Psychological state assessment method and system based on deep learning

Abstract

The invention discloses a psychological state assessment method and system based on deep learning, and belongs to the technical field of data processing, and the method comprises the steps: training a deep learning model according to historical representation data and an actual psychological state assessment result corresponding to the historical representation data, obtaining a psychological state assessment rule, and obtaining a psychological state assessment result. And the real-time representation data is processed by adopting the psychological state assessment rule in the psychological assessment process of the testee, and the target psychological state assessment result corresponding to the testee is obtained, so that the requirements on workers can be effectively reduced, and the workers can be effectively assisted in improving the psychological state assessment efficiency.

Description

Technical Field

[0001] The present invention belongs to the technical field of data processing, and particularly relates to a method and system for psychological state evaluation based on deep learning. Background Art

[0002] Artificial evaluation of psychological state is a commonly used method in the field of mental health, mainly through face-to-face communication between professionals and individuals, observing behavioral manifestations, and filling out questionnaires. During the evaluation process, psychologists or counselors comprehensively analyze aspects such as an individual's emotions, cognition, and behavior based on professional knowledge and experience. This method emphasizes individual uniqueness and situational factors and can deeply understand the complex reasons behind psychological states. However, artificial evaluation is highly subjective, time-consuming, and limited by the professional level of the evaluator, and there may be certain biases. Nevertheless, it still has important value in personalized counseling, treatment planning, and in-depth intervention. With the accelerating social pace, mental health problems have become increasingly prominent. Traditional methods for psychological state evaluation mainly rely on questionnaires and expert interviews, and have problems such as strong subjectivity and low efficiency. Summary of the Invention

[0003] The present invention provides a method and system for psychological state evaluation based on deep learning to solve the problems of strong subjectivity and low efficiency in the process of psychological state evaluation in the prior art.

[0004] On the one hand, the present invention provides a method for psychological state evaluation based on deep learning, including: Obtaining historical characterization data in the historical psychological evaluation input by a staff member and the corresponding actual psychological state evaluation result of the historical characterization data; Training a deep learning model according to the historical characterization data and the corresponding actual psychological state evaluation result of the historical characterization data to obtain a psychological state evaluation rule; During the psychological evaluation of a subject, collecting real-time characterization data, and processing the real-time characterization data using the psychological state evaluation rule to obtain the target psychological state evaluation result corresponding to the subject.

[0005] Further, obtaining historical characterization data in the historical psychological evaluation input by a staff member and the corresponding actual psychological state evaluation result of each historical characterization data includes: Obtaining electroencephalogram data, voice data, and facial expression image data of each question in the historical psychological evaluation input by a staff member to obtain historical characterization data of each question; wherein, the electroencephalogram data and the facial expression image data are both collected in multiple types at a preset data sampling frequency; Obtain the actual psychological state evaluation result corresponding to the historical representation data input by the staff; among them, the actual psychological state evaluation results corresponding to all the test questions of the same person are the same.

[0006] Further, according to the historical representation data and the actual psychological state evaluation result corresponding to the historical representation data, train the deep learning model to obtain the psychological state evaluation rule, including: Use a convolutional neural network to construct a first deep learning model for electroencephalogram data, use a convolutional neural network to construct a second deep learning model for facial expression image data, use a convolutional neural network to construct a third deep learning model for speech data, and use a convolutional neural network to construct a fourth deep learning model; For the first deep learning model, use the electroencephalogram data as the actual input of the first deep learning model, and use the actual psychological state evaluation result corresponding to the historical representation data as the expected output to train the first deep learning model to obtain the trained first deep learning model; For the second deep learning model, use the facial expression image data as the actual input of the second deep learning model, and use the actual psychological state evaluation result corresponding to the historical representation data as the expected output to train the second deep learning model to obtain the trained second deep learning model; For the third deep learning model, use the time-domain image corresponding to the speech data as the actual input of the third deep learning model, and use the actual psychological state evaluation result corresponding to the historical representation data as the expected output to train the third deep learning model to obtain the trained third deep learning model; For the fourth deep learning model, use the electroencephalogram data as the actual input of the first deep learning model to obtain the electroencephalogram features output by the fully connected layer in the first deep learning model; use the facial expression image data as the actual input of the second deep learning model to obtain the facial expression features output by the fully connected layer in the second deep learning model; use the time-domain image corresponding to the speech data as the actual input of the third deep learning model to obtain the speech features output by the fully connected layer in the third deep learning model; form a feature data matrix in a fixed order with the electroencephalogram features, facial expression features, and speech features of the same person, and use the feature data matrix as the input of the fourth deep learning model, and use the actual psychological state evaluation result corresponding to the historical representation data as the expected output to train the fourth deep learning model to obtain the trained fourth deep learning model; According to the trained first deep learning model, the trained second deep learning model, the trained third deep learning model, and the trained fourth deep learning model, obtain the psychological state evaluation rule.

