Cloud computing data processing platform for mental health of teenagers

Abstract

The invention discloses a cloud computing data processing platform for mental health of teenagers, and relates to the technical field of data processing, and the platform comprises a data collection module which collects teenager multi-modal data, carries out the standardization processing, and carries out the facial feature extraction and voice feature extraction based on the collected data; and the emotion analysis module is used for constructing a double-branch deep neural network model to respectively calculate an enthusiasm score and an awakening degree score. According to the method, the enthusiasm score and the awakening degree score can be calculated through the double-branch structure of the model, the enthusiasm score reflects the positive and negative polarities of the emotion and helps to recognize the direction of the emotion, and the awakening degree score represents the intense degree of the emotion and helps to recognize the intensity of the emotion; through fuzzy logic rule definition and multiple emotion category division, the system can be more flexible when judging the emotion of the teenager, emotion data processing is distributed on different nodes through a three-level architecture, the requirements of different numbers of users are met, and the processing efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a cloud computing data processing platform for adolescent mental health. Background Art

[0002] In the context of the rapid development of information technology today, adolescent mental health issues have gradually attracted great attention from all walks of life. With the accelerated pace of modern life and the superposition of multiple pressures such as academic, social and family, the psychological problems of adolescents are becoming more complex and diversified.

[0003] Traditional emotion recognition systems usually process data on a single device, making it difficult for the system to support real-time emotion monitoring and long-term trend analysis of emotion data. This is especially true in the application scenario of adolescent mental health. It is inconvenient to perform more sophisticated emotion cycle identification and event correlation analysis based on the complex correlation between emotional fluctuations and life events. The lack of effective emotion feature extraction and cross-modal data fusion mechanisms makes it difficult to achieve high-precision, multi-dimensional emotion analysis in emotion judgment, which hinders the accurate analysis and management of adolescent mental health. Summary of the invention

[0004] In view of the above-mentioned problems existing in the existing cloud computing data processing platform for adolescent mental health, the present invention is proposed.

[0005] Therefore, the problem to be solved by the present invention is that traditional emotion recognition systems usually perform data processing on a single device, which makes it difficult for the system to support real-time emotion monitoring and long-term trend analysis of emotion data. Especially in the application scenario of adolescent mental health, it is inconvenient to urgently need more sophisticated emotion cycle identification and event correlation analysis based on the complex correlation between emotion fluctuations and life events. The lack of effective emotion feature extraction and cross-modal data fusion mechanism makes it difficult to achieve high-precision and multi-dimensional emotion analysis in emotion judgment, which hinders the accurate analysis and management of adolescent mental health.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: a cloud computing data processing platform for adolescent mental health, comprising:

[0007] The data collection module collects multimodal data of teenagers, performs standardization processing, and extracts facial features and voice features based on the collected data;

[0008] The emotion analysis module builds a two-branch deep neural network model to calculate the positivity score and arousal score respectively, calculates the valence and emotional arousal values, and defines fuzzy logic rules to map to a two-dimensional plane to judge the emotions of adolescents;

[0009] The data distribution module uses a three-level cloud architecture to deploy models and store data;

[0010] The emotion cycle analysis module calculates the emotion intensity based on a two-dimensional plane, recursively decomposes the emotion intensity sequence using empirical mode decomposition, and identifies the FFT main frequency of the low-frequency IMF to calculate the main cycle;

[0011] The event correlation analysis module calculates the correlation between adolescent life events and low-frequency IMFs, determines positively correlated adolescent events, and traces back the corresponding emotional judgments;

[0012] The data storage module encrypts and stores the identification and judgment data.

[0013] As a preferred solution of the cloud computing data processing platform for adolescent mental health described in the present invention, wherein: the collection of adolescent multimodal data and standardization processing include:

[0014] Collect multimodal data of adolescents, including facial expressions and voice data;

[0015] Use video cameras to collect facial expressions in real time, and use audio equipment to simultaneously collect voice features;

[0016] A synchronous clock mechanism is used to align time series and perform data standardization.

[0017] As a preferred solution of the cloud computing data processing platform for adolescent mental health of the present invention, wherein: the facial feature extraction and voice feature extraction based on the collected data include:

[0018] Based on the facial expression data of teenagers, for each frame of facial image, according to the position coordinates of the selected facial feature points, the Euclidean distance of the displacement of facial feature points between different frames is calculated, and the calculated distance values ​​are combined into a facial feature vector;

[0019] The synchronously collected and cloud data are windowed and FFT is performed to extract frequency domain features.

[0020] The frequency domain features are filtered using a Mel filter, the cepstral coefficients are calculated as emotional features, and the MFCC feature values ​​of each frame are combined into a speech feature vector.

