Psychological stress grade identification method based on deep emotional neural network

Abstract

The invention discloses a psychological stress grade identification method based on a deep emotional neural network, and relates to the technical field of computers. Comprising the following steps: information acquisition: acquiring past pressure information and real-time multi-modal feature information of a user; according to the invention, derivation is carried out according to the weight distribution of the previous pressure information through the set weight processing method to obtain the derived pressure information under different weight distributions, so that the pressure identification model under different weight distributions can be established, and the relationship between different weight distributions and the pressure identification model can be recorded through the set directory table. According to the method, the corresponding pressure identification model is searched according to the weight distribution, and the target pressure level is obtained according to the real-time multi-modal feature information and the pressure level processed by the target pressure identification model, so that the effects of fully considering the difference of individual expression modes, adapting to different expression modes of individuals, reducing the probability of influencing the mental health of the individuals and improving the mental health of the individuals are achieved. And the credibility and the use experience of the user are improved.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and specifically to a method for identifying the psychological stress level based on a deep emotion neural network. Background Technique

[0002] Psychological stress refers to the psychological and physiological reactions that an individual generates when facing various stressors in life, work, or study. It may manifest as emotions and physical reactions such as anxiety, tension, irritability, insomnia, etc. Appropriate stress can motivate people to strive, but excessive stress may affect physical and mental health and even trigger mental illnesses.

[0003] A method and device for identifying psychological stress based on a neural network with the patent publication number CN111063420A obtains the photoplethysmogram (PPG) data of a user whose psychological stress is to be identified; performs cutting processing on the PPG data to obtain the cut PPG data; performs equal-length processing on the cut PPG data to obtain equal-length PPG data; inputs the equal-length PPG data into a psychological stress recognition model to obtain the psychological stress information of the user; wherein, the psychological stress recognition model is pre-trained on a preset neural network model using PPG training data labeled with psychological stress information. Based on the above solution, it is possible to identify psychological stress based on a neural network.

[0004] In the above and similar principles, most of the emotion neural networks for identifying psychological stress by neural networks are trained under clear signal conditions, lacking robustness to complex noise environments, and not fully considering the differences in individual expression methods, resulting in the model being difficult to identify the key emotion features representing stress, performing poorly among different populations, being unable to be adjusted to adapt to the different expression methods of individuals, and at the same time misidentifying users with high stress as having low stress, resulting in insufficient psychological intervention, affecting the mental health of individuals, and frequent false alarms may cause distrust and bad experiences of users, affecting the promotion and adoption of the system. Therefore, the present invention is proposed. Summary of the Invention

[0005] The purpose of the present invention is to provide a method for identifying the psychological stress level based on a deep emotion neural network to solve the problems raised in the above background technique.

[0006] To achieve the above purpose, the present invention provides the following technical solution: A method for identifying the psychological stress level based on a deep emotion neural network, including: Information acquisition: Obtain past stress information and the real-time multimodal feature information of the user. The past stress information includes past multimodal feature information and the stress information corresponding to the past multimodal feature information; Information processing: Obtain target noise by acquiring application scenario noise information, and perform noise reduction processing on real-time multimodal feature information through an information processing method based on the target noise to obtain processed real-time multimodal feature information; It is characterized in that it includes: Pressure level division: Perform pressure level division on the pressure information through a level division method to obtain a pressure level, and the pressure level includes several sub-pressure levels; Characteristic weight distribution: Derive derivative pressure information from the past pressure information according to different characteristic weights through a weight processing method based on the past pressure information; Model establishment: Establish a pressure recognition model under different characteristic weight distributions based on the derivative pressure information and the past pressure information to obtain a pressure recognition model and a directory table; Pressure level recognition: Determine the change weight by judging the user's expression pressure method, select the corresponding pressure recognition model based on the directory table through a selection method to obtain a target pressure recognition model, import the processed real-time multimodal feature information into the target pressure recognition model to obtain a target pressure, and obtain a target pressure level based on the target pressure and the pressure level.

