Psychological disease early warning method and system based on artificial intelligence

Abstract

The invention relates to the field of psychological disease early warning, and discloses a psychological disease early warning method and system based on artificial intelligence, and the method comprises the steps: carrying out the feature extraction and data fusion of collected and preprocessed physiological and behavior data, carrying out the emotion analysis of the emotional state of a user through the fused data, and carrying out the early warning of psychological diseases. The method comprises the following steps: preliminarily judging whether individualized emotion is abnormal or not, comparing physiological features and behavior features of abnormal emotion according to an analysis result of emotion abnormality, performing time sequence analysis on the physiological features and behavior features of known psychological disease symptoms, detecting whether a user has psychological diseases or not, and further evaluating and analyzing the severity of the diseases. According to the method and the system, the disease is subjected to graded early warning, so that the user is helped to realize own emotion change, the psychological disease is discovered and intervened as soon as possible, the analysis and detection of the psychological state are more accurate and reliable, the detection efficiency of the psychological disease is improved, and the disease risk is reduced.

Description

Technical Field

[0001] The present invention relates to the field of mental illness early warning technology, and more specifically to a mental illness early warning method and system based on artificial intelligence. Background Art

[0002] In recent years, East Asian technology has shown unique advantages in the field of AI-driven mental illness early warning. The "KR102021009876A patent" proposed by the Seoul National University team in South Korea has achieved fine-grained recognition of depressive tendencies by building a joint speech-text embedding model, and its F1 value is 12% higher than that of traditional methods. The "JP2020156789A patent" of the RIKEN Institute of Physical and Chemical Research in Japan innovatively introduces brain waves. As a biomarker, wave power ratio provides a neurophysiological basis for the objective assessment of anxiety disorders. These patented technologies demonstrate the great potential of artificial intelligence in the field of early warning and diagnosis of mental illness. However, the use of artificial intelligence technology in China to prevent mental illness is a long way to go. The prediction of mental illness in China is more focused on behavioral data and weak intervention in daily scenarios. It mainly analyzes the user's behavioral data and mental health data through deep learning technology, thereby identifying and detecting the user's mental illness; However, the above process still has the following disadvantages: First, the existing early warning methods for mental illness lack the ability to integrate multi-source data and the complete process of data collection, processing, analysis and early warning, which results in inaccurate and unreliable analysis and detection of mental states, thus reducing the efficiency of mental illness detection. Second, the existing mental illness warning methods only collect a large amount of mental health data and directly use the mental health data for the prediction of mental illness. They lack the ability to first observe the abnormal changes in the user's emotions, and then remind the user of abnormal emotional fluctuations, and then detect the mental illness based on the abnormal emotional changes. They cannot help users realize that abnormal emotional changes may lead to mental illness, resulting in the inability to detect whether the user suffers from a mental illness early and accurately. Summary of the invention

[0003] In order to overcome the above-mentioned defects of the prior art, the present invention provides a mental illness early warning method and system based on artificial intelligence to solve the problems existing in the above-mentioned background technology.

[0004] The present invention provides the following technical solution: a mental illness early warning method based on artificial intelligence, comprising: S1: used to collect the user's physiological data and behavioral data, and upload the collected physiological data and behavioral data to the cloud database for data preprocessing; S2: Create a comprehensive mental state image by extracting features from the preprocessed physiological and behavioral data and then fusing the extracted features; S3: Based on the fused data, the deep learning model is used to analyze the user's emotional state, and a preliminary judgment is made as to whether the individual emotion is abnormal, and abnormal emotions are fed back to the user end; S4: Based on the analysis results of abnormal emotions, the physiological and behavioral characteristics of abnormal emotions are compared with the physiological and behavioral characteristics of known mental illness symptoms for time series analysis to obtain the disease assessment coefficient, which is used to detect whether the user has a mental illness; S5: Based on the disease detection results, further evaluate and analyze the severity of the disease, calculate the severity assessment coefficient, and issue a graded warning for the disease.