[0021] As a preferred solution of the cloud computing data processing platform for adolescent mental health described in the present invention, wherein: the construction of a dual-branch deep neural network model calculates the positivity score and the arousal score respectively, and calculates the valence and emotional arousal value, including:

[0022] Different facial feature areas are divided based on the face parts, and a two-branch deep neural network model is constructed, including input layer, convolution layer, LSTM layer and output layer;

[0023] The input layer takes facial features and speech features as multimodal data input, and the convolution layer performs spatial feature extraction on the input features to extract local features of facial expressions and speech;

[0024] The LSTM layer processes the time series relationship of features, captures the continuous dynamic changes of facial expressions and the fluctuations of voice emotions, and performs dual-branch emotion scoring. The LSTM layer predicts the output of the positive and negative polarity vectors of emotions through the time series, and outputs the positivity score through the Softmax activation function to indicate the direction of the emotion, which is expressed as:

[0025] Sj = Soft(Wj·P+bj);

[0026] Where Sj represents the positivity score, Soft represents the Softmax activation function, P represents the feature vector output by the convolutional layer, Wj and bj represent the weight and bias of the positivity branch respectively;

[0027] The goal of the branch is to extract the intensity of facial and voice features and calculate the arousal of emotions, expressed as:

[0028] S h =ReLU(W h ·P+b h );

[0029] Where Sh represents the arousal score, ReLU represents the ReLU activation function, Wh and bh represent the weight and bias of the arousal branch respectively;

[0030] The output layer outputs the positivity score and the arousal score;

[0031] A multimodal dataset containing multiple emotion samples is used for model training. The cross entropy loss function is selected to calculate the difference between the model's category probability and the actual label. The Adam optimizer is used for gradient descent optimization. The model parameters including weights and bias terms are updated. If the model loss no longer decreases significantly during continuous iterations, the iteration is stopped and the model parameters are output to update the model.

[0032] Based on the positivity scores and arousal scores of different facial features and voice features, the facial emotion valence and voice valence are calculated respectively, expressed as:

[0033]

[0034] Where N and M represent the number of facial features and voice features, respectively, Xf and Xv represent the facial emotion valence and voice valence, respectively, Sj,f,i and Sj,v,k represent the positivity score of the i-th facial feature and the k-th voice feature, respectively;

[0035] The intensity of the overall emotion is extracted based on the facial and voice arousal scores and is used as the emotional arousal value, expressed as:

[0036]

[0037] where Hf and Hv represent the emotional arousal values ​​of facial emotional valence and voice valence, respectively, and Sh,f,i and Sh,v,k represent the arousal score of the i-th facial feature and the arousal score of the k-th voice feature, respectively.

[0038] As a preferred solution of the cloud computing data processing platform for adolescent mental health of the present invention, wherein: the definition of fuzzy logic rules is mapped to a two-dimensional plane to judge the emotions of adolescents, including:

[0039] Define fuzzy logic rules based on calibrated historical data;

[0040] The fused valence and arousal values ​​are calculated by the fuzzy operator based on fuzzy logic rules and expressed as:

[0041]

[0042] Among them, Xr and Hr represent the final fusion valence and fusion arousal, respectively;

[0043] Based on the calculated data of the fused arousal level, the sum of the mean and the standard deviation is calculated as the arousal level threshold;

[0044] The fusion valence and fusion arousal are standardized and converted into coordinate positions in a two-dimensional plane, wherein the fusion valence is used as the x-coordinate in the two-dimensional plane, the arousal threshold of the fusion arousal is used as the y-coordinate origin in the two-dimensional plane, and the fusion arousal greater than the arousal threshold is used as the positive number of the y-coordinate, and the fusion arousal less than the arousal threshold is used as the negative number of the y-coordinate. Based on the mapping of the fusion valence and the fusion arousal in the two-dimensional plane coordinates, the emotional area is divided on the two-dimensional plane;

[0045] Based on the four quadrants of the two-dimensional plane coordinates, corresponding to the emotions of "calm positive", "anxious negative", "active excitement" and "indifferent calm", the emotions of adolescents were judged according to the fusion valence and fusion arousal.

[0046] As a preferred solution of the cloud computing data processing platform for adolescent mental health described in the present invention, the three-level cloud architecture is used to deploy models and store data, including:

[0047] Based on the collection of multimodal data of adolescents, a three-level cloud architecture is adopted, including edge nodes, local cloud centers and central cloud servers for cloud computing data transmission;

[0048] The edge nodes will perform preliminary processing on the emotional data collected in real time, and the local cloud center will perform secondary processing and storage on the data from the edge nodes, and transmit the processed data to the central cloud server. The central cloud server executes the dual-branch deep neural network model and judges the emotions of teenagers.

[0049] As a preferred solution of the cloud computing data processing platform for adolescent mental health described in the present invention, wherein: the emotion intensity is calculated based on a two-dimensional plane, the emotional intensity sequence is recursively decomposed using empirical mode decomposition, and the FFT main frequency of the low-frequency IMF is identified to calculate the main period, including:

[0050] Based on the x-coordinate and y-coordinate of the coordinate in the two-dimensional plane, the emotion intensity Sq is calculated, which is expressed as:

[0051]

[0052] Based on the emotion data at fixed time intervals, the emotion intensity sequence X(t) is constructed. The emotional intensity sequence is recursively decomposed using the empirical mode decomposition (EMD) to obtain several intrinsic mode function (IMF) components and residual terms, which are expressed as:

[0053]

[0054] Where IMFk(t) is the kth IMF component, representing fluctuations of different frequencies, r(t) is the residual term, representing the non-periodic trend in the sequence, and n is the total number of IMF components;

[0055] In the FFT calculation results, the frequency component with the highest energy in the spectrum is identified as the main frequency, and the frequency threshold is determined based on historical data. If the main frequency is greater than or equal to the frequency threshold, it means that the corresponding IMF component IMFk(t) is a high-frequency IMF, indicating short-term fluctuations in emotions. If the main frequency is less than the frequency threshold, it means that the corresponding IMF component IMFk(t) is a low-frequency IMF component, indicating long-term fluctuations in emotions.