[0007] Furthermore, the level division method includes: presetting the number of level divisions, the optimal pressure value range, and the decreasing value. The optimal pressure value range is the first sub-pressure level, and the optimal pressure value range includes a first boundary value and a second boundary value. The value of the first boundary value is greater than the second boundary value. Obtain a third boundary value based on the second boundary value combined with the decreasing value, integrate the second boundary value and the third boundary value to obtain a sub-optimal pressure value range, and the sub-optimal pressure value range is the second sub-pressure level. Obtain the third sub-pressure level based on the third boundary value combined with the generation decreasing value and the third boundary value, and integrate the first sub-pressure level, the second sub-pressure level, and the third sub-pressure level to obtain the pressure level.

[0008] Furthermore, the weight processing method includes: obtaining the basic weight distribution of the past multimodal feature information in the past pressure information to obtain a basic weight, and the basic weight includes sub-weights corresponding to individual feature information in the past multimodal feature information. Preset an adjustment factor, perform the same increase and reduction on the sub-weights in the basic weight based on the adjustment factor to obtain a derivative weight, obtain sub-derivative feature information based on the derivative weight and the past multimodal feature information, obtain the pressure of the sub-derivative feature information to obtain sub-derivative pressure, establish a correspondence between the sub-derivative feature information and the sub-derivative pressure, and integrate the sub-derivative feature information and the sub-derivative pressure to obtain derivative pressure information.

[0009] Furthermore, the process of establishing the pressure recognition model is: Training data acquisition and processing: Preprocess the derived stress information and past stress information to obtain the preprocessed derived stress information and preprocessed past stress information. Both the preprocessed derived stress information and the preprocessed past stress information are training data; Model selection and training: Select a deep learning model as the model matrix, and import the training data into the model matrix for the model matrix to be trained to obtain an initial model; Model adjustment and output: Preset verification information, which includes verification feature information and verification stress. Import the verification feature information into the initial model to obtain the result stress, judge the result stress and the verification stress to obtain a judgment result, and optimize and adjust the initial model based on the judgment result to obtain a stress recognition model; Model marking: Obtain marking information based on the weights of the training data, establish an association relationship between the marking information and the stress recognition model to obtain a target association, and establish a directory table for storing the target association.

[0010] Furthermore, the process of judging the user's stress expression method to obtain the change weight is as follows: Preset a basic weight, a fixed value, and a change threshold. The basic weight includes several sub-basic weights corresponding to single sub-modal feature information. Establish time nodes, and monitor the changes of single sub-modal feature information in the real-time multi-modal feature information based on adjacent time nodes to obtain sub-change values. Judge the relationship between all sub-change values and the change threshold, extract the sub-change values that exceed the change threshold to obtain target change values, add the fixed value to the sub-basic weights of the target change values to obtain target weights, extract the sub-change values that do not exceed the change threshold to obtain fixed change values, judge the number of fixed change values to obtain the reduction number, average the fixed value based on the reduction number to obtain an average value, subtract the average value from the sub-basic weights of the fixed change values to obtain fixed weights, and integrate all target weights and fixed weights to obtain the change weight.

[0011] Furthermore, the information processing method includes: Amplify and reduce the target noise to obtain the derived noise, establish a noise library for storing the target noise and the derived noise, extract the voice information in the multi-modal feature information to obtain the information to be processed, establish a timeline, match the information to be processed with the target noise and the derived noise in the noise library based on the timeline to obtain a matching result, extract the voice information corresponding to the target noise and the derived noise in the information to be processed based on the matching result to obtain the processed information to be processed, and replace the information to be processed in the multi-modal feature information with the processed information to be processed to obtain the processed real-time multi-modal information.

[0012] Further, the selection method includes: traversing a directory table based on a variation weight to obtain a traversal result. When the traversal result indicates that there is a weight corresponding to the variation weight in the target table, extracting the pressure recognition model corresponding to the weight of the variation weight in the target table to obtain a target pressure recognition model. When the traversal result indicates that there is no weight corresponding to the variation weight in the directory table, extracting the pressure recognition model with the weight closest to the variation weight in the directory table to obtain a target pressure recognition model.

