Intelligent newborn nursing system and method based on cry recognition

Abstract

The invention relates to an intelligent newborn nursing system and method based on cry recognition. The system comprises a real-time data acquisition module for acquiring real-time sound, environment data and a surrounding image sequence of the newborn; the voiceprint anomaly judgment module filters the sound signals to extract features, classifies cry types through a support vector machine, and fuses environment data to generate an anomaly demand identifier; the image attitude decision-making module takes the identifier as a trigger, analyzes the image sequence to obtain abnormal attitude information of the newborn, and integrates related data to obtain an emergency demand comprehensive decision-making basis; the instruction iteration planning module generates a response instruction sequence based on the basis, combines execution feedback update parameters, fuses historical nursing data and a current deviation level, and generates a nursing execution plan containing dynamic execution opportunity, hierarchical monitoring nodes and an adaptive feedback period. According to the system, the problems that traditional nursing depends on manpower, response lags behind and the misjudgment rate is high are solved through cooperation of the modules, and continuous and accurate intelligent nursing is provided for newborns.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of artificial intelligence, and particularly relates to a new-born baby intelligent nursing system and method based on cry sound recognition. BACKGROUND

[0002] In the daily nursing scene of new-born babies, the new-born babies usually convey physiological or environmental discomfort and other needs through cries, and the traditional nursing mode relies on manual real-time on-duty, which has problems such as response delay, strong subjectivity of demand judgment, and insufficient continuity of nursing. The existing new-born baby nursing technology based on sound recognition mostly only makes single-dimensional type judgment on cries, lacks multi-source fusion analysis of environmental data (such as temperature, humidity, light, and sound decibel) and new-born baby posture information, leads to low demand recognition accuracy and high abnormal state misjudgment rate, and is difficult to realize personalized and dynamic nursing response adjustment according to the physiological rhythm of new-born babies and historical nursing data, so it cannot meet the fine and intelligent nursing needs of new-born babies. SUMMARY

[0003] Therefore, it is necessary to provide a new-born baby intelligent nursing system and method based on cry sound recognition, which can effectively solve the problems of traditional nursing relying on manual work, response lag, and high misjudgment rate, and provide continuous, accurate, and intelligent nursing support for new-born babies.

[0004] In a first aspect, the application provides a new-born baby intelligent nursing system based on cry sound recognition, comprising:

[0005] A real-time data acquisition module is configured to acquire real-time sound signals of new-born babies, environmental data, and surrounding image sequences.

[0006] A voiceprint abnormality judgment module is configured to extract feature parameters from the filtered real-time sound signals and obtain cry classification results by using a support vector machine, and obtain abnormal demand identifiers by fusing the environmental data.

[0007] An image posture decision module is configured to obtain new-born baby abnormal posture information by performing posture analysis on the surrounding image sequences based on the abnormal demand identifiers, and obtain a comprehensive decision basis for judging emergency demands by combining the cry classification results and the environmental data.

[0008] An instruction iteration planning module is configured to generate a response instruction sequence according to the comprehensive decision basis, update the response instruction sequence based on feedback data after execution, and generate a nursing execution plan by fusing historical nursing data and a current deviation level.

[0009] In one embodiment, the voiceprint abnormality judgment module is further configured to:

[0010] Filter the collected real-time sound signals of new-born babies to obtain filtered sound signal data.

[0011] The feature extraction algorithm is used to extract sound feature parameters including volume, frequency, and duration from the sound signal data.

[0012] The extracted sound feature parameters are input into a preset support vector machine algorithm model, and a classification result of the newborn cry type is obtained through model operation; the classification result includes a hunger type cry, a pain type cry, and an uncomfortable type cry.

[0013] The environment data of the environment in which the newborn is located is synchronously collected; the environment data includes temperature data, humidity data, illumination intensity data, and sound decibel data.

[0014] The environment data and the classification result of the cry type are fused by using a weighted fusion algorithm to obtain a fusion result, and the fusion result is judged based on a preset abnormality judgment threshold to determine whether there is an environment-related abnormality demand, thereby obtaining a binary judgment result.

[0015] If the binary judgment result is that there is an environment-related abnormality demand, a corresponding abnormality demand identifier is generated; the abnormality demand identifier includes cry type information, environment data characteristics, and an abnormality level.

[0016] In one embodiment, based on the abnormality demand identifier, posture analysis is performed on the surrounding image sequence to obtain newborn abnormal posture information, and the cry classification result and the environment data are combined to obtain a comprehensive decision basis for judging the emergency demand, including:

[0017] The convolutional neural network is used to extract features from the acquired surrounding image sequence frame by frame to obtain image feature data containing texture, contour, and target area.

[0018] The image feature data is input into a posture analysis model, the deviation of the newborn's limb and face key point coordinates is calculated and compared with a standard posture template to determine the newborn's abnormal posture information; the abnormal posture information includes limb distortion and abnormal facial expression.

[0019] The multi-modal data fusion algorithm is used to integrate the abnormal posture information, the classification result of the cry type, and the corresponding environment data to obtain an emergency demand judgment result.

[0020] If the confidence of the emergency demand judgment result exceeds a preset threshold, the posture key point dynamic change information of the newborn, the frequency time sequence fluctuation information of the cry signal, and the instantaneous abnormal fluctuation information of the environment data are extracted respectively to generate a comprehensive decision basis including demand type and emergency level.

[0021] In one embodiment, the deviation of the newborn's limb and face key point coordinates is calculated by the following formula:

[0022]

[0023] wherein, represents the newborn in the frame image, the comprehensive coordinate deviation value of the group of key points, represents the frame, actual pixel coordinates of the key point, the reference pixel coordinates of the key point in the preset standard posture template, weight coefficient of the key point, represents the inter-frame time decay coefficient, the acquisition time stamp of the frame and the frame.

[0024] In one of the embodiments, the instruction iteration planning module is further configured to:

[0025] generate a response instruction sequence according to the comprehensive decision basis, and execute the response instruction sequence to obtain real-time neonatal care state data; the care state data includes cry intensity decay value, limb posture recovery degree, environmental parameter adjustment compliance rate, and physiological sign stability value.

[0026] compare the care state data with the preset care compliance threshold, divide the deviation level according to the deviation degree, and match the corresponding dynamic adjustment coefficient.

[0027] adjust the parameters of the response instruction sequence according to the dynamic adjustment coefficient to obtain an updated response instruction sequence; the parameters include alarm volume gradient value, temperature adjustment step amplitude, and execution frequency adaptive interval.