[0056] Based on the low-frequency IMF component, the fast Fourier transform FFT is used to perform frequency domain analysis on the low-frequency IMF component, which is expressed as:

[0057]

[0058] Where F(IMFn) represents the frequency domain representation of the nth low-frequency IMF component, f represents the frequency, T represents the total number of emotion sequences in the period, u represents the imaginary unit, 2πft / T represents the phase of frequency f at time t, and ei·2πft / T represents mapping the nth low-frequency IMF component in the time domain to the frequency domain;

[0059] In the FFT calculation result, the frequency component with the highest energy in the spectrum is identified as the main frequency, and the main period is determined based on the inverse of the main frequency.

[0060] As a preferred solution of the cloud computing data processing platform for adolescent mental health of the present invention, wherein: the calculation of the correlation between adolescent life events and low-frequency IMF includes:

[0061] The life events of adolescents in the cycle time are counted and encoded into binary sequence values ​​based on the cycle time. A time series Et corresponding to the cycle time content is created for each event, and the Pearson correlation coefficient is used to calculate the correlation between the event series and the low-frequency IMF.

[0062] As a preferred solution of the cloud computing data processing platform for adolescent mental health of the present invention, the method of judging positively correlated adolescent events and tracing back the corresponding emotional judgments includes:

[0063] Based on the sum of the mean and standard deviation of the correlation of historical data as the corresponding correlation threshold, if the calculated correlation between the event and the low-frequency fluctuation of emotions is greater than or equal to the correlation threshold, it means that there is a significant positive correlation between the event and the low-frequency fluctuation of emotions. For each significantly correlated time point t, the two-dimensional plane coordinates of the emotion intensity sequence at time point t are traced back to further determine the specific emotions triggered by the event.

[0064] As a preferred solution of the cloud computing data processing platform for adolescent mental health of the present invention, wherein: the encrypted storage of identification and determination data includes:

[0065] Combining the life events of adolescents and the specific emotions triggered by them, as well as the main cycle extracted by the low-frequency IMF component of the intensity of emotions, the association between the long-term fluctuation characteristics of emotions and specific periodic events is identified;

[0066] The emotion judgment data, extracted main cycle data and correlation data are diverted by time period and evenly distributed to multiple central cloud server nodes through a load balancer. In the process of data diversion to each node, the TLS protocol is used to encrypt data transmission. On the central cloud server node, the data is encrypted and stored using the AES-256 encryption algorithm.

[0067] The beneficial effects of the present invention are as follows: the dual-branch structure of the model can be used to calculate the positivity score and the arousal score respectively, wherein the positivity score reflects the positive and negative polarity of the emotion and helps to identify the direction of the emotion, and the arousal score represents the intensity of the emotion and helps to identify the strength of the emotion. Through the definition of fuzzy logic rules and the classification of multiple emotion categories, the system can be more flexible in judging the emotions of adolescents. Through the three-level architecture, the processing of emotion data is distributed on different nodes to adapt to the needs of different numbers of users and improve the processing efficiency. By classifying IMF components of different frequencies as high-frequency or low-frequency components, the separation of short-term emotional fluctuations and long-term emotional trends can be achieved, thereby supporting more detailed emotional cycle analysis. Through correlation analysis, the periodic correlation between life events and emotional fluctuations can be identified, which is helpful to formulate emotional management measures before the event occurs and reduce the emotional pressure of adolescents during the event. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0069] Figure 1 This is a schematic diagram of the structure of the cloud computing data processing platform for adolescent mental health.

[0070] Figure 2 Schematic diagram of the process of cloud computing data processing platform for adolescent mental health. DETAILED DESCRIPTION

[0071] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings.

[0072] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0073] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor is it a separate or selective embodiment that is mutually exclusive with other embodiments.

[0074] Example 1, reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, and which provides a cloud computing data processing platform for adolescent mental health. The cloud computing data processing platform for adolescent mental health includes:

[0075] S1, collect multimodal data of adolescents, perform standardization, and extract facial features and voice features based on the collected data;

[0076] Preferably, the multimodal data of adolescents are collected and standardized, including:

[0077] Collect multimodal data of adolescents, including facial expressions and voice data;

[0078] Use video cameras to collect facial expressions in real time, and use audio equipment to simultaneously collect voice features;

[0079] A synchronous clock mechanism is used to align time series and perform data standardization.

[0080] By collecting multimodal data, including facial expressions and voice, we can capture the emotional characteristics and psychological state of adolescents more comprehensively. By using a synchronous clock mechanism to align the time series of different data sources, the system can accurately capture the synchronization between facial expressions and voice, ensuring that each modal data can be analyzed at the same time scale, providing effective technical support for long-term mental health tracking, emotional trend analysis and personalized intervention.

[0081] Further, facial feature extraction and voice feature extraction are performed based on the collected data, including:

[0082] Based on the facial expression data of teenagers, for each frame of facial image, according to the position coordinates of the selected facial feature points, the Euclidean distance of the displacement of facial feature points between different frames is calculated, and the calculated distance values ​​are combined into a facial feature vector;

[0083] The synchronously collected and cloud data are windowed and FFT is performed to extract frequency domain features.

