State feature analysis method and system based on wake period information and dream characteristics

Abstract

This invention relates to the field of artificial intelligence technology, specifically to a method and system for state feature analysis based on waking-time information and dream characteristics. The method includes: acquiring multi-source waking-time information of a target individual and performing structured processing to generate waking-time information features; simultaneously, acquiring their dream text data and performing deep analysis to extract multi-dimensional dream features; subsequently, fusing the two types of features to generate a joint feature vector; inputting this joint feature vector into a pre-trained state feature analysis model to calculate a quantitative evaluation result of the state features; and finally outputting the evaluation result. This invention, through systematic fusion analysis of multi-dimensional information from an individual's waking-time and deep features of dream content, achieves a more comprehensive, objective, and dynamic quantitative auxiliary evaluation of mental state, outputting specific results including scores, trends, and risk levels, effectively overcoming the shortcomings of traditional evaluation methods such as single-dimensionality and reliance on subjective instantaneous reports.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, specifically to a method and system for analyzing state features based on waking information and dream characteristics. Background Technology

[0002] An individual's mental state is a comprehensive reflection of their emotional, cognitive, and behavioral characteristics. Accurate assessment of mental state is crucial for maintaining mental health, early intervention, and personal life adjustment. Traditionally, the assessment of mental state has relied primarily on standardized psychological questionnaires, structured clinical interviews, or observational judgments by professionals. While these methods have some validity, they have several limitations: First, they are mostly discrete-time point sampling assessments, making it difficult to monitor long-term, continuous changes in an individual's state at low cost and high frequency; second, the assessment process largely depends on the individual's subjective recollection and statements, which may be influenced by social desirability effects, current emotions, or cognitive biases; third, these methods are usually based on conscious information in a waking state and fail to systematically incorporate the rich content of unconscious or subconscious activities (such as dreams) as an analytical dimension.

[0003] In reality, dreams, as a common psychological experience during sleep, are not entirely random in terms of content, emotional tone, and narrative structure. They are often intricately linked to an individual's waking experiences, emotional state, unresolved conflicts, and deep cognitive patterns. Psychological research indicates that dreams may reflect an individual's stress levels, core themes of emotional distress, and potential adaptive resources. However, current technological advancements lack a technical solution for automatically and systematically co-modeling and analyzing the characteristics of multi-source information from an individual's waking period (such as daily experiences, behavioral logs, emotional records, and physiological signals) with dream text features. Most research on the link between dreams and mental states remains at the level of ex post facto analysis based on human experience or small-sample qualitative research, failing to achieve large-scale, objective, and quantifiable auxiliary assessments.

[0004] Therefore, it is necessary to provide an innovative technical solution that can comprehensively utilize artificial intelligence, natural language processing, and multimodal information fusion technology to automatically collect and structure multidimensional information of individuals during their waking hours. At the same time, it can deeply analyze the semantic, emotional, thematic, and narrative features of dream texts and effectively correlate and fuse the two to output quantitative, dynamic, and multidimensional mental state auxiliary assessment results, thereby making up for the shortcomings of existing assessment methods in terms of continuity, objectivity, and information integration.

[0005] Therefore, the existing technology still needs further development. Summary of the Invention

[0006] The purpose of this invention is to overcome the above-mentioned technical deficiencies and provide a method and system for analyzing state characteristics based on waking information and dream features, so as to solve the problems existing in the prior art.

[0007] To achieve the above-mentioned technical objectives, according to a first aspect of the present invention, the present invention provides a state feature analysis method based on waking-time information and dream characteristics, comprising: S1. Obtain multi-source lucidity information related to the target individual, and convert the lucidity information into structured data that can be processed by a computer; S2. Obtain the dream text data of the target individual, and perform feature analysis on the dream text data to extract multi-dimensional dream features; S3. Perform feature fusion processing on the structured features corresponding to the lucidity information and the multidimensional dream features to generate a joint feature vector for representing the individual's overall state; S4. Input the joint feature vector into the pre-trained state feature analysis model to calculate the state feature quantitative evaluation result; wherein, the state feature analysis model is trained based on the waking-time information features, dream features and their corresponding labeled state data of the sample individual; S5. Output the quantitative evaluation results of the state characteristics.

[0008] Specifically, the quantitative assessment results of the state characteristics include at least one of the following: state characteristic score, state characteristic vector, state change trend index, risk assessment level information, or summary of life status recommendations; wherein, the state characteristic score is obtained by fusing the calculation results of each dimension sub-feature in the joint feature vector; the state change trend index is obtained by comparing the difference of the joint feature vector or the change of the state characteristic score between the current time period and the historical time period.

[0009] Specifically, the lucidity information includes at least one of the following: individual experience narrative text, emotional state record text, daily life behavior log, social relationship and interpersonal interaction record, information exposure or media exposure content; step S1 further includes: Semantic embedding, topic modeling, and emotion polarity and intensity analysis are performed on text-based lucidity information, and event encoding and frequency statistics are performed on behavioral record information to generate structured features corresponding to the lucidity information.

[0010] Specifically, the multidimensional dream features include at least two of the following extracted from the dream text data: emotional features, theme features, narrative structure complexity features, semantic coherence features, and imagery element features; wherein, the emotional features include emotional valence and emotional intensity sequence, the theme features are extracted through a theme model, and the narrative structure complexity features are calculated based on the completeness of the event chain and the number of turning points.