[0028] fuse and analyze the updated response instruction sequence, neonatal historical care data, and current deviation level to generate a care execution plan; the historical care data includes past abnormal response effect and physiological rhythm characteristics; the care execution plan includes dynamic execution timing, hierarchical monitoring node, and adaptive feedback cycle.

[0029] In one of the embodiments, the updated response instruction sequence, neonatal historical care data, and current deviation level are fused and analyzed to generate a care execution plan, including:

[0030] extract the core parameters of the updated response instruction sequence, associate the neonatal historical care data and the current deviation level, and construct a multi-dimensional fusion analysis data set.

[0031] Based on the fusion analysis dataset combined with the circadian rhythm characteristics, the dynamic execution timing rules are designed, the density standards of the hierarchical monitoring nodes are formulated according to the current deviation level, and the preliminary design scheme of each execution element is formed.

[0032] Referring to the past abnormal response effect in the historical care data, the logic of the dynamic execution timing rules, the hierarchical monitoring nodes and the adaptive feedback cycle is optimized, and a complete care execution plan is generated.

[0033] If the adaptation degree of each element in the care execution plan and the core parameters of the updated response instruction sequence meets the preset threshold range, the final executable care execution plan is output.

[0034] In the second aspect, the application also provides a new-born intelligent care method based on cry recognition, which comprises the following steps:

[0035] Obtaining real-time sound signals, environmental data and surrounding image sequences of the new-born;

[0036] After filtering the real-time sound signals, the feature parameters are extracted and the support vector machine is used to obtain the cry classification result, and the abnormal demand identifier is obtained by fusing the environmental data;

[0037] Based on the abnormal demand identifier, the posture analysis is performed on the surrounding image sequences to obtain the abnormal posture information of the new-born, and the comprehensive decision basis for judging the emergency demand is obtained by combining the cry classification result and the environmental data;

[0038] According to the comprehensive decision basis, a response instruction sequence is generated, and the response instruction sequence is updated based on the feedback data after execution, and the care execution plan is generated by fusing the historical care data and the current deviation level.

[0039] In the third aspect, the application also provides a computer device, which comprises a memory and a processor, the memory stores a computer program, and the processor implements the steps of the foregoing system when executing the computer program.

[0040] In the fourth aspect, the application also provides a computer readable storage medium, which stores a computer program, and the computer program is executed by the processor to perform the steps of the foregoing system.

[0041] The new-born intelligent nursing system based on cry recognition, the new-born intelligent nursing method, the computer device and the storage medium, the real-time data acquisition module acquires real-time sound signals, environment data and surrounding image sequences of the new-born, and provides multi-source original data support for subsequent analysis; the voiceprint abnormality determination module receives the real-time sound signals, filters out environmental interference through filtering processing first, then extracts core characteristic parameters such as volume, frequency and duration, completes cry type classification through a support vector machine algorithm, then performs abnormality checking by fusing the environment data collected in real time, and generates an abnormal demand identifier containing cry type, environment characteristics and abnormality level; the image posture decision module takes the abnormal demand identifier as a trigger condition, carries out posture analysis on the surrounding image sequences, acquires new-born abnormal posture information through key point coordinate deviation calculation and standard template comparison, further integrates the cry classification result and the environment data, and obtains a comprehensive decision basis for judging emergency demand through multi-modal fusion operation; the instruction iteration planning module generates an initial response instruction sequence based on the comprehensive decision basis, collects feedback data (such as new-born nursing state changes and instruction execution effects) after the execution of the instruction, adjusts and updates the core parameters of the response instruction sequence accordingly, and generates a nursing execution plan containing a dynamic execution time, a hierarchical monitoring node and an adaptive feedback period by fusing the deviation level of the new-born historical nursing data (such as past abnormal response effects and physiological rhythm characteristics) and the current nursing state. Through the data flow closed loop and function cooperation among the modules, the system realizes the full-process automatic processing from multi-source data acquisition, feature extraction, demand determination to response optimization, compensates for the limitations of single-dimensional analysis through multi-modal data fusion, improves the accuracy of new-born demand recognition and the reliability of abnormal state judgment; through the feedback-driven instruction iteration and the personalized plan generation supported by the historical data, the dynamic adjustment and individual adaptation of the nursing strategy are realized, effectively solving the problems of traditional nursing such as dependence on manual work, response lag and high misjudgment rate, and providing continuous, accurate and intelligent nursing support for new-borns. BRIEF DESCRIPTION OF DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related art, the following will briefly introduce the drawings needed to be used in the embodiment or related art description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creating any inventive labor.

[0043] Figure 1 A structure block diagram of a new-born intelligent nursing system based on cry recognition provided by the embodiment of the present application;

[0044] Figure 2 A flowchart of a new-born intelligent nursing method based on cry recognition provided by the embodiment of the present application. DETAILED DESCRIPTION

[0045] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application will be further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and do not limit the present application.

[0046] In one of the embodiments, as shown in the accompanying drawings, Figure 1 The present application provides a new-born intelligent care system based on cry recognition, which can include:

[0047] The real-time data acquisition module 101 is used to acquire real-time sound signals of new-borns, environmental data and surrounding image sequences.

[0048] Specifically, the synchronization acquisition of multi-source data is completed by the preset sensing device and image acquisition device. The real-time sound signals of new-borns are acquired by a directional microphone array, the environmental data are acquired by sensors deployed in the new-born care area (including temperature, humidity, light intensity and sound decibel data), and the surrounding image sequences are acquired by a high-definition camera for multi-frame continuous acquisition. The acquisition timing is consistent with the sound signals and environmental data to ensure the time correlation of the data. The module directly outputs the three types of raw data collected to the subsequent voiceprint abnormality judgment module 102 and image posture decision module 103.

[0049] The voiceprint abnormality judgment module 102 is used to extract feature parameters after filtering the real-time sound signals and obtain the cry classification results by using a support vector machine, and obtain an abnormal demand identifier by fusing the environmental data.

[0050] Further, the module receives the real-time sound signals output by the real-time data acquisition module 101, first pre-processes them by using a digital filtering algorithm to filter out invalid signals such as environmental noise and device interference to obtain pure sound signal data; then extracts core parameters such as volume, frequency and duration reflecting the characteristics of the cry from the filtered sound signals by using a feature extraction algorithm; inputs the extracted feature parameters into a preset support vector machine algorithm model to obtain the classification results of the new-born cry types (including hunger cry, pain cry and discomfort cry) by the pattern matching operation of the model; simultaneously receives the environmental data output by the real-time data acquisition module 101, fuses and calculates the environmental data and the cry classification results by using a weighted fusion algorithm, completes the abnormality verification based on a preset abnormality judgment threshold, finally generates an abnormal demand identifier containing the cry type information, environmental data characteristics and abnormality level, and outputs the identifier together with the cry classification results and environmental data to the image posture decision module 103.