[0005] Preferably, the S1 registers a user account by inputting the user's personal information in the wearable device, binds the sensor in the wearable device to the user's account information, and then guides the user to wear the wearable device correctly, so as to monitor the user's physiological data and behavioral data in real time, wherein the physiological data includes heart rate, skin resistance, respiratory rate and body temperature, and the behavioral data includes sleep pattern, amount of exercise, frequency of social activities, and mobile phone usage habits, and then performs data preprocessing operations including: synchronizing the physiological data and behavioral data to a unified time reference, data cleaning, and data standardization, and the preprocessing time complexity is O(N), where N is the amount of data.

[0006] Preferably, S2 extracts physiological characteristics and behavioral characteristics from the preprocessed physiological data and behavioral data, wherein the physiological characteristics include heart rate variability characteristics, skin electrical activity characteristics, and respiratory periodic change characteristics, and the behavioral characteristics include activity level characteristics, social interaction frequency characteristics, and sleep quality index characteristics, merges the physiological characteristics and behavioral characteristics into a feature matrix to form a feature set, and uses PCA to reduce the dimension and fuse the merged feature set; The specific steps of using PCA to reduce the dimension and fuse the merged feature set are: Step S211: Calculate the covariance matrix between each pair of features in the feature matrix; Step S212: performing eigendecomposition on the covariance matrix and obtaining eigenvalues ​​and eigenvectors; Step S213: sorting the eigenvalues ​​in descending order and selecting the eigenvectors corresponding to the first few eigenvalues ​​as principal components according to the eigenvalues; Step S214: reconstructing the data using the selected principal components to obtain a new feature set, that is, a fused feature set, and using the fused feature set to create a mental state image; By mapping the fused features to the representation space of psychological states, the feature set is converted into a psychological state image using data visualization technology. In the psychological state image, the typical emotion distribution can be identified through cluster analysis, and the movement trajectory of the individual in the representation space can be dynamically visualized to capture the changing trend of the user's psychological state.

[0007] Preferably, S3 uses the principal component feature vector of the psychological state image as the input feature of the deep learning model, and outputs a probability distribution of different emotional states, and calculates the formula for the probability value of the emotional state as follows: , represents the original score of the emotion category corresponding to the i-th neuron in the output layer, represents the jth raw score when traversing all emotion categories, M represents the total number of emotion categories in the output layer, e represents the base of the natural logarithm, and the emotional state of the current user is predicted based on the emotional state probability value; By analyzing and calculating the mean and standard deviation of the probability distribution of each emotion category in the normal state, the emotion probability threshold is set for each emotion category based on the mean and standard deviation. The specific calculation formula is: ,in, represents the emotion probability threshold of the i-th emotion category, represents the average probability value of all normal samples in the i-th emotion category, represents the standard deviation of the probability value of the normal sample in the i-th emotion category, and k represents a constant used to adjust the strictness of the threshold. The corresponding emotional state probability threshold Comparison is used to determine whether there is an abnormality in a certain emotion category of the user; if , it is considered that the i-th emotion category is abnormal, and then the emotion state probability values ​​of all emotion categories and the corresponding emotion state probability thresholds are analyzed and calculated. The specific calculation formula is: , where L represents the cumulative probability of emotion, represents the weight coefficient of the i-th emotion category, which is calculated by dividing the emotion cumulative probability L by the preset cumulative probability threshold Compare and preliminarily judge whether individual emotions are abnormal. When the abnormal emotion judgment result is given to the user, the user is considered to have an abnormal emotion, and the abnormal emotion judgment result is fed back to the user to warn that the user's emotion change is abnormal.

[0008] Preferably, S4 uses emotion recognition technology to identify the abnormal emotion period of the user based on the judgment result of abnormal emotion, and extracts the physiological characteristics and behavioral characteristics of the abnormal emotion period, and compares the physiological characteristics and behavioral characteristics of the abnormal emotion with the physiological characteristics and behavioral characteristics of known psychological disease symptoms in a time series, and then calculates the disease assessment coefficient P according to the time series comparison analysis result; By setting a disease threshold , the disease assessment coefficient P and the disease threshold Comparison is used to detect whether the user has a mental illness; when When the user is considered to have a high possibility of mental illness, the test result is transmitted to S5 for further analysis of the severity of the disease. When the user has a mental illness, it is considered that the possibility of the user having a mental illness is small, and the physiological and behavioral characteristics of the user's abnormal emotions will continue to be monitored.