[0084] The frequency domain features are filtered using a Mel filter, the cepstral coefficients are calculated as emotional features, and the MFCC feature values ​​of each frame are combined into a speech feature vector.

[0085] By calculating the Euclidean distance of facial feature points in consecutive frames, the system can quickly extract key information of facial movement and effectively compress the data volume. By extracting MFCC features through FFT and Mel filters, the frequency domain analysis of speech data can be quickly completed to extract key emotional features. By calculating the displacement of facial feature points and extracting the cepstral coefficients of speech features, the system can capture subtle changes in adolescents' emotions. Through the displacement of facial feature points in multiple frames, the system can recognize subtle changes in facial expressions, especially complex expressions (such as anger with anxiety). The MFCC feature vector retains the frequency domain feature information of speech and can accurately reflect emotion-related changes in intonation, speech rate, pitch, etc., supporting the recognition of more complex emotional states.

[0086] S2, construct a two-branch deep neural network model to calculate the positivity score and arousal score respectively, calculate the valence and emotional arousal values, and define fuzzy logic rules to map to a two-dimensional plane for adolescent emotion judgment;

[0087] Preferably, a two-branch deep neural network model is constructed to calculate the positivity score and the arousal score respectively, and to calculate the valence and emotional arousal values, including:

[0088] Different facial feature areas are divided based on the face parts, and a two-branch deep neural network model is constructed, including input layer, convolution layer, LSTM layer and output layer;

[0089] The input layer takes facial features and speech features as multimodal data input, and the convolution layer performs spatial feature extraction on the input features to extract local features of facial expressions and speech;

[0090] The LSTM layer processes the time series relationship of features, captures the continuous dynamic changes of facial expressions and the fluctuations of voice emotions, and performs dual-branch emotion scoring. The LSTM layer predicts the output of the positive and negative polarity vectors of emotions through the time series, and outputs the positivity score through the Softmax activation function to indicate the direction of the emotion, which is expressed as:

[0091] Sj = Soft(Wj·P+bj);

[0092] Where Sj represents the positivity score, Soft represents the Softmax activation function, P represents the feature vector output by the convolutional layer, Wj and bj represent the weight and bias of the positivity branch respectively;

[0093] The goal of the branch is to extract the intensity of facial and voice features and calculate the arousal of emotions, expressed as:

[0094] Sh = ReLU (Wh·P + bh);

[0095] Where Sh represents the arousal score, ReLU represents the ReLU activation function, Wh and bh represent the weight and bias of the arousal branch respectively;

[0096] The output layer outputs the positivity score and the arousal score;

[0097] A multimodal dataset containing multiple emotion samples is used for model training. The cross entropy loss function is selected to calculate the difference between the model's category probability and the actual label. The Adam optimizer is used for gradient descent optimization. The model parameters including weights and bias terms are updated. If the model loss no longer decreases significantly during continuous iterations, the iteration is stopped and the model parameters are output to update the model.

[0098] Based on the positivity scores and arousal scores of different facial features and voice features, the facial emotion valence and voice valence are calculated respectively, expressed as:

[0099]

[0100] Where N and M represent the number of facial features and voice features, respectively, Xf and Xv represent the facial emotion valence and voice valence, respectively, Sj,f,i and Sj,v,k represent the positivity score of the i-th facial feature and the k-th voice feature, respectively;

[0101] The intensity of the overall emotion is extracted based on the facial and voice arousal scores and is used as the emotional arousal value, expressed as:

[0102]

[0103] where Hf and Hv represent the emotional arousal values ​​of facial emotional valence and voice valence, respectively, and Sh,f,i and Sh,v,k represent the arousal score of the i-th facial feature and the arousal score of the k-th voice feature, respectively.

[0104] By taking facial features and voice features as multimodal inputs, the model can integrate multiple sources of emotional signals, effectively capture subtle changes in emotions, and improve the accuracy of identifying complex emotions. Through dynamic analysis of time series, the model can identify gradual changes in emotions, such as the gradual change from relaxation to tension, thereby providing basic support for accurately evaluating emotional states. The dual-branch structure of the model can calculate the positivity score and arousal score respectively. The positivity score reflects the positive and negative polarity of emotions and helps identify the direction of emotions. The arousal score represents the intensity of emotions and helps identify the intensity of emotions. Through these branch outputs, the system can distinguish emotions in multiple dimensions and achieve a more comprehensive and detailed evaluation of emotions.

[0105] By taking a weighted average of the positivity scores of different facial features (such as eyebrows, corners of the mouth, etc.) and voice features, we can get the overall facial and voice emotional valence, which can accurately reflect the overall emotional direction and help track the long-term changing trends of emotions. The intensity of emotions can be obtained through the facial and voice arousal scores, which can distinguish the intensity of emotions, such as excitement and calmness, and facilitate dynamic tracking of adolescents' emotional fluctuations.