[0007] Further, according to the trained first deep learning model, the trained second deep learning model, the trained third deep learning model, and the trained fourth deep learning model, a psychological state evaluation rule is obtained, including: Remove the output layers of the trained first deep learning model, the trained second deep learning model, and the trained third deep learning model, and connect the fully connected layers of the trained first deep learning model, the trained second deep learning model, and the trained third deep learning model to the feature data splicing layer to output electroencephalogram features, facial expression features, and speech features to the feature data splicing layer; The feature data splicing layer forms a feature data matrix by arranging the electroencephalogram features, facial expression features, and speech features corresponding to the same person in a fixed order, and uses the feature data matrix as the input of the fourth deep learning model to form a psychological state evaluation rule.

[0008] Further, the training methods of the first deep learning model, the second deep learning model, the third deep learning model, and the fourth deep learning model are the same. All of them are trained using an intelligent hybrid optimization algorithm, and the training process includes: Randomly initialize the hyperparameters of the deep learning model to be trained, and form a vector with the hyperparameters after random initialization to obtain a parameter individual. Repeat to obtain multiple different parameter individuals; where the deep learning model to be trained is the first deep learning model, the second deep learning model, the third deep learning model, or the fourth deep learning model; For any parameter individual, obtain the loss function value corresponding to the parameter individual, and determine the optimal individual according to the loss function values corresponding to all parameter individuals; According to the optimal individual, select the corresponding search space for each parameter individual to obtain the parameter individual after one search; For the parameter individual after one search, perform information fusion according to the optimal individual to obtain the parameter individual after the second search; For the parameter individual after the second search, perform a joint search in a random matching manner to obtain the parameter individual after the third search; For the parameter individual after the third search, perform a global search using an adaptive global jump search method to obtain the parameter individual after the fourth search; Judge whether the maximum number of training times has been reached. If so, according to the parameter individual after the fourth search, re-obtain the optimal individual, and use the hyperparameters included in the optimal individual as the final hyperparameters of the deep learning model to be trained to obtain the trained deep learning model to be trained. Otherwise, return to the step of obtaining the loss function value corresponding to the parameter individual.

[0009] Further, according to the optimal individual, a corresponding search space is selected for each parameter individual, and the parameter individual after one search is obtained, including: An adaptive weighting factor is obtained according to the current training times; For any parameter individual, an adaptive position search information is obtained by using the adaptive weighting factor and the cosine function; According to the optimal individual and the adaptive position search information, a corresponding search space is selected for the parameter individual, and the parameter individual after one search is obtained.

[0010] Further, for the parameter individual after one search, information fusion is performed according to the optimal individual, and the parameter individual after two searches is obtained, including: For the parameter individual after one search, a rotation search factor is obtained according to the current training times; According to the rotation search factor and the optimal individual, information fusion is performed on the parameter individual after one search, and the parameter individual after two searches is obtained.

[0011] Further, for the parameter individual after two searches, a joint search is performed in a random matching manner, and the parameter individual after three searches is obtained, including: For the parameter individual after two searches, a first random parameter individual and a second random parameter individual are randomly matched for the parameter individual; wherein, the loss function value of the parameter individual is greater than the loss function value of its corresponding first random parameter individual; Based on the current training times, an adaptive weighting coefficient is obtained, and according to the first random parameter individual, the second random parameter individual and the adaptive weighting coefficient, a joint search is performed on the parameter individual after two searches, and the parameter individual after three searches is obtained.

[0012] Further, for the parameter individual after three searches, a global search is performed in an adaptive global jump search manner, and the parameter individual after four searches is obtained, including: Based on the current training times, an adaptive chaotic search factor is obtained; According to the adaptive chaotic search factor, a global search is performed on the parameter individual after three searches, and the parameter individual after four searches is obtained.