[0009] Preferably, S5 analyzes the severity assessment coefficient based on the disease detection analysis result, and divides the severity of the disease into different levels according to the severity assessment coefficient, and the divided levels are: low risk, medium risk and high risk, and calculates the severity assessment coefficient Z by comparing and analyzing the abnormality of physiological characteristics of physiological signals in the abnormal emotional period and physiological signals in the normal state, and the abnormality of behavioral characteristics of behavioral patterns in the abnormal emotional period and the individual's regular behavioral patterns; Set low-risk thresholds based on severity of disease and high risk threshold ,when When When When , it indicates that the severity of the disease is in the high-risk range, and the severity of the disease is sent to the user end for early warning according to the risk level.

[0010] To achieve the above object, the present invention provides the following technical solution: a mental illness early warning system based on artificial intelligence, implementing the above mental illness early warning method based on artificial intelligence, comprising: Data collection module: used to collect the user's physiological data and behavioral data, and upload the collected physiological data and behavioral data to the cloud database for data preprocessing; Feature fusion module: extracts features from preprocessed physiological and behavioral data, and then fuses the extracted features to create a comprehensive mental state image; Emotional analysis module: Based on the fused data, the deep learning model is used to analyze the user's emotional state, preliminarily determine whether the individual emotion is abnormal, and provide feedback prompts of abnormal emotions to the user end; Disease detection module: Based on the analysis results of abnormal emotions, the physiological and behavioral characteristics of abnormal emotions are compared with the physiological and behavioral characteristics of known psychological disease symptoms for time series analysis to obtain the disease assessment coefficient, which is used to detect whether the user has a psychological disease; Disease severity assessment module: Based on the disease detection results, further evaluate and analyze the severity of the disease, calculate the severity assessment coefficient, and provide graded warnings for the disease.

[0011] Technical effects and advantages of the present invention: The present invention creates a comprehensive mental state image by collecting and preprocessing the physiological data and behavioral data of the user, and extracting features and fusing data on the preprocessed physiological data and behavioral data. The present invention performs emotion analysis on the emotional state of the user through the fused data, preliminarily determines whether the individualized emotion is abnormal, and feeds back the abnormal emotion to the user end. Through the analysis results of abnormal emotions, the physiological characteristics and behavioral characteristics of the abnormal emotions are compared, and time series analysis is performed with the physiological characteristics and behavioral characteristics of known psychological disease symptoms to obtain a disease assessment coefficient, which is used to detect whether the user has a psychological disease. Through the disease detection results, the severity of the disease is further evaluated and analyzed, and the severity assessment coefficient is calculated to perform graded warnings on the disease. The present invention can fuse multi-source data and execute a complete data analysis and warning process, so that the analysis and detection of the mental state are more accurate and reliable, and the detection efficiency of psychological diseases is further improved. By observing the abnormal changes in the user's emotions, the psychological diseases are further detected based on the abnormal emotional changes, which helps to help the user realize his or her emotional changes and take corresponding measures. It can more accurately determine whether the user may suffer from a psychological disease, which is helpful for early detection and intervention, and reduces the risk of disease. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 It is a diagram of the method steps of the present invention.

[0013] Figure 2 It is a system structure block diagram of the present invention. DETAILED DESCRIPTION

[0014] The technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the present invention. In addition, the forms of the various structures recorded in the following embodiments are only examples. The artificial intelligence-based mental illness early warning method and system involved in the present invention are not limited to the various structures recorded in the following embodiments. All other implementations obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0015] like Figure 1 This embodiment provides a method for early warning of mental illness based on artificial intelligence, including: S1: used to collect the user's physiological data and behavioral data, and upload the collected physiological data and behavioral data to the cloud database for data preprocessing.