[0013] Further, integrating real-time multimodal feature information, variation weight, target pressure level, and target pressure recognition model to obtain user information, establishing a repository for storing user information, obtaining a recognition requirement, and searching for user information corresponding to the recognition requirement in the repository based on the recognition requirement to obtain a search result.

[0014] Compared with the prior art, the beneficial effects of the present invention are: This method for recognizing psychological stress levels based on a deep emotional neural network derives derivative stress information under different weight distributions through the set weight processing method according to the weight distribution of past stress information, so as to establish stress recognition models under different weight distributions. Through the set directory table, the relationship between different weight distributions and stress recognition models can be recorded, so as to find the corresponding stress recognition model according to the weight distribution, and obtain the target stress level based on the target stress recognition model, the processed real-time multimodal feature information, and the stress level, so as to fully consider the differences in individual expression methods, adapt to different expression methods of individuals, reduce the probability of affecting individual mental health, and improve the user's trust and usage experience.

[0015] At the same time, through the set information processing method, noise removal processing is performed on the voice information in the real-time multimodal feature information to obtain the processed real-time multimodal feature information. Through the denoising process, the accuracy of the multimodal feature information can be improved, and further the accuracy of subsequent recognition of psychological stress levels can be improved, which is beneficial for use. Through the set level division method, the stress level is set according to the distribution of stress information according to the usage requirements, so that the stress level meets the requirements of the administrator. Through the set weight processing method, derivative stress information under different weight distributions is derived according to the weight distribution of past stress information, so as to establish stress recognition models under different weight distributions.

[0016] Meanwhile, a table of contents is established to store the target relationships, so as to obtain the corresponding stress recognition model according to the corresponding weights later. By training a single stress recognition model with the training data under a single weight distribution, the training efficiency of the stress recognition model can be improved. At the same time, the stress recognition models under different weight distributions are separated to facilitate the processing of multi-modal feature information under the corresponding weight distributions simultaneously, so as to improve the efficiency of stress recognition. By matching the user's recognition requirements with the user information in the repository to obtain a search result, according to the search result, it can be judged whether the user has used this method for psychological stress recognition, and the result of the user's last psychological stress level recognition can be obtained, so as to further judge the change of the user's psychological state and improve the user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is a schematic diagram of the main process structure of the present invention; Figure 2 is a schematic diagram of the structure for obtaining the target stress level of the present invention; Figure 3 is a schematic diagram of the change of characteristic values of the present invention; Figure 4 is a schematic diagram of the table of contents structure of the present invention; Figure 5 is a schematic diagram of the derivative weights of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0019] Identifying the psychological stress level can help detect the high-pressure state of an individual early, achieve scientific evaluation and timely intervention, thereby preventing mental health problems and promoting physical and mental health development. The method for identifying the psychological stress level based on a deep emotion neural network uses deep learning technology to automatically judge the psychological stress level according to the user's emotional expressions (such as voice, text, physiological signals).

[0020] As Figures 1-5 shown, the present invention provides a technical solution: a method for identifying the psychological stress level based on a deep emotion neural network, including: Information acquisition: acquiring past stress information and the user's real-time multi-modal feature information, where the past stress information includes past multi-modal feature information and the stress information corresponding to the past multi-modal feature information; It should be noted that the real-time multimodal feature information and the past multimodal feature information include voice, facial expressions, physiological signals, and behavioral actions. The past stress information can be obtained from the operation logs of the corresponding models for identifying the psychological stress level in the past, and the real-time multimodal feature information is obtained by installing the corresponding sensor and camera components.

[0021] Information processing: Obtain the target noise from the application scenario noise information, and perform noise reduction processing on the real-time multimodal feature information based on the target noise through the information processing method to obtain the processed real-time multimodal feature information; It should be noted that the target noise is the noise information of the application scenario, which can be specifically obtained by installing a noise sensor in the application scenario. The sound information in the real-time multimodal feature information is subjected to noise reduction processing through the set information processing method to obtain the processed real-time multimodal feature information. Through noise reduction processing, the accuracy of the multimodal feature information can be improved, and then the accuracy of subsequent identification of the psychological stress level can be improved, which is beneficial for use.