[0013] On the other hand, the present invention provides a psychological state evaluation system based on deep learning, including: a historical data acquisition module, a deep learning module, and a psychological state evaluation module; The historical data acquisition module is used to acquire the historical representation data in the historical psychological evaluation input by the staff and the actual psychological state evaluation result corresponding to the historical representation data; The deep learning module is used to train a deep learning model based on the historical representation data and the actual psychological state evaluation results corresponding to the historical representation data, so as to obtain psychological state evaluation rules; The psychological state evaluation module is used to collect real-time representation data during the psychological evaluation of the subject, and process the real-time representation data by using the psychological state evaluation rules to obtain the target psychological state evaluation results corresponding to the subject.

[0014] A psychological state evaluation method based on deep learning provided by the present invention trains a deep learning model based on the historical representation data and the actual psychological state evaluation results corresponding to the historical representation data to obtain psychological state evaluation rules, and processes the real-time representation data by using the psychological state evaluation rules during the psychological evaluation of the subject to obtain the target psychological state evaluation results corresponding to the subject. It can not only effectively reduce the requirements for staff, but also effectively assist the staff in improving the evaluation efficiency of psychological states. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The accompanying drawings herein are incorporated into the specification and form a part of the specification, showing embodiments consistent with the present invention, and are used together with the specification to explain the principles of the present invention.

[0016] Figure 1 It is a schematic flowchart of a psychological state evaluation method based on deep learning provided by an embodiment of the present invention.

[0017] Figure 2 It is a schematic structural diagram of a psychological state evaluation system based on deep learning provided by an embodiment of the present invention.

[0018] Through the above accompanying drawings, specific embodiments of the present invention have been shown, and there will be more detailed descriptions hereinafter. These drawings and the textual descriptions are not intended to limit the scope of the inventive concept in any way, but to illustrate the concept of the present invention to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0019] Here, the exemplary embodiments will be described in detail, and the examples are shown in the accompanying drawings. When the following description refers to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present invention as detailed in the appended claims.

[0020] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0021] AsFigure 1 As shown in Figure 1 , an embodiment of the present invention provides a method for evaluating mental state based on deep learning, including: S101. Obtain historical representation data in the historical mental evaluation process input by a staff member and the corresponding actual mental state evaluation result of the historical representation data; The historical representation data in the historical mental evaluation process can be the electroencephalogram data, voice data, and facial expression image data of each test question of the test subject in the historical mental evaluation process, so as to realize multi-source data fusion recognition, which is beneficial to improving the accuracy of mental state evaluation.

[0022] The actual mental state evaluation result can be whether the mental state of the test subject is healthy. For example, assuming that the identification and evaluation are for depression, then the actual mental state evaluation result can be that the test subject has a risk of depression or the test subject does not have a risk of depression.

[0023] S102. Train a deep learning model according to the historical representation data and the corresponding actual mental state evaluation result of the historical representation data to obtain a mental state evaluation rule; Training the deep learning model through the historical representation data and the corresponding actual mental state evaluation result of the historical representation data can standardize the mental state evaluation process. Only when obtaining the historical representation data and the corresponding actual mental state evaluation result of the historical representation data, one expert or multiple experts can perform joint annotation to achieve a relatively standardized mental state evaluation, reduce the requirements for subsequent staff, and improve efficiency. Further, even if the staff member who inputs the historical representation data in the historical mental evaluation process and the corresponding actual mental state evaluation result of the historical representation data is not an expert, through the method provided by the embodiment of the present invention, the evaluation process can also be standardized, assisting the staff to complete the custom evaluation task, reducing the instability of subsequent manual evaluation, and improving the evaluation efficiency.

[0024] S103. During the mental evaluation process of the test subject, collect real-time representation data, and process the real-time representation data using the mental state evaluation rule to obtain the target mental state evaluation result corresponding to the test subject.

[0025] Processing the real-time representation data through the mental state evaluation rule to obtain the target mental state evaluation result corresponding to the test subject can effectively assist the staff to complete the custom evaluation task, improve work efficiency, and avoid the uncertainty brought by manual evaluation.