[0106] Furthermore, fuzzy logic rules are defined and mapped to a two-dimensional plane for determining adolescent emotions, including:

[0107] Define fuzzy logic rules based on calibrated historical data:

[0108] Rule 1: If the “facial valence” is high and the “voice arousal” is low, the emotion is “calm and positive”;

[0109] Rule 2: If the “facial valence” is low and the “voice arousal” is high, the emotion is “anxious negative”;

[0110] Rule 3: If both “facial valence” and “voice valence” are high, the emotion is “positive excitement”;

[0111] Rule 4: If both "Facial Arousal" and "Voice Arousal" are low, the emotion is "Cold and Calm";

[0112] The fused valence and arousal values ​​are calculated by the fuzzy operator based on fuzzy logic rules and expressed as:

[0113]

[0114] Among them, Xr and Hr represent the final fusion valence and fusion arousal, respectively;

[0115] Based on the calculated data of the fused arousal level, the sum of the mean and the standard deviation is calculated as the arousal level threshold;

[0116] The fusion valence and fusion arousal are standardized and converted into coordinate positions in a two-dimensional plane, wherein the fusion valence is used as the x-coordinate in the two-dimensional plane, the arousal threshold of the fusion arousal is used as the y-coordinate origin in the two-dimensional plane, and the fusion arousal greater than the arousal threshold is used as the positive number of the y-coordinate, and the fusion arousal less than the arousal threshold is used as the negative number of the y-coordinate. Based on the mapping of the fusion valence and the fusion arousal in the two-dimensional plane coordinates, the emotional area is divided on the two-dimensional plane;

[0117] Among them, the first quadrant is high valence and high arousal, indicating that the emotion is "positive excitement", the second quadrant is low valence and high arousal, indicating that the emotion is "positive excitement", the first quadrant is high valence and high arousal, indicating that the emotion is "anxious negative", the third quadrant is low valence and low arousal, indicating that the emotion is "indifferent and calm", and the fourth quadrant is high valence and low arousal, indicating that the emotion is "calm and positive";

[0118] Based on the four quadrants of coordinates in a two-dimensional plane, adolescents' emotions were judged according to fusion valence and fusion arousal.

[0119] Through the definition of fuzzy logic rules and the classification of multiple emotion categories, the system can be more flexible in judging adolescents' emotions. Different from the traditional rigid classification method, fuzzy logic allows dynamic emotion judgment based on the integrated valence and arousal scores, thereby improving the accuracy of emotion judgment, especially in the boundary emotion states (such as the state between positivity and anxiety). By integrating the valence and arousal of face and voice, the system can identify multiple emotional states on a two-dimensional plane. Based on the division of four quadrants, it can distinguish between multiple emotional states such as "calm and positive", "anxious and negative", "positive and excited" and "indifferent and calm", providing multi-dimensional and detailed judgment for emotion analysis.

[0120] S3, which uses a three-level cloud architecture to deploy models and store data;

[0121] Preferably, a three-level cloud architecture is used to deploy models and store data, including:

[0122] Based on the collection of multimodal data of adolescents, a three-level cloud architecture is adopted, including edge nodes, local cloud centers and central cloud servers for cloud computing data transmission;

[0123] The edge nodes will perform preliminary processing on the emotional data collected in real time, and the local cloud center will perform secondary processing and storage on the data from the edge nodes, and transmit the processed data to the central cloud server. The central cloud server executes the dual-branch deep neural network model and judges the emotions of teenagers.

[0124] The processing of emotion data is distributed on different nodes through a three-level architecture, which not only improves processing efficiency, but also improves the fault tolerance of the system. When a node fails, other nodes can still work normally to ensure the continuity of emotion recognition. The deployment of edge nodes and local cloud centers is flexible and supports expansion based on the scale of users to adapt to the needs of different numbers of users. Data is processed and streamlined at each level to ensure that the data transmitted to the central cloud server is highly structured critical data, thereby reducing the demand for bandwidth and improving transmission efficiency.

[0125] S4, calculates the emotion intensity based on the two-dimensional plane, recursively decomposes the emotion intensity sequence using empirical mode decomposition, and identifies the FFT main frequency of the low-frequency IMF to calculate the main period;

[0126] Preferably, the emotion intensity is calculated based on a two-dimensional plane, the emotion intensity sequence is recursively decomposed using empirical mode decomposition, and the FFT main frequency of the low-frequency IMF is identified to calculate the main period, including,

[0127] Based on the x-coordinate and y-coordinate of the coordinate in the two-dimensional plane, the emotion intensity Sq is calculated, which is expressed as:

[0128]

[0129] Based on the emotion data at fixed time intervals, the emotion intensity sequence X(t) is constructed. The emotional intensity sequence is recursively decomposed using the empirical mode decomposition (EMD) to obtain several intrinsic mode function (IMF) components and residual terms, which are expressed as:

[0130]

[0131] Where IMFk(t) is the kth IMF component, representing fluctuations of different frequencies, r(t) is the residual term, representing the non-periodic trend in the sequence, and n is the total number of IMF components;

[0132] In the FFT calculation results, the frequency component with the highest energy in the spectrum is identified as the main frequency, and the frequency threshold is determined based on historical data. If the main frequency is greater than or equal to the frequency threshold, it means that the corresponding IMF component IMFk(t) is a high-frequency IMF, indicating short-term fluctuations in emotions. If the main frequency is less than the frequency threshold, it means that the corresponding IMF component IMFk(t) is a low-frequency IMF component, indicating long-term fluctuations in emotions.