[0022] It is characterized by including: Stress level division: Based on the stress information, perform stress level division through the level division method to obtain the stress level, and the stress level includes several sub-stress levels; It should be noted that the stress level is set according to the distribution of the stress information and the usage requirements through the set level division method, so that the stress level meets the requirements of the administrator.

[0023] Feature weight distribution: Based on the past stress information, perform derivation on the past stress information according to different feature weights through the weight processing method to obtain the derived stress information; It should be noted that different weight distribution-derived stress information is obtained through the set weight processing method according to the weight distribution of the past stress information, so as to establish stress recognition models under different weight distributions.

[0024] Model establishment: Based on the derived stress information and the past stress information, establish stress recognition models under different feature weight distributions to obtain the stress recognition models and the directory table; It should be noted that a stress recognition model is established according to the derived stress information and the past stress information. The number of stress recognition models is the same as the number of different weight distributions. Through the set directory table, the relationship between different weight distributions and the stress recognition models can be recorded, so as to find the corresponding stress recognition model according to the weight distribution, so as to fully consider the differences in individual expression methods, adapt to the different expression methods of individuals, reduce the probability of affecting individual mental health, and improve the user's trust and usage experience.

[0025] Pressure level identification: Determine the change weight by judging the user's expression pressure method. Based on the directory table, select the pressure recognition model corresponding to the change weight through the selection method to obtain the target pressure recognition model. Import the processed real-time multi-modal feature information into the target pressure recognition model to obtain the target pressure, and obtain the target pressure level based on the target pressure and the pressure level.

[0026] It should be noted that the process of determining the change weight by judging the user's expression pressure method, that is, according to the user's expression method, determine the weight distribution that conforms to the user's individual characteristics. The process of obtaining the target pressure recognition model by selecting the pressure recognition model corresponding to the change weight through the selection method based on the directory table is to find the pressure recognition model related to the user according to the weight distribution that conforms to the user's individual characteristics to obtain the target pressure recognition model. By importing the processed real-time multi-modal feature information into the target pressure recognition model, the target pressure is obtained. According to the obtained target pressure and the pressure level, find the distribution of the target pressure in the pressure level to obtain the target pressure level.

[0027] In the specific usage process, by obtaining the address information of the administrator or the user, and feedbacking the generated target pressure level to the administrator or the user according to the address information, the pressure recognition result can be presented to the administrator or the user.

[0028] Such as Figure 1 shown, the level division method includes: preset level division quantity, optimal pressure value interval and decreasing value. The optimal pressure value interval is the first sub-pressure level, and the optimal pressure value interval includes the first boundary value and the second boundary value. The value of the first boundary value is greater than the second boundary value. Based on the second boundary value and the decreasing value, the third boundary value is obtained. Integrate the second boundary value and the third boundary value to obtain the sub-optimal pressure value interval, and the sub-optimal pressure value interval is the second sub-pressure level. Based on the third boundary value and the decreasing value combined with the third boundary value, the third sub-pressure level is obtained. Integrate the first sub-pressure level, the second sub-pressure level and the third sub-pressure level to obtain the pressure level.

[0029] It should be noted that the level division quantity is the specific number of pressure levels, the optimal pressure value interval is the pressure value interval with the smallest reaction pressure value, and the decreasing value is the decreasing numerical value. By selecting the smaller boundary in the optimal pressure value interval and decreasing it in combination with the decreasing value, the second sub-pressure level is obtained, and so on to obtain several sub-pressure levels. By integrating all sub-pressure levels, the pressure level is obtained. Through the setting of the level division method, the corresponding pressure level can be formulated according to the administrator's own needs, improving the administrator's usage experience.