[0016] In this embodiment, the S1 registers a user account by inputting the user's personal information in the wearable device, binds the sensor in the wearable device to the user's account information, and then guides the user to wear the wearable device correctly, so as to monitor the user's physiological data and behavioral data in real time. The physiological data includes heart rate, skin resistance, respiratory rate and body temperature, and the behavioral data includes sleep pattern, amount of exercise, frequency of social activities, and mobile phone usage habits. Then, the data preprocessing operations include: synchronizing the physiological data and behavioral data to a unified time base, data cleaning and data standardization, and the preprocessing time complexity is O(N), where N is the amount of data.

[0017] It should be specifically stated that, by designing an integrated system, the latest generation of heart rate monitors, skin resistance sensors and biometric sensors are integrated into a wearable device, and the sensors are set to work under the same time base. The user registers and logs in personal information in the wearable device, binds the sensor of the wearable device to the user's account, and then guides the user to put the wearable device close to the wrist, so as to monitor the user's heart rate changes in real time, measure the user's skin electrical activity, and capture the user's physiological and behavioral patterns, such as facial expressions, body movements, etc. The data collected by the sensor is uploaded to the cloud database in real time through the wireless network. At the same time, the wearable device can be bound to the smartphone; according to the linear relationship between the time of the preprocessing operation and the amount of data, the preprocessing time complexity is calculated to be O(N), where N is the amount of data; For example, assuming that 1000 pieces of user physiological data and 400 pieces of user behavioral data are collected every day, the total number of data is 1400. The preprocessing time complexity at this time is calculated to be O(1400), which shows that as the amount of data increases, the time required for preprocessing will also increase accordingly, and the rate of increase is a linear relationship.

[0018] S2: Create a comprehensive mental state image by extracting features from the preprocessed physiological and behavioral data and then fusing the extracted features.

[0019] In this embodiment, S2 extracts physiological characteristics and behavioral characteristics from the preprocessed physiological data and behavioral data, wherein the physiological characteristics include heart rate variability characteristics, skin electrical activity characteristics, and respiratory periodic change characteristics, and the behavioral characteristics include activity level characteristics, social interaction frequency characteristics, and sleep quality index characteristics. The physiological characteristics and behavioral characteristics are merged into a feature matrix to form a feature set, and PCA is used to reduce the dimension and fuse the merged feature set; The specific steps of using PCA to reduce the dimension and fuse the merged feature set are: Step S211: Calculate the covariance matrix between each pair of features in the feature matrix; Step S212: performing eigendecomposition on the covariance matrix and obtaining eigenvalues ​​and eigenvectors; Step S213: sorting the eigenvalues ​​in descending order and selecting the eigenvectors corresponding to the first few eigenvalues ​​as principal components according to the eigenvalues; Step S214: reconstructing the data using the selected principal components to obtain a new feature set, that is, a fused feature set, and using the fused feature set to create a mental state image; By mapping the fused features to the representation space of psychological states, the feature set is converted into a psychological state image using data visualization technology. In the psychological state image, the typical emotion distribution can be identified through cluster analysis, and the movement trajectory of the individual in the representation space can be dynamically visualized to capture the changing trend of the user's psychological state.

[0020] It should be specifically stated that the method for extracting physiological characteristics is: calculating the statistical parameters of the RR interval through time domain analysis, including the mean, standard deviation, and coefficient of variation; converting the RR interval sequence to the frequency domain through fast Fourier transform, analyzing and calculating the power of different frequency bands as the characteristic value of heart rate variability; removing noise by using a bandpass filter and extracting the specific frequency components of the EDA signal to calculate the standard deviation of the skin electrode activity as the characteristic value of the skin electrode activity; extracting the respiratory signal by using wavelet transform or other signal processing techniques as the characteristic value of respiratory periodic changes; the method for extracting behavioral characteristics is: calculating the average acceleration, activity count, and activity intensity through accelerometer data analysis as the characteristic value of activity level; analyzing the mode of social interaction, such as interaction duration and interaction type, and calculating the characteristic value of social interaction frequency; staging sleep by using a sleep monitoring algorithm, extracting sleep structure characteristics, and calculating the characteristic values ​​of sleep quality indicators, including sleep efficiency, sleep latency, and number of awakenings; The specific process of reducing the dimension and fusing the merged feature set by using PCA is as follows: Assume that the feature matrix is ​​X, where each row represents a sample, i.e., a user's full-day physiological and behavioral records on a certain day, and each column represents a feature, i.e., physiological features and behavioral features, and the covariance matrix is ​​calculated as , where n represents the number of samples, and then the eigenvalues ​​of the matrix C are calculated by calling the eigendecomposition function in the numerical calculation library And the eigenvector v, the specific calculation formula is , based on the eigenvalue, the corresponding eigenvector is selected as the principal component, and the original feature matrix X is projected onto the new space composed of the principal components to obtain the reduced dimension feature matrix .