[0051] The image posture decision module 103 is used for identifying the posture of the surrounding image sequence based on the abnormal demand identification to obtain the abnormal posture information of the newborn, and combining the crying sound classification result and the environmental data to obtain the comprehensive decision basis for judging the emergency demand.

[0052] Illustratively, the abnormal demand identification output by the voiceprint abnormality judgment module 102 is taken as the trigger condition, the surrounding image sequence transmitted by the real-time data acquisition module 101 is received, and the crying sound classification result and the environmental data output by the voiceprint abnormality judgment module 102 are received. First, the convolutional neural network is used to extract the features of the surrounding image sequence frame by frame to obtain the image feature data containing the texture, contour and target region of the newborn; the image feature data is input into the posture analysis model, the abnormal posture information (including limb distortion and abnormal facial expression) of the newborn is determined by calculating the deviation of the limb and facial key point coordinates and comparing with the standard posture template; then, the multi-modal data fusion algorithm is used to integrate and operate the abnormal posture information, the crying sound classification result and the environmental data to obtain the comprehensive decision basis of the emergency demand containing the demand type, the emergency degree and the confidence, which is output to the instruction iteration planning module 104.

[0053] The instruction iteration planning module 104 is used for generating the response instruction sequence according to the comprehensive decision basis and updating the response instruction sequence based on the feedback data after execution, and generating the nursing execution plan by fusing the historical nursing data and the current deviation level.

[0054] The comprehensive decision basis output by the image posture decision module 103 is received, and the initial response instruction sequence is generated based on the basis, the instruction sequence contains the core parameters such as the alarm volume gradient value, the temperature adjustment step amplitude and the execution frequency adaptive interval; in the execution process of the response instruction sequence, the real-time nursing state data (including the crying sound intensity attenuation value, the limb posture recovery degree, the environmental parameter adjustment compliance rate and the physiological sign stability value) of the newborn fed back by the monitoring equipment is collected, the feedback data is compared with the preset nursing compliance threshold, the deviation level is divided according to the deviation degree and the corresponding dynamic adjustment coefficient is matched, the core parameters of the response instruction sequence are adjusted accordingly to obtain the updated response instruction sequence; at the same time, the historical nursing data (including the past abnormal response effect and the physiological rhythm characteristics) of the newborn is called, the updated response instruction sequence is fused and analyzed with the historical nursing data and the current deviation level to generate the nursing execution plan containing the dynamic execution time, the hierarchical monitoring node and the adaptive feedback cycle, and the closed-loop optimization and personalized adaptation of the nursing strategy are realized.

[0055] The above-mentioned new-born intelligent nursing system based on cry recognition, the real-time data acquisition module acquires real-time sound, environmental data and surrounding image sequence of the new-born, and provides multi-source original data; the voiceprint abnormality determination module filters and extracts features from the sound signal, classifies the cry type through a support vector machine, and generates an abnormal demand identifier containing the cry type, environmental features and abnormality level by fusing the environmental data; the image posture decision module is triggered by the identifier, analyzes the image sequence to obtain abnormal posture information of the new-born, integrates related data, and obtains an emergency demand comprehensive decision basis through multi-modal fusion; the instruction iteration planning module generates a response instruction sequence based on the basis, updates parameters combined with execution feedback, fuses historical nursing data and current deviation level, and generates a nursing execution plan containing dynamic execution time, hierarchical monitoring nodes and adaptive feedback period. The system realizes full-process automation through module cooperation and data flow closed loop, improves recognition accuracy and judgment reliability through multi-modal fusion, solves the problems of traditional nursing relying on manual, response lag and high misjudgment rate through instruction iteration and personalized plan, and provides continuous and accurate intelligent nursing for new-borns.

[0056] In one of the embodiments, the voiceprint abnormality determination module can also be used for:

[0057] Step S101, filtering the collected real-time sound signal of the new-born to obtain filtered sound signal data.

[0058] Step S102, extracting sound feature parameters including volume, frequency and duration from the sound signal data using a feature extraction algorithm.

[0059] Specifically, for the volume feature, the amplitude mean and peak value of the sound signal data within the effective time length are calculated to quantify the intensity of the cry; for the frequency feature, the time-domain sound signal is converted to the frequency domain using fast Fourier transform to extract the fundamental frequency, harmonic frequency and frequency distribution interval to capture the pitch characteristics of the cry; for the duration feature, the effective duration of a single cry and the interval time between adjacent cries are determined by detecting the starting and ending thresholds of the sound signal. Finally, the above three types of feature parameters are integrated into a structured sound feature parameter set.

[0060] Step S103, inputting the extracted sound feature parameters into a preset support vector machine algorithm model to obtain the classification result of the new-born cry type through model operation; the classification result includes hunger type cry, pain type cry and discomfort type cry.

[0061] Preferably, the support vector machine algorithm model has been trained in advance by a large number of newborn cry samples labeled with "hunger class, pain class, and discomfort class", and a stable classification decision boundary is formed. In the model operation process, first, the input sound feature parameters are mapped to a high-dimensional feature space, the distance between the feature parameters and the decision boundary of each class is calculated through the kernel function, and the corresponding class of the current cry is determined according to the distance minimization principle; then the explicit classification result of the newborn cry type is output, and the classification result is limited to three categories of hunger class cry, pain class cry, and discomfort class cry.

[0062] Step S104, synchronously collecting environment data of the environment where the newborn is located; the environment data includes temperature data, humidity data, illumination intensity data, and sound decibel data.

[0063] Step S105, using a weighted fusion algorithm to fuse the environment data and the classification result of the cry type to obtain a fusion result, and judging whether there is an environment-related abnormal demand based on a preset abnormal judgment threshold to obtain a binary judgment result.

[0064] Preferably, first, based on the influence degree of each data dimension on the abnormal demand judgment, fixed or dynamic weight coefficients are respectively assigned to each dimension of the environment data and the cry type classification result, and then the above quantized environment data and cry classification result are linearly weighted and summed by the weighted fusion algorithm to obtain a fusion result representing the "sound-environment" correlation relationship; wherein the preset abnormal judgment threshold is calibrated and determined through a plurality of newborn care scene experimental data, and can be individually adjusted according to the actual care demand, and then the calculated fusion result is compared with the preset abnormal judgment threshold, if the fusion result is greater than the preset abnormal judgment threshold, it is determined that there is an environment-related abnormal demand, otherwise it is determined that there is not, and finally a binary judgment result of "existence" or "nonexistence" is output.

