thinking training unit requires a lot of oxygen and increases the breathing rate. If the matching degree is less than 60%, the baby's body will feel very uncomfortable and the effect of thinking training will not be achieved. If the matching degree is greater than 80%, the baby's body will be damaged due to the excessively fast heart rate.

[0127] Similarly, the physiological information corresponding to the action training unit is matched with the physiological information corresponding to the emotion training unit, and the physiological information corresponding to the thinking training unit is matched with the physiological information corresponding to the emotion training unit. Both must meet the preset threshold range. Preferably, the preset threshold range is not limited to between 60% and 80%, and can also be pre-set according to specific action, emotion, and thinking training.

[0128] That is, the cloud server's deep learning training module and behavioral feature extraction module further assist in personalized monitoring and training of infants, helping them grow healthily, develop good thinking habits and character, and effectively control their emotions.

[0129] It should be noted that the above-described specific embodiments are illustrative only. Those skilled in the art may devise various solutions based on the disclosure of the present invention, and such solutions fall within the scope of the present invention and are intended to be protected by the present invention. Those skilled in the art should understand that the present description and its accompanying drawings are intended to be illustrative only and are not intended to limit the scope of the claims. The scope of protection of the present invention is defined by the claims and their equivalents.

Claims

1. An intelligent analysis system for infant needs, comprising at least a first intelligent terminal (10), a second intelligent terminal (20) and a cloud server (30), characterized in that: The first intelligent terminal (10) sends automatically collected physiological information, sound information and / or behavioral information of the baby to the cloud server (30), The second intelligent terminal (20) sends the manually input life record information and / or actual emotional state information of the baby to the cloud server (30), The cloud server (30) includes an emotional state analysis module (33) for performing theoretical emotional state analysis using an emotional state analysis algorithm based on the physiological information, the sound information and / or the behavioral information of the infant. The cloud server (30) further comprises at least a behavior feature extraction module and a deep learning training module, wherein the behavior feature extraction module extracts the behavior feature parameters of the infant based on a specific pattern established by the deep learning training module, and matches the corresponding action training unit, thinking training unit and emotion training unit based on the behavior feature parameters, wherein the specific pattern comprises at least different emotion patterns, The deep learning training module selects the action training unit, thinking training unit, and emotion training unit whose matching degree is within a set threshold range based on the specific pattern of the infant to form an infant training program; The behavior feature extraction module extracts behavior feature parameters based on the CNN model, at least one behavior feature parameter corresponds to a specific pattern of the infant; The behavior feature extraction module uses cosine similarity to match behavior feature parameters with behavior models, and uses at least one action training unit, at least one thinking training unit, and at least one emotion training unit corresponding to the maximum similarity value as elements of the infant training program.

2. The infant needs intelligent analysis system according to claim 1, characterized in that: The behavior feature extraction module extracts behavior feature parameters of the infant based on the physiological information, the sound information and / or the behavior information of the infant, and selects a matching infant training program based on analysis of the behavior feature parameters.

3. The infant needs intelligent analysis system according to claim 2, characterized in that: The behavior feature extraction module includes a matching recognition module, The matching recognition module uses cosine similarity to compare the behavioral feature parameters with the behavioral models in the database (31) and outputs a set of similarity values, sorts the similarity values ​​from large to small, and outputs the infant training program corresponding to the maximum similarity value in the database (31).

4. The infant needs intelligent analysis system according to claim 3, characterized in that: The matching recognition module in the behavior feature extraction module matches the physiological information corresponding to the action training unit with the physiological information corresponding to the thinking training unit. If the matching degree is between 60% and 80%, the action training unit and the thinking training unit are coordinated; if the matching degree is less than 60%, the action training unit and the thinking training unit are not coordinated.

5. The infant demand intelligent analysis system according to claim 4, characterized in that: The matching recognition module in the behavior feature extraction module matches the physiological information corresponding to the action training unit with the physiological information corresponding to the emotion training unit, and the matching degree needs to meet the range of the preset threshold; The matching recognition module matches the physiological information corresponding to the thinking training unit with the physiological information corresponding to the emotion training unit, and the matching degree needs to be within the range of a preset threshold.

6. The infant demand intelligent analysis system according to claim 5, characterized in that: The deep learning training module uses CNN algorithm to train a CNN model; The deep learning training module transfers the trained CNN model to the behavior feature extraction module; The behavior feature extraction module uses a CNN model to extract the physiological information, the sound information and / or the behavior information of the infant to obtain behavior feature parameters.

7. The infant demand intelligent analysis system according to claim 6, characterized in that: The cloud server (30) further includes a teaching module (32), wherein the teaching module (32) completes a teaching process of a theoretical emotional state stored in a preset database (31) based on actual emotional state information input by the second intelligent terminal (20), and the emotional state analysis module (33) instructs the second intelligent terminal (20) to issue a demand prompt and / or a warning prompt to the current user based on demand information corresponding to the actual emotional state of the infant.

8. The infant demand intelligent analysis system according to claim 7, characterized in that: The teaching module (32) in the cloud server (30) analyzes and completes the teaching process based on the life record information and / or actual emotional state information input by the second intelligent terminal (20), wherein: The teaching module (32) pre-configures parameters of an actual emotional state information analysis algorithm based on at least two types of abnormal emotional state information.

9. The infant demand intelligent analysis system according to claim 8, characterized in that: The emotional state analysis module (33) stores the life record information and / or actual emotional state information of the infant input by the user to the second intelligent terminal (20) in the form of text, voice, or video, and the physiological information, sound information, and / or behavioral information automatically collected by the first intelligent terminal (10) in a correlated manner in a preset database (31).

Images