[0030] In the specific implementation process, the number of level divisions can be set to four, namely the first sub-pressure level, the second sub-pressure level, the third sub-pressure level, and the fourth sub-pressure level. The optimal pressure value range is 75 - 100 for the pressure value. At this time, the lower the pressure value, the greater the pressure, and the decreasing value is 25. At this time, the pressure value range of the first pressure level is 75 - 100, the pressure value range of the second pressure level is 50 - 75, the pressure value range of the third pressure level is 25 - 50, and the pressure value range of the fourth pressure level is 0 - 25.

[0031] As Figure 1 and Figure 5 shown, the weight processing method includes: obtaining the basic weight distribution of the past multimodal feature information in the past pressure information to obtain the basic weight. The basic weight includes the sub-weights corresponding to the individual feature information in the past multimodal feature information, a preset adjustment factor, increasing and decreasing the sub-weights in the basic weight by the same amount based on the adjustment factor to obtain the derivative weight, obtaining the sub-derivative feature information based on the derivative weight in combination with the past multimodal feature information, obtaining the sub-derivative pressure of the sub-derivative feature information, establishing the correspondence between the sub-derivative feature information and the sub-derivative pressure, and integrating the sub-derivative feature information and the sub-derivative pressure to obtain the derivative pressure information.

[0032] It should be noted that the process of obtaining the basic weight distribution of the past multimodal feature information in the past pressure information to obtain the basic weight, that is, obtaining the weight distribution or attention of processing the past multimodal feature information in the production log of the past used model to obtain the basic weight. The adjustment factor is a specific value. By increasing and decreasing the sub-weights in the basic weight by the corresponding amount based on the adjustment factor to obtain the derivative weight, the process of obtaining the sub-derivative pressure of the sub-derivative feature information can be obtained by directly obtaining the pressure generated by the multimodal feature information under the same weight distribution, or by obtaining it in the pressure recognition model under the corresponding weight distribution. By integrating the sub-derivative feature information and the sub-derivative pressure to obtain the derivative pressure information, the training data for training the pressure recognition model under different weight distributions in the subsequent training can be obtained.

[0033] In the specific implementation process, extract the feature contribution degree or attention weight from the existing pressure recognition model (such as multimodal fusion network, attention mechanism, feature importance analysis) to obtain the basic weight. According to the derivative weight of the derivative pressure information, find the pressure recognition model corresponding to the derivative weight in the prior art. Import the sub-derivative feature information into the pressure recognition model to obtain the sub-derivative pressure, and integrate the sub-derivative pressure and the sub-derivative feature information to obtain the derivative pressure information. As Figure 5 shown, the basic weight, derivative weight 1, and derivative weight 2 include five features. The area size of the sub-weights in the basic weight, derivative weight 1, and derivative weight 2 represents the corresponding weight, and the reduced area is the same as the increased area in the overall weight.

[0034] As Figure 1 and Figure 4 shown, the process of establishing a pressure recognition model is as follows: Training data acquisition and processing: Preprocess the derived pressure information and the past pressure information to obtain the preprocessed derived pressure information and the preprocessed past pressure information. Both the preprocessed derived pressure information and the preprocessed past pressure information are training data; It should be noted that preprocessing the derived pressure information and the past pressure information can improve the accuracy of the derived pressure information and the past pressure information. Both the preprocessed derived pressure information and the preprocessed past pressure information are training data, that is, the preprocessed derived pressure information is a single set of training data, and the preprocessed past pressure information is also a single set of training data.

[0035] Model selection and training: Select a deep learning model as the model matrix, and import the training data into the model matrix for the model matrix to train to obtain an initial model; It should be noted that the deep learning model can be a convolutional neural network, etc. By importing the training data into the model matrix for the model to train, an initial model can be obtained.

[0036] Model adjustment and output: Preset verification information, where the verification information includes verification feature information and verification pressure. Import the verification feature information into the initial model to obtain the result pressure, judge the result pressure and the verification pressure to obtain a judgment result, and optimize and adjust the initial model based on the judgment result to obtain a pressure recognition model; It should be noted that the verification information can select single or multiple pieces of information from the training data, and rename the multi-modal feature information and pressure in the information to obtain the verification feature information and the verification pressure. By importing the verification feature information into the initial model, the result pressure under the verification feature information can be obtained, and the initial model can be optimized and adjusted according to the difference between the result pressure and the verification pressure to obtain a pressure recognition model.