[0065] Step S106, if the binary judgment result is that there is an environment-related abnormal demand, an abnormal demand identifier corresponding thereto is generated; the abnormal demand identifier includes cry type information, environment data characteristics, and an abnormal level.

[0066] Specifically, the collected real-time sound signal of the newborn is first filtered to remove environmental interference to obtain pure filtered sound signal data; then a feature extraction algorithm is used to extract core sound feature parameters such as volume, frequency, and duration from the sound signal data; the extracted sound feature parameters are input into a preset support vector machine algorithm model, and through pattern matching operation of the model, a newborn cry type classification result containing hunger cry, pain cry, and discomfort cry is obtained; during the whole sound signal processing process, the temperature data, humidity data, light intensity data, and sound decibel data of the environment where the newborn is located are synchronously collected to ensure the time sequence consistency of the two types of data; then a weighted fusion algorithm is used to quantitatively fuse the environmental data and the cry type classification result to obtain a fusion result, and based on a preset abnormality judgment threshold, the fusion result is standardized to determine whether there is an environment-related abnormal demand, and a binary judgment result of "existence / nonexistence" is output; if the binary judgment result is an environment-related abnormal demand, a corresponding abnormal demand identifier is generated, which clearly contains cry type information, environmental data characteristics, and abnormality level.

[0067] The embodiment guarantees the basic accuracy of cry classification through sound signal preprocessing and feature extraction, realizes accurate division of cry types relying on support vector machine algorithm, constructs a "sound-environment" two-dimensional analysis system through synchronous collection and weighted fusion of environmental data, and avoids one-sidedness of single data dimension judgment; the preset abnormality judgment threshold makes the abnormal demand judgment standard unified and quantifiable, the generated abnormal demand identifier integrates core decision information, which not only ensures the logical closed loop of data processing and clear and traceable data flow, but also improves the accuracy and reliability of environment-related abnormal demand judgment, provides high-quality and targeted input data for subsequent image posture analysis and response instruction generation, and effectively reduces the risk of abnormal state misjudgment.

[0068] In one of the embodiments, based on the abnormal demand identifier, posture analysis is performed on the surrounding image sequence to obtain newborn abnormal posture information, and the comprehensive decision basis for judging emergency demand is obtained in combination with the cry classification result and the environmental data, which can include the following preferred steps:

[0069] Step S201: The acquired surrounding image sequence is frame by frame subjected to feature extraction by using a convolutional neural network to obtain image feature data containing texture, contour, and target region.

[0070] The convolutional neural network comprises multiple convolutional layers and pooling layers. The convolutional layers capture local gray level changes of the image through sliding window operation to generate a texture feature map. The pooling layers down-sample the texture feature map to retain key information and reduce data dimension, while extracting overall contour features of the image. The foregoing features are integrated through a fully connected layer to locate and segment a target region where the newborn is located, and output structured image feature data containing texture details, contour shape, and spatial position information of the target region.

[0071] In step S202, the image feature data is input into a posture analysis model to calculate the coordinate deviation of the newborn's limbs and facial key points and compare it with a standard posture template to determine the abnormal posture information of the newborn. The abnormal posture information includes limb distortion and abnormal facial expression.

[0072] Preferably, the model first locates the key points of the newborn's limbs (wrist, elbow, knee, ankle, etc.) and face (eye corner, mouth corner, nose wing, etc.) based on the image feature data to obtain the actual pixel coordinates of each key point. The deviation value of the actual coordinates of each key point from the reference coordinates of the corresponding key point in the preset standard posture template is quantified through a coordinate deviation calculation formula, and then each deviation value is compared with a preset posture abnormality judgment threshold. If the deviation value of any key point exceeds the threshold or the deviation values of multiple key points satisfy the abnormality judgment condition, it is determined that the corresponding abnormality exists, and finally the abnormal posture information of the newborn including limb distortion and abnormal facial expression is output.

[0073] In step S203, a multi-modal data fusion algorithm is used to integrate the abnormal posture information, the classification results of the crying type, and the corresponding environmental data to obtain the emergency demand judgment result.

[0074] The multi-modal data fusion algorithm using feature layer fusion first standardizes the three types of data to unify the data dimension and unit. Then, according to the influence degree of each data dimension on the emergency demand judgment, a weight coefficient is assigned, and a weighted sum operation is performed on the standardized features. Through this integration process, the emergency demand judgment result containing the demand correlation degree (the matching degree of abnormal posture, crying type, and environmental data), abnormality degree, and confidence is output.

[0075] In step S204, if the confidence of the emergency demand judgment result exceeds a preset threshold, the dynamic change information of the posture key points of the newborn, the frequency time sequence fluctuation information of the crying signal, and the instantaneous abnormal fluctuation information of the environmental data are extracted respectively to generate a comprehensive decision basis containing the demand type and the emergency degree.

[0076] In the first step, the convolutional neural network is used to extract features from the acquired surrounding image sequence frame by frame, taking the abnormal demand identification as the trigger condition. The local texture features, overall contour features, and spatial features of the newborn target area are extracted step by step through the convolutional layers and pooling layers of the network, and finally the structured image feature data containing texture, contour, and target area are output. Then the image feature data is input into the preset posture analysis model. The model first locates the key points of the newborn's limbs and face (such as wrist, elbow, eye corner, mouth corner, etc.) and obtains their actual pixel coordinates. The deviation value of the actual coordinates of the key points from the reference coordinates of the corresponding key points in the standard posture template is quantified by an innovative coordinate deviation calculation formula. Then the deviation value is compared with the posture abnormality judgment threshold to determine whether the newborn has abnormal posture information such as limb distortion and abnormal facial expression. Subsequently, taking the abnormal posture information as the core, combining the crying sound type classification results obtained in the early stage and the corresponding environmental data, a multi-modal data fusion algorithm is used to integrate and weight the features of the three types of data, and the emergency demand judgment result containing the demand correlation degree, abnormality degree, and confidence is obtained. Finally, it is judged whether the confidence of the emergency demand judgment result exceeds the preset threshold. If it exceeds, the dynamic change information of the posture key points of the newborn in the continuous frames, the frequency time fluctuation information of the crying sound signal over time, and the instantaneous abnormal fluctuation information of the environmental data are further extracted. The three types of dynamic information are fused with the static data obtained in the early stage to generate a comprehensive decision basis that clearly contains the demand type and the emergency level.