[0026] In the embodiment of the present invention, obtaining the historical representation data in the historical mental evaluation process input by a staff member and the corresponding actual mental state evaluation result of each historical representation data includes: Obtain the electroencephalogram data, voice data, and facial expression image data for each test question in the historical psychological assessment process input by the staff, and obtain the historical representation data for each test question; among them, multiple electroencephalogram data and facial expression image data are collected at a preset data sampling frequency. Obtain the actual psychological state assessment result corresponding to the historical representation data input by the staff; among them, the actual psychological state assessment results corresponding to the historical representation data of all test questions corresponding to the same person are the same.

[0027] For example, assume there are A test questions. For each test question, B electroencephalogram data and facial expression image data are collected within the specified answering time, and one voice data for answering a question is collected at the same time. Then the historical representation data corresponding to each test question includes one voice data for answering a question, B electroencephalogram data, and B facial expression image data.

[0028] In an embodiment of the present invention, according to the historical representation data and the actual psychological state assessment result corresponding to the historical representation data, train a deep learning model to obtain a psychological state assessment rule, including: Use a convolutional neural network to construct a first deep learning model for electroencephalogram data, use a convolutional neural network to construct a second deep learning model for facial expression image data, use a convolutional neural network to construct a third deep learning model for voice data, and use a convolutional neural network to construct a fourth deep learning model; For the first deep learning model, use the electroencephalogram data as the actual input of the first deep learning model, and use the actual psychological state assessment result corresponding to the historical representation data as the expected output, and train the first deep learning model to obtain the trained first deep learning model; For the second deep learning model, use the facial expression image data as the actual input of the second deep learning model, and use the actual psychological state assessment result corresponding to the historical representation data as the expected output, and train the second deep learning model to obtain the trained second deep learning model; For the third deep learning model, use the time-domain image corresponding to the voice data as the actual input of the third deep learning model, and use the actual psychological state assessment result corresponding to the historical representation data as the expected output, and train the third deep learning model to obtain the trained third deep learning model; For the fourth deep learning model, the electroencephalogram data is used as the actual input of the first deep learning model to obtain the electroencephalogram features output by the fully connected layer in the first deep learning model; the facial expression image data is used as the actual input of the second deep learning model to obtain the facial expression features output by the fully connected layer in the second deep learning model; the time-domain image corresponding to the voice data is used as the actual input of the third deep learning model to obtain the voice features output by the fully connected layer in the third deep learning model; the electroencephalogram features, facial expression features, and voice features corresponding to the same person are combined into a feature data matrix in a fixed order, and the feature data matrix is used as the input of the fourth deep learning model, and the actual psychological state evaluation result corresponding to the historical representation data is used as the expected output to train the fourth deep learning model to obtain the trained fourth deep learning model; Combined with the above example, it can be determined that the feature data matrix should contain A*(2B + 1) rows of data, and each row of data is data with the same dimension output by the fully connected layer. By identifying the multi-dimensional data of the subject during the psychological evaluation process, the psychological state evaluation can be effectively realized.

[0029] According to the trained first deep learning model, the trained second deep learning model, the trained third deep learning model, and the trained fourth deep learning model, obtain the psychological state evaluation rule.

[0030] In the embodiment of the present invention, according to the trained first deep learning model, the trained second deep learning model, the trained third deep learning model, and the trained fourth deep learning model, obtaining the psychological state evaluation rule includes: Remove the output layers of the trained first deep learning model, the trained second deep learning model, and the trained third deep learning model, and connect the fully connected layers of the trained first deep learning model, the fully connected layers of the trained second deep learning model, and the fully connected layers of the trained third deep learning model to the feature data splicing layer to output the electroencephalogram features, facial expression features, and voice features to the feature data splicing layer; The feature data splicing layer combines the electroencephalogram features, facial expression features, and voice features corresponding to the same person into a feature data matrix in a fixed order, and uses the feature data matrix as the input of the fourth deep learning model to form a psychological state evaluation rule.

[0031] In the embodiment of the present invention, the training methods of the first deep learning model, the second deep learning model, the third deep learning model, and the fourth deep learning model are the same, and all use the intelligent hybrid optimization algorithm for training, and the training process includes: Randomly initialize the hyperparameters of the deep learning model to be trained, and form a vector with the hyperparameters after random initialization to obtain a parameter individual, and repeat to obtain multiple different parameter individuals; wherein, the deep learning model to be trained is the first deep learning model, the second deep learning model, the third deep learning model or the fourth deep learning model; It should be noted that the parameter individual should include all the hyperparameters to be trained of the deep learning model to be trained.