[0037] Model marking: Obtain marking information based on the weights of the training data, establish an association relationship between the marking information and the pressure recognition model to obtain a target association, and establish a directory table for storing the target association.

[0038] It should be noted that through the process of model marking, that is, separating the pressure recognition models trained with training data under different weight distributions, and establishing a directory table for storing the target relationship, so as to obtain the corresponding pressure recognition model according to the corresponding weight later. Training a single pressure recognition model with the training data under a single weight distribution can improve the training efficiency of the pressure recognition model, and at the same time separate the pressure recognition models under different weight distributions to facilitate the processing of multi-modal feature information under the corresponding weight distributions to improve the efficiency of pressure recognition. As Figure 4As shown, the target table includes several pressure recognition models and the weight distribution corresponding to the training data.

[0039] As Figure 1 shown, the process of determining the change weight according to the user's expression pressure method is as follows: preset the basic weight, fixed value and change threshold. The basic weight includes several sub-basic weights corresponding to the single sub-modal feature information. Establish time nodes, and based on adjacent time nodes, monitor the change of the single sub-modal feature information in the real-time multi-modal feature information to obtain the sub-change value. Judge the relationship between all sub-change values and the change threshold, extract the sub-change values exceeding the change threshold to obtain the target change value, add the fixed value to the sub-basic weight of the target change value to obtain the target weight, extract the sub-change values not exceeding the change threshold to obtain the fixed change value, judge the number of fixed change values to obtain the reduction number, average the fixed value based on the reduction number to obtain the average value, subtract the average value from the sub-basic weight of the fixed change value to obtain the fixed weight, and integrate all target weights and fixed weights to obtain the change weight.

[0040] It should be noted that the basic weight, fixed value and change threshold are formulated according to the actual usage situation, and the time nodes are formulated according to the actual usage situation, specifically as multiple time points under the same time interval. By adding the fixed value to the sub-basic weight of the target change value to obtain the target weight and subtracting the average value from the sub-basic weight of the fixed change value to obtain the fixed weight, the weight can be changed according to the change of the user's own characteristics while ensuring the stability of the overall weight, so as to determine the corresponding weight that conforms to the multi-modal feature information and make the subsequent pressure recognition result conform to the user's personal characteristics.

[0041] In the specific usage process, when the multi-modal feature information contains five features, the sub-basic weights can be 0.15, 0.2, 0.3, 0.2 and 0.15 respectively, corresponding to the five features. The five features include feature one, feature two, feature three, feature four and feature five. When the sub-change value of feature four exceeds the change threshold, add the fixed value to the sub-basic weight of feature four, and average the fixed value into four parts to reduce the sub-basic weights of feature one, feature two, feature three and feature five. Integrate the increased and reduced sub-basic weights to obtain the change weight. Specifically, to judge whether the sub-change value exceeds the change threshold can be as Figure 3 shown. By determining the slope of the change threshold, and then comparing the slope of the change trend of the paragraph feature values of adjacent time nodes with the slope of the change threshold, it is possible to complete the judgment of whether the sub-change value exceeds the change threshold.

[0042] As Figure 2As shown in the figure, the information processing method includes: performing noise amplification and noise reduction based on the target noise to obtain derived noise, establishing a noise library for storing the target noise and the derived noise, extracting the voice information in the multi-modal feature information to obtain the information to be processed, establishing a timeline, matching the information to be processed with the target noise and the derived noise in the noise library based on the timeline to obtain a matching result, extracting the voice information corresponding to the target noise and the derived noise in the information to be processed based on the matching result to obtain the processed information to be processed, and replacing the information to be processed in the multi-modal feature information with the processed information to be processed to obtain the processed real-time multi-modal information.