[0077] The embodiment guarantees the accuracy of image feature extraction through the convolutional neural network, realizes the quantitative determination of abnormal posture by relying on coordinate deviation calculation and standard template comparison, integrates the three types of core data of "image-sound-environment" by means of multi-modal data fusion algorithm, and makes up for the limitations of single-dimensional analysis. Through confidence threshold screening and dynamic fluctuation information supplement, the reliability and comprehensiveness of the emergency demand judgment are further improved. The generated comprehensive decision basis contains not only static judgment results but also dynamic change characteristics, providing high-quality, multi-dimensional decision support for the accurate generation of subsequent response instruction sequences, effectively reducing the risk of emergency demand misjudgment and omission.

[0078] In one embodiment, the coordinate deviation of the newborn's limb and face key points can be calculated by the following formula:

[0079]

[0080] wherein, represents the comprehensive coordinate deviation value of the target newborn's group key points in the frame image, represents the actual pixel coordinates of the th key point in the frame, a reference pixel coordinate representing a preset standard posture template key point, a reference pixel coordinate representing a preset standard posture template key point, a weight coefficient representing a preset standard posture template key point, a weight coefficient representing a preset standard posture template key point, a total number of key points to be detected in a single frame image, an inter-frame time decay coefficient, a collection timestamp of a first frame, a collection timestamp of a first frame, a collection timestamp of a first frame.

[0081] The coordinate deviation calculation formula of the embodiment can improve the pertinence of deviation calculation by introducing a dynamic weight coefficient of the key point, assigning weights according to the importance difference of different parts of the newborn's limbs and face in posture anomaly determination, and fusing an inter-frame time decay factor to effectively filter transient interference caused by newborn breathing, slight limb micro-movement, etc., and reduce the risk of misjudgment. The Euclidean distance is used to quantify the deviation of the actual coordinates of the key points in a single frame from the reference coordinates, and the collection timestamp of the adjacent frame is associated and analyzed, taking into account the precise quantification of single-frame deviation and the continuity of multi-frame posture changes, avoiding the limitations of single coordinate comparison. Through the cooperative operation of multiple parameters, the scientific quantification of the key point coordinate deviation of the newborn's limbs and face is realized, providing reliable numerical support for accurate determination of abnormal posture information, and further ensuring the accuracy of subsequent emergency demand judgment and nursing response.

[0082] In one of the embodiments, the instruction iteration planning module can also be used to:

[0083] Step S301, generate a response instruction sequence according to the comprehensive decision basis, and execute the response instruction sequence to obtain real-time newborn care state data; the care state data includes cry intensity decay value, limb posture recovery degree, environmental parameter adjustment compliance rate, and physiological sign stability value.

[0084] Step S302, compare the care state data with the preset care compliance threshold, divide the deviation level according to the deviation degree, and match the corresponding dynamic adjustment coefficient.

[0085] Step S303, adjust the parameters of the response instruction sequence according to the dynamic adjustment coefficient to obtain an updated response instruction sequence; the parameters include alarm volume gradient value, temperature adjustment step amplitude, and execution frequency adaptive interval.

[0086] Preferably, according to the obtained dynamic adjustment coefficient, and the alarm volume gradient value, temperature adjustment step size, execution frequency adaptive interval dynamic adjustment coefficient and deviation level corresponding one by one in the original response instruction sequence, the higher the deviation level, the stronger the numerical adaptability of the adjustment coefficient. In specific operation, the dynamic adjustment coefficient is quantized and operated with the three types of parameters respectively. The adjustment of the alarm volume gradient value focuses on the gradient amplitude of volume increment / decrement, the adjustment of the temperature adjustment step size targets the step precision of the environment temperature approaching the suitable range, and the adjustment of the execution frequency adaptive interval surrounds the time interval of instruction repeated execution; through the targeted parameter correction, the original response instruction sequence is more suitable for the deviation degree of the current care scene, and finally an updated response instruction sequence with optimized parameters is output.

[0087] In step S304, the updated response instruction sequence is combined with the neonatal historical care data and the current deviation level for fusion analysis to generate a care execution plan; the historical care data includes past abnormal response effects and physiological rhythm characteristics; the care execution plan includes dynamic execution time, hierarchical monitoring nodes, and adaptive feedback period.

[0088] Further, the output updated response instruction sequence is received, and the neonatal historical care data (including past abnormal response effects and physiological rhythm characteristics) and the current deviation level are called to construct a multi-dimensional fusion analysis data set. In the fusion analysis process, the dynamic execution time adapted to the individual work-rest is designed in combination with the physiological rhythm characteristics (such as neonatal sleep-wake cycle, feeding interval, etc.) in the historical care data; the hierarchical monitoring node density is determined according to the current deviation level, the higher the deviation level, the more intensive the monitoring node distribution; the length rule of the adaptive feedback period is optimized referring to the past abnormal response effects, to ensure that the feedback efficiency and the adjustment effect match; the dynamic execution time, the hierarchical monitoring node, and the adaptive feedback period are integrated to form a complete care execution plan that is logically coherent and adapted to individual needs.

[0089] Firstly, according to the previously generated comprehensive decision basis, a response instruction sequence containing core parameters such as alarm volume gradient value, temperature adjustment step size, and execution frequency adaptive interval is generated. After executing the sequence, real-time neonatal care state data is collected through monitoring equipment, which specifically includes cry intensity attenuation value, limb posture recovery degree, environmental parameter adjustment compliance rate, and physiological sign stability value. Then, the collected real-time care state data is compared with the preset care compliance threshold item by item. According to the deviation degree of the actual data and the threshold, the deviation level is divided into mild, moderate, and severe. Based on the quantitative standard of each deviation level, the corresponding dynamic adjustment coefficient is matched. Then, the core parameters of the original response instruction sequence are corrected according to the dynamic adjustment coefficient, and the updated response instruction sequence that adapts to the current care scene is obtained. Finally, the neonatal historical care data (including past abnormal response effects and physiological rhythm characteristics) are retrieved. The updated response instruction sequence, historical care data, and current deviation level are subjected to multidimensional fusion analysis. The dynamic execution time is determined in combination with the physiological rhythm characteristics. The hierarchical monitoring node density is set according to the deviation level. The adaptive feedback period is optimized by referring to the past abnormal response effects. Finally, a complete care execution plan containing dynamic execution time, hierarchical monitoring node, and adaptive feedback period is generated.