[0032] For any parameter individual, obtain the loss function value corresponding to the parameter individual, and determine the optimal individual according to the loss function values corresponding to all parameter individuals; For example, after applying the parameter individual to the deep learning model to be trained, use the actual output and the expected output of the deep learning model to be trained to obtain the root mean square loss function value or the cross entropy loss function value, and obtain the loss function value corresponding to the parameter individual.

[0033] According to the optimal individual, select the corresponding search space for each parameter individual to obtain the parameter individual after one search; For the parameter individual after one search, perform information fusion according to the optimal individual to obtain the parameter individual after the second search; For the parameter individual after the second search, perform a joint search in a random matching manner to obtain the parameter individual after the third search; For the parameter individual after the third search, perform a global search using an adaptive global jump search method to obtain the parameter individual after the fourth search; Judge whether the maximum number of training times has been reached. If so, re-obtain the optimal individual according to the parameter individual after the fourth search, and use the hyperparameters included in the optimal individual as the final hyperparameters of the deep learning model to be trained to obtain the trained deep learning model to be trained. Otherwise, return to the step of obtaining the loss function value corresponding to the parameter individual.

[0034] Optionally, after each search, the parameter individual can be processed for out-of-bounds to ensure the effectiveness of algorithm training.

[0035] In the prior art, the gradient descent and backpropagation methods are often used to optimize the hyperparameters of the deep learning model. Although it has a relatively fast training speed, it is extremely easy to fall into local optima, which may lead to poor task execution ability and thus poor psychological state evaluation effect. Therefore, the embodiments of the present invention provide an intelligent hybrid optimization algorithm, which can effectively ensure the algorithm training speed while enhancing the global search ability of the algorithm, thereby enhancing the algorithm's recognition ability of data relationships and ultimately improving the evaluation accuracy of psychological states.

[0036] In an embodiment of the present invention, according to the optimal individual, a corresponding search space is selected for each parameter individual, and the parameter individual after one search is obtained, including: According to the current number of training times, the adaptive weighting factor is obtained as:

[0037] Wherein, represents the adaptive weighting factor, represents the first random number between (0, 1), S represents the search control constant, e represents the natural constant, t represents the current number of training times, T represents the maximum number of training times, and cos represents the cosine function; For any parameter individual, the adaptive position search information is obtained by using the adaptive weighting factor and the cosine function as:

[0038] Wherein, represents the t th training process, the i th parameter individual, i = 1, 2,..., P, and P represents the total number of parameter individuals, represents a random angle between (0, ), represents pi, represents the adaptive position search information corresponding to the parameter individual; According to the optimal individual and the adaptive position search information, a corresponding search space is selected for the parameter individual, and the parameter individual after one search is obtained as:

[0039] Wherein, represents the optimal individual, represents the parameter individual after one search .

[0040] The one - search provided by the embodiment of the present invention enables each parameter individual to select a corresponding search space around the optimal individual based on its own position, which can effectively improve the search speed of the algorithm, and gradually converge as the algorithm progresses, improving the search efficiency near the optimal solution.

[0041] In an embodiment of the present invention, for the parameter individual after one search, information fusion is performed according to the optimal individual, and the parameter individual after two searches is obtained, including: For the parameter individual after one search, according to the current number of training times, the rotation search factor is obtained as:

[0042] Wherein, represents the rotation search factor, represents the exponential function with the natural constant e as the base, represents the cosine function, represents the cosine function, represents the pi, T represents the maximum number of training times, and k represents the rotation search control parameter; According to the rotation search factor and the optimal individual, perform information fusion on the parameter individuals after one search to obtain the parameter individuals after the second search as:

[0043] wherein, represents the k-th parameter individual after the first search in the t-th training process, k = 1, 2,..., P, and P represents the total number of parameter individuals, represents the parameter individual after the second search , represents the second random number between (0, 1), represents the spiral search range coefficient, and , represents the third random number between (0, 1).

[0044] The second search provided by the embodiment of the present invention can enable the parameter individuals to effectively exchange information with the optimal individuals, further improve the search speed of the algorithm, and at the same time improve the search ability for unknown local areas, and improve the ability of the algorithm to jump out of the local optimum.