[0043] It should be noted that by performing noise amplification and noise reduction based on the target noise to obtain derived noise, the samples excluding the target noise can be increased, and the data volume for processing the target noise in the subsequent process can be reduced. The specific amplification factor and reduction factor are determined according to the actual usage situation. The timeline is a time process, that is, the generation time process of the multi-modal feature information. Extracting the voice information corresponding to the target noise and the derived noise in the information to be processed based on the matching result to obtain the processed information to be processed means performing similarity matching of the target noise and the derived noise in the information to be processed, and removing the information corresponding to the target noise and the derived noise in the information to be processed to obtain the processed information to be processed.

[0044] As Figure 2 shown in the figure, the selection method includes: traversing the directory table based on the change weight to obtain a traversal result. When the traversal result feedback indicates that there is a weight corresponding to the change weight in the target table, extracting the pressure recognition model corresponding to the change weight in the target table to obtain the target pressure recognition model. When the traversal result feedback indicates that there is no weight corresponding to the change weight in the directory table, extracting the pressure recognition model with the weight closest to the change weight in the directory table to obtain the target pressure recognition model.

[0045] It should be noted that through the set selection method, the target pressure recognition model can be obtained by selecting the pressure recognition model corresponding to or close to the change weight in the directory table according to the change weight, achieving the function of screening the pressure recognition model that meets the change weight.

[0046] Integrating the real-time multi-modal feature information, the change weight, the target pressure level, and the target pressure recognition model to obtain user information, establishing a storage repository for storing the user information, obtaining the recognition requirement, and searching for the user information corresponding to the recognition requirement in the storage repository based on the recognition requirement to obtain a search result.

[0047] It should be noted that user information is obtained by recording and integrating real-time multi-modal feature information, change weights, target stress levels, and target stress recognition models. The user information can be understood as a user profile. A repository is established to store the user profile. The identification requirement is for users who need to perform psychological stress recognition. By matching the identification requirements of the users with the user information in the repository, a search result is obtained. According to the search result, it can be determined whether the user has used this method to perform psychological stress recognition, and the result of the user's last psychological stress level recognition can be obtained, so as to further judge the change in the user's psychological state and improve the user experience.

[0048] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended embodiments and their equivalents.

Claims

1. A method for identifying psychological stress levels based on a deep emotional neural network, comprising: Information acquisition: Obtain past stress information and the user's real-time multi-modal feature information. The past stress information includes past multi-modal feature information and stress information corresponding to the past multi-modal feature information; Information processing: Obtain the target noise by acquiring the application scenario noise information, and perform noise reduction processing on the real-time multi-modal feature information through an information processing method based on the target noise to obtain the processed real-time multi-modal feature information; It is characterized by including: Stress level division: Perform stress level division through a level division method based on the stress information to obtain the stress level. The stress level includes several sub-stress levels; Feature weight distribution: Derive the derivative stress information from the past stress information according to different feature weights through a weight processing method based on the past stress information; Model establishment: Establish a stress recognition model under different feature weight distributions based on the derivative stress information and the past stress information to obtain the stress recognition model and the directory table; Stress level recognition: Judge the user's way of expressing stress to obtain the change weight, select the stress recognition model corresponding to the change weight from the directory table through a selection method to obtain the target stress recognition model, import the processed real-time multi-modal feature information into the target stress recognition model to obtain the target stress, and obtain the target stress level based on the target stress and the stress level.

2. The method for identifying the psychological stress level based on a deep emotion neural network according to claim 1, characterized in that: The level division method includes: presetting the number of level divisions, the optimal stress value range and the decreasing value. The optimal stress value range is the first sub-stress level, and the optimal stress value range includes the first boundary value and the second boundary value. The value of the first boundary value is greater than the second boundary value. Obtain the third boundary value by combining the second boundary value with the decreasing value, and integrate the second boundary value and the third boundary value to obtain the sub-optimal stress value range. The sub-optimal stress value range is the second sub-stress level. Obtain the third sub-stress level by combining the third boundary value with the generation decreasing value and the third boundary value. Integrate the first sub-stress level, the second sub-stress level and the third sub-stress level to obtain the stress level.