[0021] S3: Based on the fused data, the deep learning model is used to analyze the user's emotional state, and a preliminary judgment is made as to whether the individualized emotions are abnormal, and abnormal emotions are fed back to the user.

[0022] In this embodiment, S3 uses the principal component feature vector of the psychological state image as the input feature of the deep learning model, outputs a probability distribution of different emotional states, and calculates the formula of the emotional state probability value as follows: , represents the original score of the emotion category corresponding to the i-th neuron in the output layer, represents the jth raw score when traversing all emotion categories, M represents the total number of emotion categories in the output layer, e represents the base of the natural logarithm, and the emotional state of the current user is predicted based on the emotional state probability value; By analyzing and calculating the mean and standard deviation of the probability distribution of each emotion category in the normal state, the emotion probability threshold is set for each emotion category based on the mean and standard deviation. The specific calculation formula is: ,in, represents the emotion probability threshold of the i-th emotion category, represents the average probability value of all normal samples in the i-th emotion category, represents the standard deviation of the probability value of the normal sample in the i-th emotion category, and k represents a constant used to adjust the strictness of the threshold. The corresponding emotional state probability threshold Comparison is used to determine whether there is an abnormality in a certain emotion category of the user; if , then it is considered that the i-th emotion category is abnormal, and then the emotion state probability values ​​of all emotion categories and the emotion cumulative probability L of the corresponding emotion state probability threshold are analyzed and calculated. By comparing the emotion cumulative probability L with the preset cumulative probability threshold Compare and preliminarily judge whether individual emotions are abnormal. When the abnormal emotion judgment result is given to the user, the user is considered to have an abnormal emotion, and the abnormal emotion judgment result is fed back to the user to warn that the user's emotion change is abnormal.

[0023] It should be noted that the specific calculation formula for the cumulative probability of emotion is: , where L represents the cumulative probability of emotion, represents the weight coefficient of the i-th emotion category; The steps for building a deep learning model include: Step 1: Collect a large number of psychological state images and corresponding emotional state labels, standardize, normalize, frame-process and enhance the collected images, use PCA to extract principal component feature vectors from the images, and use the principal component feature vectors and corresponding emotional state labels as the data set; Step 2: Divide the data set into training set, validation set and test set. The division ratio can be set to 70% training set, 15% validation set and 15% test set. Step 3: Choose a deep learning model architecture suitable for emotion recognition, define the size of the input layer to match the dimension of the principal component feature vector, add one or more hidden layers according to the complexity of the problem, define the output layer with the same number of neurons as the number of emotion state categories, and use the softmax activation function to output the probability distribution; Step 4: Use the training set data to train the model, use the validation set to adjust hyperparameters and prevent overfitting, and use the test set to evaluate the performance of the trained model; For example, suppose the output layer has three neurons, corresponding to the categories "normal", "anxious", and "sad", and the raw scores output by the model are , calculated using the softmax function, and the emotional state probability value is According to this probability distribution, the predicted emotional state can be judged as "anxiety".

[0024] S4: Based on the analysis results of abnormal emotions, by comparing the physiological and behavioral characteristics of abnormal emotions with the physiological and behavioral characteristics of known mental illness symptoms, a time series analysis is performed to obtain the disease assessment coefficient, which is used to detect whether the user has a mental illness.