[0045] In the embodiment of the present invention, for the parameter individuals after the second search, a random matching method is adopted for joint search to obtain the parameter individuals after the third search, including: For the parameter individuals after the second search, randomly match the first random parameter individual and the second random parameter individual for the parameter individuals; wherein, the loss function value of the parameter individual is greater than the loss function value of its corresponding first random parameter individual; Based on the current training times, obtain the adaptive weighting coefficient as:

[0046] wherein, represents the adaptive weighting coefficient, represents the first range control constant of the adaptive weighting coefficient, and is set to 0.6; represents the second range control constant of the adaptive weighting coefficient, and is set to 0.8; represents the transformation control factor of the adaptive weighting coefficient, and is set to 5; Based on the first random parameter individual, the second random parameter individual, and the adaptive weighting coefficient, perform a joint search on the parameter individuals after the second search to obtain the parameter individuals after the third search as follows:

[0047] where, represents the m-th parameter individual after the second search in the t-th training process, represents the parameter individual after the third search , m = 1, 2, …, P, where P represents the total number of parameter individuals, represents the first learning factor, represents the second learning factor, represents the fourth random number between (0, 1), represents the fifth random number between (0, 1), represents the first random parameter individual, represents the second random parameter individual.

[0048] The third search provided by the embodiments of the present invention can perform local information exchange and, in combination with the adaptive weighting coefficient, can effectively improve the local search ability, ensuring the local search ability in the later stage of the algorithm and the further search ability in unknown regions.

[0049] In the embodiments of the present invention, for the parameter individuals after the third search, an adaptive global jump search method is adopted for global search to obtain the parameter individuals after the fourth search, including: Based on the current training times, obtain the adaptive chaotic search factor as:

[0050] where, represents the adaptive chaotic search factor in the t-th training process, represents the adaptive chaotic search factor in the (t + 1)-th training process, represents the chaotic search control parameter, and is set to 0.7; Based on the adaptive chaotic search factor, perform a global search on the parameter individuals after the third search to obtain the parameter individuals after the fourth search as follows:

[0051] where, represents the n-th parameter individual after the third search in the t-th training process, represents the parameter individual after the fourth search , represents the upper limit vector composed of the upper limit values of the hyperparameter individuals, denotes the lower bound vector composed of the lower bound values of hyperparameter individuals, and sin denotes the sine function. denotes a random angle within [0, 2 . denotes the parameter individual and the upper bound vector the Euclidean distance between them. denotes the parameter individual and the lower bound vector the Euclidean distance between them.

[0052] Optionally, after four searches, an annealing simulation algorithm or a greedy algorithm can also be used to control the four-search process to ensure the search speed of the algorithm.

[0053] The four searches provided by the embodiments of the present invention can effectively provide global search capabilities, enabling the algorithm to have the ability to jump out of local optima, thereby improving the training effect of the algorithm.

[0054] Through the mutual cooperation of the above four search processes, the training effect of the algorithm can be effectively improved, and finally the evaluation accuracy of the mental state can be improved.

[0055] As Figure 2 shown, the embodiments of the present invention provide a mental state evaluation system based on deep learning, including: a historical data acquisition module 201, a deep learning module 202, and a mental state evaluation module 203; The historical data acquisition module 201 is used to acquire historical characterization data in the historical psychological evaluation input by the staff and the actual mental state evaluation results corresponding to the historical characterization data; The deep learning module 202 is used to train a deep learning model according to the historical characterization data and the actual mental state evaluation results corresponding to the historical characterization data to obtain mental state evaluation rules; The mental state evaluation module 203 is used to collect real-time characterization data during the psychological evaluation of the subject and process the real-time characterization data using the mental state evaluation rules to obtain the target mental state evaluation results corresponding to the subject.

[0056] The mental state evaluation system based on deep learning provided by the embodiments of the present invention can execute the above method technical solutions, and its principles and beneficial effects are similar, so they will not be elaborated here.

[0057] A method for evaluating mental state based on deep learning provided by the present invention trains a deep learning model according to the historical representation data and the actual mental state evaluation results corresponding to the historical representation data to obtain a mental state evaluation rule, and processes the real-time representation data by using the mental state evaluation rule during the mental evaluation of the subject to obtain the target mental state evaluation result corresponding to the subject, which can not only effectively reduce the requirements for the staff, but also effectively assist the staff in improving the evaluation efficiency of the mental state.