3. A method for identifying the level of psychological stress based on a deep emotion neural network according to claim 1, characterized in that: The weight processing method includes: obtaining the basic weight distribution of the past multi-modal feature information in the past stress information to obtain the basic weight. The basic weight includes the sub-weights corresponding to the individual feature information in the past multi-modal feature information. Preset the adjustment factor, perform the same increase and reduction on the sub-weights in the basic weight based on the adjustment factor to obtain the derivative weight, obtain the sub-derivative feature information based on the derivative weight and the past multi-modal feature information, obtain the sub-derivative stress of the sub-derivative feature information, establish the corresponding relationship between the sub-derivative feature information and the sub-derivative stress, and integrate the sub-derivative feature information and the sub-derivative stress to obtain the derivative stress information.

4. A method for identifying the psychological stress level based on a deep emotion neural network according to claim 1, characterized in that: The process of establishing the stress recognition model is: Training data acquisition and processing: Preprocess the derivative stress information and the past stress information to obtain the preprocessed derivative stress information and the preprocessed past stress information. Both the preprocessed derivative stress information and the preprocessed past stress information are training data; Model selection and training: Select a deep learning model as the model matrix, import the training data into the model matrix, and train the model matrix to obtain the initial model; Model adjustment and output: Preset verification information, which includes verification feature information and verification pressure. Import the verification feature information into the initial model to obtain the result pressure, judge the result pressure and the verification pressure to obtain a judgment result, and optimize and adjust the initial model based on the judgment result to obtain a pressure recognition model; Model marking: Obtain marking information based on the weights of the training data, establish an association relationship between the marking information and the pressure recognition model to obtain a target association, and establish a directory table for storing the target association.

5. A method for identifying psychological stress level based on a deep emotion neural network according to claim 1, characterized in that: The process of judging the user's expression pressure method to obtain the change weight is as follows: Preset the basic weight, fixed value, and change threshold. The basic weight includes several sub-basic weights corresponding to the single sub-modal feature information. Establish time nodes, and monitor the changes of the single sub-modal feature information in the real-time multi-modal feature information based on adjacent time nodes to obtain sub-change values. Judge the relationship between all sub-change values and the change threshold, extract the sub-change values that exceed the change threshold to obtain target change values, add the fixed value to the sub-basic weights of the target change values to obtain target weights, extract the sub-change values that do not exceed the change threshold to obtain fixed change values, judge the number of fixed change values to obtain the reduction number, average the fixed value based on the reduction number to obtain the average value, subtract the average value from the sub-basic weights of the fixed change values to obtain fixed weights, and integrate all target weights and fixed weights to obtain the change weight.

6. The method for identifying psychological stress level based on deep emotion neural network according to claim 1, wherein: The information processing method includes: Amplify and reduce the target noise to obtain the derivative noise, establish a noise library for storing the target noise and the derivative noise, extract the voice information in the multi-modal feature information to obtain the information to be processed, establish a timeline, and match the information to be processed with the target noise and the derivative noise in the noise library based on the timeline to obtain a matching result. Based on the matching result, extract the voice information corresponding to the target noise and the derivative noise in the information to be processed to obtain the processed information to be processed, and replace the information to be processed in the multi-modal feature information with the processed information to be processed to obtain the processed real-time multi-modal information.

7. A method for identifying the level of psychological stress based on a deep emotion neural network according to claim 1, characterized in that: The selection method includes: Traverse the directory table based on the change weight to obtain a traversal result. When the traversal result feedback is that there is a weight corresponding to the change weight in the target table, extract the pressure recognition model corresponding to the weight of the change weight in the target table to obtain the target pressure recognition model. When the traversal result feedback is that there is no weight corresponding to the change weight in the directory table, extract the pressure recognition model with the weight closest to the change weight in the directory table to obtain the target pressure recognition model.

8. A method for identifying the psychological stress level based on a deep emotion neural network according to claim 1, characterized in that: Integrate the real-time multi-modal feature information, change weight, target pressure level, and target pressure recognition model to obtain user information, establish a storage library for storing user information, obtain the recognition requirement, and search for the user information corresponding to the recognition requirement in the storage library based on the recognition requirement to obtain a search result.

Images