[0025] In this embodiment, S4 uses emotion recognition technology to identify the abnormal emotion period of the user based on the judgment result of abnormal emotion, and extracts the physiological characteristics and behavioral characteristics of the abnormal emotion period, and compares the physiological characteristics and behavioral characteristics of the abnormal emotion with the physiological characteristics and behavioral characteristics of the known psychological disease symptoms in a time series analysis, and then calculates the disease assessment coefficient P according to the time series comparison analysis result; By setting a disease threshold , the disease assessment coefficient P and the disease threshold Comparison is used to detect whether the user has a mental illness; when When the user is considered to have a high possibility of mental illness, the test result is transmitted to S5 for further analysis of the severity of the disease. When the user has a mental illness, it is considered that the possibility of the user having a mental illness is small, and the physiological and behavioral characteristics of the user's abnormal emotions will continue to be monitored.

[0026] It should be noted that the specific analysis method for the disease assessment coefficient is as follows: Step S411: By setting the physiological characteristic vector of the abnormal emotion period to , the physiological feature vector of known mental illness symptoms is , then the similarity of physiological characteristics is ,in, represents the dot product of vector A and vector B, and Represent the magnitude of vector A and vector B respectively; Step S412: By setting the behavior feature vector of the abnormal emotion period to , the behavioral feature vector of known mental illness symptoms is , then the behavior feature similarity is ,in, represents the dot product of vector D and vector H, and Respectively represent the magnitude of vector D and vector H; Step S413: Analyze based on the physiological characteristics similarity and behavioral characteristics similarity, and calculate the disease assessment coefficient: ,in, is the weight coefficient.

[0027] S5: Based on the disease detection results, further evaluate and analyze the severity of the disease, calculate the severity assessment coefficient, and issue a graded warning for the disease.

[0028] In this embodiment, the S5 analyzes the severity assessment coefficient based on the disease detection analysis result, and divides the severity of the disease into different levels according to the severity assessment coefficient, and the levels divided are: low risk, medium risk and high risk, and calculates the severity assessment coefficient Z by comparing and analyzing the abnormality of the physiological characteristics of the physiological signals in the abnormal emotional period and the physiological characteristics of the physiological signals in the normal state, and the abnormality of the behavioral characteristics of the behavioral patterns in the abnormal emotional period and the individual's regular behavioral patterns; Set low-risk thresholds based on severity of disease and high risk threshold ,when When When When , it indicates that the severity of the disease is in the high-risk range, and the severity of the disease is sent to the user end for early warning according to the risk level.

[0029] It should be noted that the specific analysis method for the severity assessment coefficient is as follows: Step S511: By comparing and analyzing the physiological characteristic abnormality of the physiological signals in the abnormal emotion period and the physiological signals in the normal state, the calculation formula is: ,in, Indicates the degree of abnormality of physiological characteristics. represents the physiological characteristic value of the qth user during the abnormal emotion period, represents the physiological characteristic value of the qth user in the normal population, and Q represents the number of physiological characteristics; Step S512: By comparing and analyzing the behavior pattern in the abnormal emotion period with the individual's regular behavior pattern, the abnormality of the behavior characteristics is calculated as follows: ,in, Indicates the abnormality of behavioral characteristics. represents the behavioral feature value of the u-th user during the abnormal emotion period, represents the behavior feature value of the u-th user in the normal population, and U represents the number of behavior features; Step S513: Analyze the abnormality of physiological characteristics and behavioral characteristics and calculate the severity assessment coefficient: , where P represents the disease assessment coefficient, They represent weight coefficients respectively, and their specific values ​​are determined according to the importance of disease type and characteristics.

[0030] like Figure 2The present embodiment shown provides an implementation system corresponding to an anti-counterfeiting method for articles based on local feature visual information, including a data collection module, a feature fusion module, a sentiment analysis module, a disease detection module and a disease severity assessment module, wherein the data collection module is connected to the feature fusion module, the feature fusion module is connected to the sentiment analysis module, the sentiment analysis module is connected to the disease detection module, the disease detection module is connected to the disease severity assessment module, and the sentiment analysis module is connected to the disease severity assessment module.