[0058] After considering the specification and practicing the invention disclosed herein, those skilled in the art will readily conceive of other embodiments of the present invention. The present invention is intended to cover any variations, uses, or adaptations of the present invention, which follow the general principles of the present invention and include well-known common knowledge or conventional technical means in the technical field not disclosed by the present invention. It should be understood that the present invention is not limited to the exact structures described above and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the present invention is only limited by the appended claims.

Claims

1. A method for evaluating mental state based on deep learning, characterized in that, Including: Obtain historical representation data in the historical psychological assessment process input by the staff and the actual psychological state assessment results corresponding to the historical representation data; Train a deep learning model according to the historical representation data and the actual psychological state assessment results corresponding to the historical representation data to obtain a psychological state assessment rule; During the psychological assessment of the subject, collect real-time representation data, and process the real-time representation data using the psychological state assessment rule to obtain the target psychological state assessment results corresponding to the subject.

2. The method for evaluating mental state based on deep learning according to claim 1, wherein Obtain historical representation data in the historical psychological assessment process input by the staff and the actual psychological state assessment results corresponding to each historical representation data, including: Obtain electroencephalogram data, voice data, and facial expression image data for each test question in the historical psychological assessment process input by the staff to obtain historical representation data for each test question; among them, both the electroencephalogram data and the facial expression image data are collected in multiple types at a preset data sampling frequency; Obtain the actual psychological state assessment results corresponding to the historical representation data input by the staff; among them, the actual psychological state assessment results corresponding to the historical representation data corresponding to all test questions of the same person are the same.

3. The method for psychological state evaluation based on deep learning according to claim 2, wherein Train a deep learning model according to the historical representation data and the actual psychological state assessment results corresponding to the historical representation data to obtain a psychological state assessment rule, including: Construct a first deep learning model for electroencephalogram data using a convolutional neural network, construct a second deep learning model for facial expression image data using a convolutional neural network, construct a third deep learning model for voice data using a convolutional neural network, and construct a fourth deep learning model using a convolutional neural network; For the first deep learning model, use the electroencephalogram data as the actual input of the first deep learning model, and use the actual psychological state assessment results corresponding to the historical representation data as the expected output to train the first deep learning model to obtain the trained first deep learning model; For the second deep learning model, use the facial expression image data as the actual input of the second deep learning model, and use the actual psychological state assessment results corresponding to the historical representation data as the expected output to train the second deep learning model to obtain the trained second deep learning model; For the third deep learning model, use the time-domain image corresponding to the voice data as the actual input of the third deep learning model, and use the actual psychological state assessment results corresponding to the historical representation data as the expected output to train the third deep learning model to obtain the trained third deep learning model; For the fourth deep learning model, the electroencephalogram (EEG) data is used as the actual input of the first deep learning model to obtain the EEG features output by the fully connected layer in the first deep learning model; the facial expression image data is used as the actual input of the second deep learning model to obtain the facial expression features output by the fully connected layer in the second deep learning model; the time-domain image corresponding to the voice data is used as the actual input of the third deep learning model to obtain the voice features output by the fully connected layer in the third deep learning model; the EEG features, facial expression features, and voice features corresponding to the same person are combined into a feature data matrix in a fixed order, and the feature data matrix is used as the input of the fourth deep learning model, and the actual psychological state evaluation result corresponding to the historical representation data is used as the expected output to train the fourth deep learning model to obtain the trained fourth deep learning model; According to the trained first deep learning model, the trained second deep learning model, the trained third deep learning model, and the trained fourth deep learning model, obtain the psychological state evaluation rule.

4. The method for evaluating mental state based on deep learning according to claim 3, wherein According to the trained first deep learning model, the trained second deep learning model, the trained third deep learning model, and the trained fourth deep learning model, obtain the psychological state evaluation rule, including: Remove the output layers of the trained first deep learning model, the trained second deep learning model, and the trained third deep learning model, and connect the fully connected layers of the trained first deep learning model, the trained second deep learning model, and the trained third deep learning model to the feature data splicing layer to output the EEG features, facial expression features, and voice features to the feature data splicing layer; The feature data splicing layer combines the EEG features, facial expression features, and voice features corresponding to the same person into a feature data matrix in a fixed order, and uses the feature data matrix as the input of the fourth deep learning model to form the psychological state evaluation rule.