[0090] The embodiment ensures the pertinence of the initial response instruction sequence. Through the quantitative comparison of care state data and compliance threshold, deviation level division, and dynamic adjustment coefficient matching, the response instruction parameters are accurately optimized, avoiding the limitations of fixed instructions. Relying on the fusion analysis of neonatal historical care data and current deviation level, the generated care execution plan can adapt to individual physiological rhythm and past care experience, taking into account the dynamics and individualization of execution. The timeliness and accuracy of care response are ensured, effectively improving the improvement efficiency and stability of neonatal care state.

[0091] In one of the embodiments, the updated response instruction sequence is combined with neonatal historical care data and current deviation level for fusion analysis to generate a care execution plan, which can include the following steps:

[0092] Step S401, the core parameters of the updated response instruction sequence are extracted, associated with neonatal historical care data and current deviation level, and a multidimensional fusion analysis data set is constructed.

[0093] Step S402, based on the fusion analysis data set, the dynamic execution time rules are designed in combination with the physiological rhythm characteristics. The density standard of hierarchical monitoring nodes is determined according to the current deviation level, and the preliminary design scheme of each execution element is formed.

[0094] Further, in terms of dynamic execution timing rule design, the physiological rhythm characteristics of the newborn (such as sleep-wake cycle, feeding interval, and work-rest regularity) are extracted from the fusion analysis data set, and the core parameter characteristics of the updated response instruction sequence (such as temperature regulation avoiding the deep sleep period of the newborn) are combined to determine the time window and priority rule of the instruction trigger, ensuring that the execution timing is adapted to the individual work of the newborn. In terms of hierarchical monitoring node density standard, the current deviation level (mild, moderate, and severe) in the data set is used to set the corresponding monitoring frequency and node interval, where mild deviation corresponds to lower monitoring density and longer node interval, moderate and severe deviation increase the monitoring density and shorten the node interval in turn, ensuring the monitoring coverage in abnormal state. Finally, the above dynamic execution timing rule and hierarchical monitoring node density standard are integrated to form a preliminary design scheme containing two core execution elements.

[0095] Step S403, referring to the past abnormal response effect in the historical care data, the logic of dynamic execution timing rule, hierarchical monitoring node and adaptive feedback cycle is optimized to generate a complete care execution plan.

[0096] Specifically, the trigger threshold and time window of the dynamic execution timing are adjusted according to the past abnormal response effect to avoid invalid execution period in past verification; the actual interval parameters of the hierarchical monitoring node are optimized to ensure that the monitoring frequency matches the abnormal treatment demand; the design logic of the adaptive feedback cycle is added to set the initial value and adjustment range of the feedback cycle according to the time length rule of "instruction execution-state improvement" in the past response effect, so that the feedback efficiency and the improvement rhythm of the newborn state are adapted; through the coordinated optimization of the dynamic execution timing rule, the hierarchical monitoring node parameters, and the adaptive feedback cycle logic, a complete care execution plan containing three core elements and logically consistent is integrated.

[0097] Step S404, if the adaptation degree of each element in the care execution plan and the core parameters of the updated response instruction sequence meets the preset threshold range, the final executable care execution plan is output.

[0098] Firstly, the core parameters such as alarm volume gradient value, temperature adjustment step size, and execution frequency adaptive interval in the updated response instruction sequence are extracted, which are associated and matched with the historical neonatal care data (including past abnormal response effect, physiological rhythm characteristics) and current deviation level, the corresponding relationship of each data dimension is determined, and a multi-dimensional fusion analysis data set covering instruction parameters, historical data, and deviation level is constructed; based on the fusion analysis data set, the physiological rhythm characteristics (such as neonatal sleep-wake cycle, feeding interval, etc.) in the historical care data are combined, the dynamic execution time rules adapted to individual work-rest are designed, and the density standards of hierarchical monitoring nodes are formulated according to the current deviation level (mild / moderate / severe) (the higher the deviation level, the denser the distribution of monitoring nodes), forming a preliminary design scheme of each execution element including dynamic execution time, hierarchical monitoring nodes; then, the trigger logic of dynamic execution time, the interval setting of hierarchical monitoring nodes, and the length rules of adaptive feedback period are optimized according to the past abnormal response effect in the historical care data, to make up for the shortcomings of the preliminary design scheme, and to generate a complete care execution plan integrating dynamic execution time, hierarchical monitoring nodes, and adaptive feedback period; finally, the adaptation degree of each element in the care execution plan to the core parameters of the updated response instruction sequence is quantitatively evaluated, and if the adaptation degree meets the preset threshold range, the final executable care execution plan is output.

[0099] The embodiment constructs a fusion analysis data set through multi-dimensional data association, ensuring the data source comprehensiveness of the execution plan design; the preliminary scheme is designed based on the physiological rhythm characteristics and deviation level as the core basis, ensuring the pertinence and rationality of the execution elements; the plan logic adaptability and reliability are improved by relying on the optimization of past abnormal response effect; through the adaptation degree verification link, the execution plan and the instruction parameters are further ensured to be matched, and execution conflicts are avoided; both the precise adaptation of the care execution plan to the individual characteristics of the neonate and the current care scene and the executable of the plan are ensured, providing a rigorous execution basis for intelligent and personalized care.

[0100] In one of the embodiments, as shown in Figure 2 The application further provides a neonatal intelligent care method based on cry recognition, which can include the following steps:

[0101] Step S501, acquiring real-time sound signals of a neonate, environment data, and surrounding image sequences;

[0102] Step S502, extracting feature parameters after filtering the real-time sound signals and obtaining cry classification results by using a support vector machine, and obtaining abnormal demand identification by fusing environment data;

[0103] Step S503, based on the abnormal demand identification, the posture of the surrounding image sequence is analyzed to obtain the abnormal posture information of the newborn, and the comprehensive decision basis for judging the emergency demand is obtained by combining the crying sound classification result and the environment data;

[0104] Step S504, according to the comprehensive decision basis, a response instruction sequence is generated, and the response instruction sequence is updated based on the feedback data after execution, and a nursing execution plan is generated by fusing historical nursing data and current deviation level.