[0031] The data collection module is used to collect the user's physiological data and behavioral data, and upload the collected physiological data and behavioral data to the cloud database for data preprocessing; The feature fusion module extracts features from the preprocessed physiological data and behavioral data, and then fuses the extracted features to create a comprehensive mental state image; The emotion analysis module analyzes the user's emotional state based on the fused data by using a deep learning model, preliminarily determines whether the individual emotion is abnormal, and provides feedback prompts of abnormal emotions to the user end; The disease detection module is based on the analysis results of abnormal emotions, and compares the physiological characteristics and behavioral characteristics of abnormal emotions with the physiological characteristics and behavioral characteristics of known psychological disease symptoms to perform time series analysis to obtain a disease assessment coefficient for detecting whether the user has a psychological disease; The disease severity assessment module further assesses and analyzes the disease severity based on the disease detection results, calculates the severity assessment coefficient, and provides a graded warning for the disease.

[0032] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

[0033] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A mental illness early warning method based on artificial intelligence, characterized in that: include: S1: used to collect the user's physiological data and behavioral data, and upload the collected physiological data and behavioral data to the cloud database for data preprocessing; S2: Create a comprehensive mental state image by extracting features from the preprocessed physiological and behavioral data and then fusing the extracted features; S3: Based on the fused data, the deep learning model is used to analyze the user's emotional state, and a preliminary judgment is made as to whether the individual emotion is abnormal, and abnormal emotions are fed back to the user end; S4: Based on the analysis results of abnormal emotions, the physiological and behavioral characteristics of abnormal emotions are compared with the physiological and behavioral characteristics of known mental illness symptoms for time series analysis to obtain the disease assessment coefficient, which is used to detect whether the user has a mental illness; S5: Based on the disease detection results, further evaluate and analyze the severity of the disease, calculate the severity assessment coefficient, and issue a graded warning for the disease.

2. The method for early warning of mental illness based on artificial intelligence according to claim 1, characterized in that: The S1 registers a user account by inputting the user's personal information into the wearable device, binds the sensor in the wearable device to the user's account information, and then guides the user to wear the wearable device correctly, so as to monitor the user's physiological data and behavioral data in real time. The physiological data includes heart rate, skin resistance, respiratory rate and body temperature, and the behavioral data includes sleep pattern, amount of exercise, frequency of social activities, and mobile phone usage habits. Then, the data preprocessing operations include: synchronizing the physiological data and behavioral data to a unified time base, data cleaning and data standardization, and the preprocessing time complexity is O(N), where N is the amount of data.

3. The method for early warning of mental illness based on artificial intelligence according to claim 1, characterized in that: S2 extracts physiological features and behavioral features from the preprocessed physiological data and behavioral data, wherein the physiological features include heart rate variability features, skin electrical activity features, and respiratory periodicity change features, and the behavioral features include activity level features, social interaction frequency features, and sleep quality index features, merges the physiological features and behavioral features into a feature matrix to form a feature set, and uses PCA to reduce the dimension and fuse the merged feature set; The specific steps of using PCA to reduce the dimension and fuse the merged feature set are: Step S211: Calculate the covariance matrix between each pair of features in the feature matrix; Step S212: performing eigendecomposition on the covariance matrix and obtaining eigenvalues ​​and eigenvectors; Step S213: sorting the eigenvalues ​​in descending order and selecting the eigenvectors corresponding to the first few eigenvalues ​​as principal components according to the eigenvalues; Step S214: reconstructing the data using the selected principal components to obtain a new feature set, that is, a fused feature set, and using the fused feature set to create a mental state image; By mapping the fused features to the representation space of psychological states, the feature set is converted into a psychological state image using data visualization technology. In the psychological state image, the typical emotion distribution can be identified through cluster analysis, and the movement trajectory of the individual in the representation space can be dynamically visualized to capture the changing trend of the user's psychological state.