5. The method for psychological state evaluation based on deep learning according to claim 4, characterized in that, The training methods of the first deep learning model, the second deep learning model, the third deep learning model, and the fourth deep learning model are the same, and all use the intelligent hybrid optimization algorithm for training, and the training process includes: Randomly initialize the hyperparameters of the deep learning model to be trained, and form a vector with the hyperparameters after random initialization to obtain a parameter individual, and repeat to obtain multiple different parameter individuals; where the deep learning model to be trained is the first deep learning model, the second deep learning model, the third deep learning model, or the fourth deep learning model; For any parameter individual, obtain the loss function value corresponding to the parameter individual, and determine the optimal individual according to the loss function values corresponding to all parameter individuals; According to the optimal individual, select the corresponding search space for each parameter individual to obtain the parameter individual after one search; For the parameter individual after one search, perform information fusion according to the optimal individual to obtain the parameter individual after the second search; For the parameter individuals after the second search, a random matching method is adopted for joint search to obtain the parameter individuals after the third search; For the parameter individuals after the third search, an adaptive global jump search method is adopted for global search to obtain the parameter individuals after the fourth search; Judge whether the maximum number of training times has been reached. If so, based on the parameter individuals after the fourth search, the optimal individual is re-obtained, and the hyperparameters included in the optimal individual are used as the final hyperparameters of the deep learning model to be trained, and the deep learning model to be trained after training is obtained. Otherwise, return to the step of obtaining the loss function value corresponding to the parameter individual.

6. The method for psychological state evaluation based on deep learning according to claim 5, wherein According to the optimal individual, a corresponding search space is selected for each parameter individual to obtain the parameter individuals after the first search, including: Obtain an adaptive weighting factor according to the current training times; For any parameter individual, an adaptive position search information is obtained by using the adaptive weighting factor and the cosine function; According to the optimal individual and the adaptive position search information, a corresponding search space is selected for the parameter individual to obtain the parameter individuals after the first search.

7. The method for evaluating mental state based on deep learning according to claim 6, wherein For the parameter individuals after the first search, information fusion is performed according to the optimal individual to obtain the parameter individuals after the second search, including: For the parameter individuals after the first search, a rotation search factor is obtained according to the current training times; According to the rotation search factor and the optimal individual, information fusion is performed on the parameter individuals after the first search to obtain the parameter individuals after the second search.

8. The method for evaluating mental state based on deep learning according to claim 7, characterized in that, For the parameter individuals after the second search, a random matching method is adopted for joint search to obtain the parameter individuals after the third search, including: For the parameter individuals after the second search, a first random parameter individual and a second random parameter individual are randomly matched for the parameter individual; among them, the loss function value of the parameter individual is greater than the loss function value of its corresponding first random parameter individual; Based on the current training times, an adaptive weighting coefficient is obtained, and based on the first random parameter individual, the second random parameter individual and the adaptive weighting coefficient, joint search is performed on the parameter individuals after the second search to obtain the parameter individuals after the third search.

9. The method for evaluating mental state based on deep learning according to claim 8, wherein For the parameter individuals after the third search, an adaptive global jump search method is adopted for global search to obtain the parameter individuals after the fourth search, including: Based on the current training times, an adaptive chaotic search factor is obtained; According to the adaptive chaotic search factor, global search is performed on the parameter individuals after the third search to obtain the parameter individuals after the fourth search.

10. A psychological state evaluation system based on deep learning, characterized in that, Including: A historical data acquisition module, a deep learning module and a mental state evaluation module; The historical data acquisition module is used to acquire the historical representation data in the historical psychological evaluation input by the staff and the actual psychological state evaluation result corresponding to the historical representation data; The deep learning module is used to train the deep learning model according to the historical representation data and the actual psychological state evaluation result corresponding to the historical representation data to obtain the mental state evaluation rule; The psychological state evaluation module is used to collect real-time representation data during the psychological evaluation of the subject, and process the real-time representation data using the psychological state evaluation rules to obtain the target psychological state evaluation result corresponding to the subject.

Images