[0105] The above-mentioned intelligent nursing method for newborns based on crying sound recognition synchronously acquires real-time sound signals of newborns, environment data (temperature, humidity, light intensity, etc.) and surrounding image sequences through a pre-set acquisition device, and constructs a multi-source original data basis; the real-time sound signals are pre-processed by filtering to remove environmental interference, and core characteristic parameters such as volume, frequency and duration are extracted, the characteristic parameters are input into a pre-trained support vector machine algorithm model to obtain a crying sound classification result (including hunger class, pain class and discomfort class), and then a weighted fusion algorithm is used to fuse and operate the classification result and the synchronously acquired environment data, and an abnormal demand identification containing the crying sound type, environment characteristics and abnormal level is generated based on a pre-set abnormality judgment threshold; the surrounding image sequence is subjected to frame-by-frame feature extraction and posture analysis based on the abnormal demand identification as a trigger condition, and newborn abnormal posture information (including limb distortion and abnormal facial expression) is obtained through key point coordinate deviation calculation and standard template comparison; the crying sound classification result, environment data and abnormal posture information are further integrated, and a comprehensive decision basis for judging the emergency demand is obtained through multi-modal data fusion operation; finally, an initial response instruction sequence is generated based on the comprehensive decision basis, and after execution, real-time nursing state data of the newborn is collected as feedback, the deviation level is divided by combining the feedback data and the deviation degree of the pre-set nursing standard threshold and matching the dynamic adjustment coefficient, the response instruction sequence parameters are updated accordingly, and multi-dimensional analysis is performed by fusing the historical nursing data of the newborn (including past abnormal response effect and physiological rhythm characteristics) and the current deviation level, and finally a nursing execution plan containing dynamic execution time, hierarchical monitoring node and adaptive feedback period is generated.

[0106] The embodiment synchronously acquires multi-source data and performs cross-dimension fusion analysis, which makes up for the limitations of single data dimension judgment, improves the accuracy of newborn demand recognition and the reliability of abnormal state judgment, realizes dynamic adjustment of nursing response, avoids the rigidity of fixed instructions, generates personalized nursing execution plan by combining historical nursing data and individual physiological rhythm characteristics, ensures the adaptability of nursing strategy to individual newborns, and realizes whole-process automation processing from data acquisition, demand judgment to response execution without human intervention, effectively solving the problems of traditional nursing relying on manual operation, response lag and high misjudgment rate, and providing continuous, accurate and intelligent nursing support for newborns.

[0107] In one of the embodiments, the present application also provides a new-born intelligent care implementation mode with targeted care measures, which can include:

[0108] The real-time data acquisition module continuously acquires real-time sound signals of the new-born, environmental data (temperature data, humidity data, light intensity data and sound decibel data) and surrounding image sequences, wherein the sound signals are synchronously captured by a directional microphone array to ensure the integrity of the crying data; the voiceprint abnormality determination module filters the real-time sound signals, extracts core characteristic parameters such as volume, frequency, duration and pitch fluctuation after filtering out environmental noise, inputs the characteristic parameters into a pre-trained support vector machine algorithm model, and obtains four types of crying classification results including hunger type crying, pain type crying, discomfort type crying and companion-seeking type crying - wherein the companion-seeking type crying is characterized by moderate volume, stable frequency, long duration and no obvious environmental abnormality correlation, and is often accompanied by slight body movement but no distortion, and no painful expression on the face; then the module fuses the four types of crying classification results and the synchronously acquired environmental data through a weighted fusion algorithm, and generates an abnormal demand identifier containing crying type information, environmental data characteristics and abnormality level based on a pre-set abnormality judgment threshold.

[0109] The image posture decision module takes the abnormal demand identifier as a trigger condition, extracts features from the surrounding image sequences frame by frame through a convolutional neural network, calculates the deviation of the key point coordinates of the new-born's body and face (the deviation calculation uses the aforementioned formula), determines whether there are abnormal posture information such as body distortion and abnormal facial expression, and then combines the crying classification results and environmental data for multi-modal fusion to obtain a comprehensive decision basis containing demand type, emergency level and confidence.

[0110] The instruction iteration planning module generates a targeted response instruction sequence based on the comprehensive decision basis, specifically including:

[0111] If the classification result is hunger type crying, and there is no significant abnormality in the environmental data and no abnormal posture in the posture analysis, a notification instruction is generated to "remind the caregiver to perform the feeding operation within 30 minutes", and the instruction parameters include notification priority (medium) and repeated reminder interval (10 minutes).

[0112] If the classification result is discomfort type crying, combined with environmental data (such as high humidity) or posture analysis (such as slight body distortion), a combined instruction is generated to "automatically play soothing white noise (volume gradient value 50-60 dB)" and "remind the caregiver to check and replace the diaper", and the environmental temperature and humidity are simultaneously adjusted to the pre-set appropriate range (temperature 24-26℃, humidity 50%-60%), and the instruction parameters include the playing time (15 minutes) and the temperature adjustment step size (0.5℃ / time).

[0113] If the classification result is a pain type cry, and the posture analysis shows limb distortion and a painful facial expression, a "high-priority alarm notification for nursing staff to immediately check the newborn's physical condition" instruction is generated, and the cry frequency time sequence fluctuation information, posture abnormality characteristics, and environmental data are recorded synchronously to provide data support for the auxiliary identification of early symptoms such as intestinal distension and skin inflammation. The instruction parameters include alarm volume gradient value (70-80 dB) and hierarchical monitoring node density (1 monitoring node per 30 seconds).

[0114] If the classification result is a companion-seeking type cry, and the posture analysis is normal and the environmental data is normal, a "notify nursing staff to perform soothing and lullaby within 15 minutes" notification instruction is generated, which can be linked to play soft lullaby music. The instruction parameters include execution frequency adaptive interval (5 minutes / time) and adaptive feedback period (20 minutes).

[0115] After the response instruction sequence is executed, the real-time nursing state data of the newborn (cry intensity decay value, limb posture recovery degree, environmental parameter adjustment compliance rate, and physiological sign stability value) is collected through the monitoring device, which is compared with the preset nursing compliance threshold to divide the deviation level into mild, moderate, and severe, and match the dynamic adjustment coefficient. For example, after the execution of the instruction corresponding to the hunger type cry, if the cry intensity does not decay (moderate deviation), the notification priority is increased to high, and the repeated reminder interval is shortened to 5 minutes. Then, combined with the historical nursing data of the newborn (such as the response effect of past feeding, the soothing time of companion-seeking type cry) and the current deviation level, the response instruction parameters are optimized to generate a nursing execution plan containing dynamic execution time, hierarchical monitoring nodes, and adaptive feedback period. If the adaptation degree of the plan elements to the core parameters of the updated response instruction sequence meets the preset threshold range, the final executable plan is output.

[0116] This embodiment adds companion-seeking type cry classification and targeted nursing measures, and each type of nursing measure is deeply associated with cry type, environmental data, and posture information, which not only improves the accuracy and practicality of nursing response, but also provides clear execution basis for nursing staff. At the same time, through data recording to assist early symptom identification, the core value of intelligent nursing is further strengthened.