[0133] Based on the low-frequency IMF component, the fast Fourier transform FFT is used to perform frequency domain analysis on the low-frequency IMF component, which is expressed as:

[0134]

[0135] Where F(IMFn) represents the frequency domain representation of the nth low-frequency IMF component (used to analyze the energy of different frequencies), f represents the frequency, T represents the total number of emotion sequences in the period, u represents the imaginary unit, 2πft / T represents the phase of frequency f at time t, and ei·2πft / T represents mapping the nth low-frequency IMF component in the time domain to the frequency domain;

[0136] In the FFT calculation result, the frequency component with the highest energy in the spectrum is identified as the main frequency, and the main period is determined based on the inverse of the main frequency.

[0137] The emotion intensity method calculated by two-dimensional coordinates is more accurate than a single data source, and can fully reflect the overall state of emotion, providing a more stable foundation for subsequent emotion fluctuation analysis and cycle identification. By classifying IMF components of different frequencies into high-frequency or low-frequency components, it is possible to separate short-term emotion fluctuations (such as sudden emotion changes) and long-term emotion trends (such as long-term depression or excitement), thereby supporting more detailed emotion cycle analysis. By performing fast Fourier transform (FFT) analysis on low-frequency IMF components, the system can identify the main frequency of the emotion intensity sequence and further determine the main cycle. The identification of the main cycle helps monitor long-term emotion changes and can help identify periodic emotion change patterns, such as weekly or monthly emotion fluctuations. This is of great significance for emotional management and mental health intervention for adolescents, and helps to timely discover and predict emotional risks. The system can identify the main cycle based on the FFT analysis results and use it to monitor the periodic characteristics of emotional fluctuations. By analyzing these periodic characteristics, it can help parents, teachers and mental health workers develop more accurate emotion management and intervention plans.

[0138] S5, calculate the correlation between adolescent life events and low-frequency IMF, determine the adolescent events with positive correlation and trace back the corresponding emotion judgment;

[0139] Preferably, the correlation between the adolescent life events and the low-frequency IMF is calculated, including,

[0140] The life events of adolescents in the cycle time are counted, and the life events of adolescents based on the cycle time are encoded into binary sequence values. A time series Et corresponding to the cycle time is created for each event, and the Pearson correlation coefficient is used to calculate the correlation between the event sequence and the low-frequency IMF, which is expressed as:

[0141]

[0142] Rev represents the correlation between events and low-frequency fluctuations of emotions, T represents the total number of data points of emotion sequences and event sequences within the period, IMFo,t represents the tth low-frequency IMF component, μ IMF and μev represent the means of the low-frequency IMF components of event sequence and emotion intensity, respectively.

[0143] By encoding adolescents' life events into binary sequences and calculating the Pearson correlation coefficient between the event sequence and the emotional low-frequency IMF component, the system can quantify the impact of life events on emotional fluctuations. Through correlation analysis, the periodic correlation between life events and emotional fluctuations can be identified. When the system detects that a life event is highly correlated with emotional fluctuations, it can generate an early warning when the event is about to occur, which helps to formulate emotional management measures before the event occurs and reduce the emotional stress of adolescents during the event. Correlation analysis can reveal the periodic laws of emotional fluctuations and link these periodic changes with life events. For example, the system can detect that the main cycle of emotional fluctuations coincides with a certain type of life event (such as monthly exams or social activities).

[0144] Furthermore, we identify positively correlated adolescent events and trace back the corresponding emotional judgments, including:

[0145] Based on the sum of the mean and standard deviation of the correlation of historical data as the corresponding correlation threshold, if the calculated correlation between the event and the low-frequency fluctuation of emotions is greater than or equal to the correlation threshold, it means that there is a significant positive correlation between the event and the low-frequency fluctuation of emotions. For each significantly correlated time point t, the two-dimensional plane coordinates of the emotion intensity sequence at time point t are traced back to further determine the specific emotions triggered by the event.

[0146] By using the sum of the mean and standard deviation of the historical data correlation as the correlation threshold, we can effectively screen out events that have a significant positive correlation with low-frequency fluctuations in emotions. After screening out significantly correlated events, we can further trace back the emotion intensity sequence and view the emotional state on a two-dimensional plane coordinate to identify the specific emotions (such as positivity, anxiety, indifference, etc.) triggered by a specific event at time point t, thereby intuitively understanding how each event affects the emotional state.

[0147] S6, encrypting and storing the identification and determination data;

[0148] Preferably, the identification and determination data are encrypted and stored, including:

[0149] Combining the life events of adolescents and the specific emotions triggered by them, as well as the main cycle extracted by the low-frequency IMF component of the intensity of emotions, the association between the long-term fluctuation characteristics of emotions and specific periodic events is identified;

[0150] The emotion judgment data, extracted main cycle data and correlation data are diverted by time period and evenly distributed to multiple central cloud server nodes through a load balancer. In the process of data diversion to each node, the TLS protocol is used to encrypt data transmission. On the central cloud server node, the data is encrypted and stored using the AES-256 encryption algorithm.