4. The method for early warning of mental illness based on artificial intelligence according to claim 1, characterized in that: S3 uses the principal component feature vector of the psychological state image as the input feature of the deep learning model, outputs a probability distribution of different emotional states, and calculates the formula of the emotional state probability value as follows: , represents the original score of the emotion category corresponding to the i-th neuron in the output layer, represents the jth raw score when traversing all emotion categories, M represents the total number of emotion categories in the output layer, e represents the base of the natural logarithm, and the emotional state of the current user is predicted based on the emotional state probability value; By analyzing and calculating the mean and standard deviation of the probability distribution of each emotion category in the normal state, the emotion probability threshold is set for each emotion category based on the mean and standard deviation. The specific calculation formula is: ,in, represents the emotion probability threshold of the i-th emotion category, represents the average probability value of all normal samples in the i-th emotion category, represents the standard deviation of the probability value of the normal sample in the i-th emotion category, and k represents a constant used to adjust the strictness of the threshold. The corresponding emotional state probability threshold Comparison is used to determine whether there is an abnormality in a certain emotion category of the user; if , it is considered that the i-th emotion category is abnormal, and then the emotion state probability values ​​of all emotion categories and the corresponding emotion state probability thresholds are analyzed and calculated. The specific calculation formula is: , where L represents the cumulative probability of emotion, represents the weight coefficient of the i-th emotion category, which is calculated by dividing the emotion cumulative probability L by the preset cumulative probability threshold Compare and preliminarily judge whether individual emotions are abnormal. When the abnormal emotion judgment result is given to the user, the user is considered to have an abnormal emotion, and the abnormal emotion judgment result is fed back to the user to warn that the user's emotion change is abnormal.

5. The method for early warning of mental illness based on artificial intelligence according to claim 1, characterized in that: Based on the abnormal emotion judgment result, S4 uses emotion recognition technology to identify the abnormal emotion period of the user, and extracts the physiological characteristics and behavioral characteristics of the abnormal emotion period, and compares the physiological characteristics and behavioral characteristics of the abnormal emotion with the physiological characteristics and behavioral characteristics of known psychological disease symptoms in a time series, and then calculates the disease assessment coefficient P according to the time series comparison analysis result; By setting a disease threshold , the disease assessment coefficient P and the disease threshold Comparison is used to detect whether the user has a mental illness; when When the user is considered to have a high possibility of mental illness, the test result is transmitted to S5 for further analysis of the severity of the disease. When the user has a mental illness, it is considered that the possibility of the user having a mental illness is small, and the physiological and behavioral characteristics of the user's abnormal emotions will continue to be monitored.

6. The method for early warning of mental illness based on artificial intelligence according to claim 1, characterized in that: The S5 analyzes the severity assessment coefficient based on the disease detection analysis result, and divides the severity of the disease into different levels according to the severity assessment coefficient, and the levels are divided into: low risk, medium risk and high risk, and calculates the severity assessment coefficient Z by comparing and analyzing the abnormality of the physiological characteristics of the physiological signals in the abnormal emotional period with the physiological signals in the normal state and the abnormality of the behavioral characteristics of the behavioral patterns in the abnormal emotional period with the individual's regular behavioral patterns; Set low-risk thresholds based on severity of disease and high risk threshold ,when When When When , it indicates that the severity of the disease is in the high-risk range, and the severity of the disease is sent to the user end for early warning according to the risk level.

7. A mental illness early warning system based on artificial intelligence, implementing a mental illness early warning method based on artificial intelligence as claimed in any one of claims 1 to 6, characterized in that: include: Data collection module: used to collect the user's physiological data and behavioral data, and upload the collected physiological data and behavioral data to the cloud database for data preprocessing; Feature fusion module: extracts features from preprocessed physiological and behavioral data, and then fuses the extracted features to create a comprehensive mental state image; Emotional analysis module: Based on the fused data, the deep learning model is used to analyze the user's emotional state, preliminarily determine whether the individual emotion is abnormal, and provide feedback prompts of abnormal emotions to the user end; Disease detection module: Based on the analysis results of abnormal emotions, the physiological and behavioral characteristics of abnormal emotions are compared with the physiological and behavioral characteristics of known psychological disease symptoms for time series analysis to obtain the disease assessment coefficient, which is used to detect whether the user has a psychological disease; Disease severity assessment module: Based on the disease detection results, further evaluate and analyze the severity of the disease, calculate the severity assessment coefficient, and provide graded warnings for the disease.

Images