[0117] It should be understood that, although the steps in the flowcharts related to the embodiments described above are shown in sequence according to the arrows, the steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, the execution of the steps is not strictly limited in sequence, and the steps can be executed in other sequences. Moreover, at least some of the steps in the flowcharts related to the embodiments described above can include multiple steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution sequence of the steps or stages is not necessarily sequential, but can be executed in rotation or alternation with at least some of the other steps or the steps or stages in the other steps.

[0118] In an embodiment, a computer device is provided, comprising a memory and a processor, the memory storing a computer program, and the processor implementing the steps of the method for a new-born baby intelligent care system based on cry recognition as described above when executing the computer program.

[0119] In an embodiment, a computer readable storage medium is provided, storing a computer program, and the computer program implementing the steps of the method embodiments described above when executed by a processor.

[0120] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts are described with reference to the parts of the method embodiments. The device embodiments described above are merely illustrative, and the components described as separate components can or can not be physically separated, and the components displayed as units can or can not be physical units, i.e., they can be located in one place, or distributed on multiple network units. Some or all of the modules can be selected to achieve the purposes of the present disclosure according to actual needs. Those skilled in the art can understand and implement it without creative labor.

[0121] The above-described embodiments only express several implementation manners of the present application, and the description is relatively specific and detailed, but it should not be understood as a limitation on the patent scope of the application. It should be pointed out that, for those skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are all within the protection scope of the present application.

Claims

1. A new-born intelligent care system based on cry recognition, characterized in that, The system comprises: a real-time data acquisition module for acquiring real-time sound signals of newborns, environmental data and surrounding image sequences; an abnormal voiceprint determination module for filtering the real-time sound signals to extract feature parameters and obtaining a crying classification result by using a support vector machine, and fusing the environmental data to obtain an abnormal demand identifier; an image posture decision module for analyzing the posture of the surrounding image sequences based on the abnormal demand identifier to obtain abnormal posture information of the newborns, combining the crying classification result and the environmental data to obtain a comprehensive decision basis for judging emergency demands; an instruction iteration planning module for generating a response instruction sequence according to the comprehensive decision basis and updating the response instruction sequence based on feedback data after execution, and fusing historical care data and a current deviation level to generate a care execution plan.

2. The system of claim 1, wherein, The abnormal voiceprint determination module is further configured to: filter the acquired real-time sound signals of newborns to obtain filtered sound signal data; extract sound feature parameters including volume, frequency and duration from the sound signal data using a feature extraction algorithm; input the extracted sound feature parameters into a preset support vector machine algorithm model to obtain a classification result of the crying type of the newborns by model operation; the classification result includes a hunger type of crying, a pain type of crying and an uncomfortable type of crying; synchronously acquire environmental data of the environment in which the newborns are located; the environmental data includes temperature data, humidity data, illumination intensity data and sound decibel data; fuse the environmental data and the classification result of the crying type using a weighted fusion algorithm to obtain a fusion result, judge whether there is an environment-related abnormal demand based on a preset abnormality judgment threshold, and obtain a binary judgment result; if the binary judgment result is that there is the environment-related abnormal demand, an abnormal demand identifier corresponding to the environment-related abnormal demand is generated; the abnormal demand identifier includes crying type information, environmental data features and an abnormality level.

3. The system of claim 1, wherein, The image posture decision module is further configured to: extract features from the acquired surrounding image sequences frame by frame using a convolutional neural network to obtain image feature data containing texture, contour and target region; input the image feature data into a posture analysis model to calculate the coordinate deviation of the key points of the limbs and face of the newborns and compare them with a standard posture template to determine the abnormal posture information of the newborns; the abnormal posture information includes limb distortion and abnormal facial expression; integrate the abnormal posture information, the classification result of the crying type and the corresponding environmental data using a multi-modal data fusion algorithm to obtain an emergency demand judgment result; if the confidence level of the emergency demand judgment result exceeds a preset threshold, the dynamic change information of the posture key points of the newborns, the frequency time fluctuation information of the crying signal and the instantaneous abnormal fluctuation information of the environmental data are extracted respectively to generate a comprehensive decision basis including the demand type and the emergency level.

4. The system of claim 3, wherein, The coordinate deviation of the key points of the limbs and face of the newborns is calculated by the following formula: wherein, denotes the newborn in the frame image comprehensive coordinate deviation value of the group of key points, denotes the actual pixel coordinates of the key point in the frame denotes the reference pixel coordinates of the key point in the preset standard posture template, denotes the weight coefficient of the key point, denotes the total number of key points to be detected in a single frame image, denotes the inter-frame time decay coefficient, denotes the acquisition time stamp of the frame and the frame .​ 5. The system of claim 1, wherein, The instruction iteration planning module is further configured to: generate a response instruction sequence according to the comprehensive decision basis, and obtain neonatal real-time care state data by executing the response instruction sequence; the care state data includes a crying sound intensity decay value, a limb posture recovery degree, an environmental parameter adjustment compliance rate, and a physiological sign stability value; compare the care state data with a preset care compliance threshold, divide a deviation level according to a deviation degree, and match a corresponding dynamic adjustment coefficient; adjust parameters of the response instruction sequence according to the dynamic adjustment coefficient to obtain an updated response instruction sequence; the parameters include an alarm volume gradient value, a temperature adjustment step amplitude, and an execution frequency adaptive interval; fuse and analyze the updated response instruction sequence, neonatal historical care data, and a current deviation level to generate a care execution plan; the historical care data includes past abnormal response effects and physiological rhythm characteristics; the care execution plan includes a dynamic execution time, a hierarchical monitoring node, and an adaptive feedback cycle.

6. The system of claim 5, wherein, The method comprises: obtaining real-time sound signals, environmental data, and surrounding image sequences of a neonate; extracting feature parameters from the real-time sound signals after filtering, obtaining a crying sound classification result by using a support vector machine, and obtaining an abnormal demand identifier by fusing the environmental data; performing posture analysis on the surrounding image sequences based on the abnormal demand identifier to obtain neonatal abnormal posture information, and obtaining a comprehensive decision basis for judging an emergency demand by combining the crying sound classification result and the environmental data; generating a response instruction sequence according to the comprehensive decision basis, updating the response instruction sequence based on feedback data after execution, and generating a care execution plan by fusing historical care data and a current deviation level. 7.A method for intelligent neonatal care based on cry recognition, characterized in that, The processor executes the computer program to implement the steps of the system in any one of claims 1 to 6. The computer program is executed by the processor to implement the steps of the system in any one of claims 1 to 6. ​ ​ ​ 8. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that, ​ 9. A computer readable storage medium having stored thereon a computer program, characterized in that, ​

Images