[0151] By combining the life events of teenagers with the specific emotional judgments triggered, the system extracts the main cycle through the low-frequency IMF component of emotional intensity, thereby effectively identifying the long-term fluctuation characteristics of emotions. By shunting the emotional judgment data, main cycle data and correlation data according to time periods, the system can realize the distributed processing of emotional data, and use the load balancer to evenly distribute the data to multiple central cloud server nodes, avoiding the overload of a single node, improving the efficiency of data processing and the system response speed, so that the system can respond quickly and provide emotional monitoring feedback in real time. During the data shunting process, the TLS protocol is used to encrypt data transmission to ensure that the data is not intercepted during network transmission. After arriving at the central cloud server node, the AES-256 encryption algorithm is used to encrypt the data for storage, effectively protecting emotional data and sensitive information, ensuring user privacy and security, and meeting industry standards for data security.

[0152] Example 2

[0153] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0154] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0155] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0156] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used to implement: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0157] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A cloud computing data processing platform for adolescent mental health, characterized by: include, The data collection module collects multimodal data of teenagers, performs standardization processing, and extracts facial features and voice features based on the collected data; The emotion analysis module builds a two-branch deep neural network model to calculate the positivity score and arousal score respectively, calculates the valence and emotional arousal values, and defines fuzzy logic rules to map to a two-dimensional plane to judge the emotions of adolescents; The data distribution module uses a three-level cloud architecture to deploy models and store data; The emotion cycle analysis module calculates the emotion intensity based on a two-dimensional plane, recursively decomposes the emotion intensity sequence using empirical mode decomposition, and identifies the FFT main frequency of the low-frequency IMF to calculate the main cycle; The event correlation analysis module calculates the correlation between adolescent life events and low-frequency IMFs, determines positively correlated adolescent events, and traces back the corresponding emotional judgments; The data storage module encrypts and stores the identification and judgment data.

2. The cloud computing data processing platform for adolescent mental health as claimed in claim 1, characterized in that: The multimodal data of adolescents are collected and standardized, including: Collect multimodal data of adolescents, including facial expressions and voice data; Use video cameras to collect facial expressions in real time, and use audio equipment to simultaneously collect voice features; A synchronous clock mechanism is used to align time series and perform data standardization.

3. The cloud computing data processing platform for adolescent mental health as claimed in claim 2, characterized in that: The facial feature extraction and voice feature extraction based on the collected data include: Based on the facial expression data of teenagers, for each frame of facial image, according to the position coordinates of the selected facial feature points, the Euclidean distance of the displacement of facial feature points between different frames is calculated, and the calculated distance values ​​are combined into a facial feature vector; The synchronously collected and cloud data are windowed and FFT is performed to extract frequency domain features. The frequency domain features are filtered using a Mel filter, the cepstral coefficients are calculated as emotional features, and the MFCC feature values ​​of each frame are combined into a speech feature vector.

4. The cloud computing data processing platform for adolescent mental health as claimed in claim 3, characterized in that: The dual-branch deep neural network model is constructed to calculate the positivity score and the arousal score, and calculate the valence and emotional arousal values, including: Different facial feature areas are divided based on the face parts, and a two-branch deep neural network model is constructed, including input layer, convolution layer, LSTM layer and output layer; The input layer takes facial features and speech features as multimodal data input, and the convolution layer performs spatial feature extraction on the input features to extract local features of facial expressions and speech; The LSTM layer processes the time series relationship of features, captures the continuous dynamic changes of facial expressions and the fluctuations of voice emotions, and performs dual-branch emotion scoring. The LSTM layer predicts the output of the positive and negative polarity vectors of emotions through the time series, and outputs the positivity score through the Softmax activation function to indicate the direction of the emotion, which is expressed as: Sj = Soft(Wj·P+bj); Where Sj represents the positivity score, Soft represents the Softmax activation function, P represents the feature vector output by the convolutional layer, Wj and bj represent the weight and bias of the positivity branch respectively; The goal of the branch is to extract the intensity of facial and voice features and calculate the arousal of emotions, expressed as: Sh = ReLU (Wh·P + bh); Where Sh represents the arousal score, ReLU represents the ReLU activation function, Wh and bh represent the weight and bias of the arousal branch respectively; The output layer outputs the positivity score and the arousal score; A multimodal dataset containing multiple emotion samples is used for model training. The cross entropy loss function is selected to calculate the difference between the model's category probability and the actual label. The Adam optimizer is used for gradient descent optimization. The model parameters including weights and bias terms are updated. If the model loss no longer decreases significantly during continuous iterations, the iteration is stopped and the model parameters are output to update the model. Based on the positivity scores and arousal scores of different facial features and voice features, the facial emotion valence and voice valence are calculated respectively, expressed as: Where N and M represent the number of facial features and voice features, respectively, Xf and Xv represent the facial emotion valence and voice valence, respectively, Sj,f,i and Sj,v,k represent the positivity score of the i-th facial feature and the k-th voice feature, respectively; The intensity of the overall emotion is extracted based on the facial and voice arousal scores and is used as the emotional arousal value, expressed as: where Hf and Hv represent the emotional arousal values ​​of facial emotional valence and voice valence, respectively, and Sh,f,i and Sh,v,k represent the arousal score of the i-th facial feature and the arousal score of the k-th voice feature, respectively.

5. The cloud computing data processing platform for adolescent mental health as claimed in claim 4, characterized in that: The defined fuzzy logic rules are mapped to a two-dimensional plane to determine the emotions of adolescents, including: Define fuzzy logic rules based on calibrated historical data; The fused valence and arousal values ​​are calculated by the fuzzy operator based on fuzzy logic rules and expressed as: Among them, Xr and Hr represent the final fusion valence and fusion arousal, respectively; Based on the calculated data of the fused arousal level, the sum of the mean and the standard deviation is calculated as the arousal level threshold; The fusion valence and fusion arousal are standardized and converted into coordinate positions in a two-dimensional plane, wherein the fusion valence is used as the x-coordinate in the two-dimensional plane, the arousal threshold of the fusion arousal is used as the y-coordinate origin in the two-dimensional plane, and the fusion arousal greater than the arousal threshold is used as the positive number of the y-coordinate, and the fusion arousal less than the arousal threshold is used as the negative number of the y-coordinate. Based on the mapping of the fusion valence and the fusion arousal in the two-dimensional plane coordinates, the emotional area is divided on the two-dimensional plane; The four quadrants based on the two-dimensional plane coordinates correspond to the emotions of "calm positive", "anxious negative", "active excitement" and "indifferent calm", and the emotions of adolescents are judged according to the fusion valence and fusion arousal.

6. The cloud computing data processing platform for adolescent mental health as claimed in claim 5, characterized in that: The three-level cloud architecture is used to deploy models and store data, including: Based on the collection of multimodal data of adolescents, a three-level cloud architecture is adopted, including edge nodes, local cloud centers and central cloud servers for cloud computing data transmission; The edge nodes will perform preliminary processing on the emotional data collected in real time, and the local cloud center will perform secondary processing and storage on the data from the edge nodes, and transmit the processed data to the central cloud server. The central cloud server executes the dual-branch deep neural network model and judges the emotions of teenagers.

7. The cloud computing data processing platform for adolescent mental health as claimed in claim 6, characterized in that: The emotion intensity is calculated based on a two-dimensional plane, and the emotion intensity sequence is recursively decomposed using empirical mode decomposition, and the FFT main frequency of the low-frequency IMF is identified to calculate the main period, including: Based on the x-coordinate and y-coordinate of the coordinate in the two-dimensional plane, the emotion intensity Sq is calculated, which is expressed as: Based on the emotion data at fixed time intervals, the emotion intensity sequence X(t) is constructed. The emotional intensity sequence is recursively decomposed using the empirical mode decomposition (EMD) to obtain several intrinsic mode function (IMF) components and residual terms, which are expressed as: Where IMFk(t) is the kth IMF component, representing fluctuations of different frequencies, r(t) is the residual term, representing the non-periodic trend in the sequence, and n is the total number of IMF components; In the FFT calculation results, the frequency component with the highest energy in the spectrum is identified as the main frequency, and the frequency threshold is determined based on historical data. If the main frequency is greater than or equal to the frequency threshold, it means that the corresponding IMF component IMFk(t) is a high-frequency IMF, indicating short-term fluctuations in emotions. If the main frequency is less than the frequency threshold, it means that the corresponding IMF component IMFk(t) is a low-frequency IMF component, indicating long-term fluctuations in emotions. Based on the low-frequency IMF component, the fast Fourier transform FFT is used to perform frequency domain analysis on the low-frequency IMF component, which is expressed as: Where F(IMFn) represents the frequency domain representation of the nth low-frequency IMF component, f represents the frequency, T represents the total number of emotion sequences in the period, u represents the imaginary unit, 2πft / T represents the phase of frequency f at time t, and ei·2πft / T represents mapping the nth low-frequency IMF component in the time domain to the frequency domain; In the FFT calculation result, the frequency component with the highest energy in the spectrum is identified as the main frequency, and the main period is determined based on the inverse of the main frequency.

8. The cloud computing data processing platform for adolescent mental health as claimed in claim 7, characterized in that: The calculation of the correlation between adolescent life events and low-frequency IMFs includes: The life events of adolescents in the cycle time are counted and encoded into binary sequence values ​​based on the cycle time. A time series Et corresponding to the cycle time content is created for each event, and the Pearson correlation coefficient is used to calculate the correlation between the event series and the low-frequency IMF.

9. The cloud computing data processing platform for adolescent mental health as claimed in claim 8, characterized in that: The positively correlated adolescent events are judged and the corresponding emotional judgments are traced back. include, Based on the sum of the mean and standard deviation of the correlation of historical data as the corresponding correlation threshold, if the calculated correlation between the event and the low-frequency fluctuation of emotions is greater than or equal to the correlation threshold, it means that there is a significant positive correlation between the event and the low-frequency fluctuation of emotions. For each significantly correlated time point t, the two-dimensional plane coordinates of the emotion intensity sequence at time point t are traced back to further determine the specific emotions triggered by the event.

10. The cloud computing data processing platform for adolescent mental health as claimed in claim 9, characterized in that: The encrypted storage of identification and determination data includes: Combining the life events of adolescents and the specific emotions triggered by them, as well as the main cycle extracted by the low-frequency IMF component of the intensity of emotions, the association between the long-term fluctuation characteristics of emotions and specific periodic events is identified; The emotion judgment data, extracted main cycle data and correlation data are diverted by time period and evenly distributed to multiple central cloud server nodes through a load balancer. In the process of data diversion to each node, the TLS protocol is used to encrypt data transmission. On the central cloud server node, the data is encrypted and stored using the AES-256 encryption algorithm.

Images