[0011] Specifically, the method further includes: Acquire multiple sets of lucidity information and corresponding dream text data of the target individual over multiple consecutive time periods; In step S3, a joint feature vector is generated for each time period, and the vectors are arranged in chronological order to form a sequence of joint feature vectors. Step S4 includes: inputting the joint feature vector sequence into the state feature analysis model, analyzing the temporal evolution pattern of the state features, and outputting a state feature analysis report reflecting the long-term trend of change.

[0012] Specifically, the multiple time periods include cycles in units of days, weeks, or months; the state feature analysis model is a recurrent neural network, a long short-term memory network, or a temporal convolutional network, used to capture long-term dependencies and periodic patterns in the joint feature vector sequence and predict future state feature trends.

[0013] Specifically, the lucidity information also includes biosignal information corresponding to the target individual, the biosignal information including at least one of electroencephalogram (EEG), electrocardiogram (ECG), or electrodermal osmosis (EDS); the method further includes: The biological signal information is preprocessed to extract at least one of its time-domain statistical features, frequency-domain power spectrum features, or nonlinear dynamic features. The extracted biological signal features are then used as part of the structured features corresponding to the awake period information and participate in the feature fusion processing in step S3.

[0014] Specifically, the feature fusion processing in step S3 includes: The structured features corresponding to the lucidity information and the multidimensional dream features are subjected to feature splicing, weighted splicing, or feature interaction fusion based on attention mechanism; wherein, the weights in the weighted splicing are set based on feature importance analysis or domain knowledge, and the feature interaction based on attention mechanism is used to dynamically calculate the correlation strength between different lucidity information features and different dream features.

[0015] Specifically, the state feature analysis model in step S4 is an ensemble learning model based on a multilayer perceptron classifier, a support vector machine, a random forest, or a deep neural network model; the training objective of the model is to minimize the difference loss between its output state feature quantification evaluation result and the sample's true state label.

[0016] According to a second aspect of the present invention, a state feature analysis system based on waking-time information and dream characteristics is provided, comprising: The lucidity information processing module is used to acquire multi-source lucidity information of the target individual and perform structured processing to generate lucidity information features; The dream feature analysis module is used to acquire the dream text data of the target individual and perform feature analysis to extract multidimensional dream features; The feature fusion module, connected to the lucid information processing module and the dream feature parsing module, is used to fuse the lucid information features with the multidimensional dream features to generate a joint feature vector; The state analysis module is connected to the feature fusion module. It integrates a pre-trained state feature analysis model to receive the joint feature vector and calculate and generate a state feature quantitative evaluation result. The result output module is connected to the state analysis module and is used to output the quantitative evaluation results of the state characteristics.

[0017] Beneficial effects: Compared with the prior art, the technical solution provided by the present invention has at least the following beneficial effects: First, this invention achieves a deep and systematic joint analysis of multi-source information during waking hours and dream features, significantly expanding the information dimensions and analytical depth of state assessment. Traditional methods often rely on a single data source (such as questionnaires), while this invention integrates waking-hour information, including individual narrative texts, daily behaviors, social interactions, information exposure, and even objective physiological signals, and fuses them with multi-dimensional features extracted from dreams, such as emotions, themes, structural complexity, and imagery. This information fusion mechanism, which transcends consciousness and subconsciousness, and subjective reports and objective data, provides a more three-dimensional and comprehensive perspective for understanding an individual's overall psychological state, revealing deeper state-related patterns beneath surface statements.

[0018] Secondly, it provides specific, quantifiable, and actionable status assessment outputs, greatly enhancing the practicality and interpretability of the results. This invention goes beyond mere "joint analysis," further outputting quantitative results in concrete forms such as status characteristic scores, multidimensional feature vectors, risk assessment levels, trend indicators, and summaries of life status recommendations. This allows users (including individuals and professionals) to intuitively and quickly grasp the overall level of their status, core problem dimensions, long-term trends, and potential risk levels, obtaining concrete improvement guidance. This effectively transforms technical analysis results into understandable and actionable reference information, facilitating long-term monitoring and targeted intervention.

[0019] Third, it supports dynamic monitoring and trend early warning based on time-series data, achieving a leap from static assessment to process management. By continuously collecting serialized data over multiple time periods and performing modeling analysis, this invention can identify the slow evolution trend, periodic fluctuation patterns, and key turning points of an individual's state, and even predict the state in the short term. This capability is of great value for early warning, intervention effect evaluation, and personalized management in the field of mental health, enabling the solution to serve proactive and preventative health management, rather than passive and reactive problem response.

[0020] Fourth, by integrating objective physiological signals with subjective textual reports, the reliability and verifiability of the assessment results are enhanced. Introducing biosignal features such as electroencephalography (EEG), electrocardiography (ECG), and conductance of skin provides physiological evidence for state assessment independent of subjective statements. When subjective feelings and physiological indicators are inconsistent, this contradiction itself can serve as an important analytical signal, suggesting potential issues that may require further attention. This reduces the bias that may arise from relying solely on subjective reports and improves the objectivity and robustness of the overall assessment.

[0021] Fifth, the technical solution of this invention features excellent modular design and scalability. Each module of the method and system has a clear function and can employ various technical paths, from feature splicing and weighted fusion to attention-based interactive fusion, while also being compatible with multiple machine learning models as the core of analysis. This design allows the solution to flexibly adapt to different application scenarios, data conditions, and technological iterations. For example, it can easily integrate updated natural language processing models to improve dream interpretation accuracy or incorporate new wearable device data sources, ensuring the technology's continued vitality and broad applicability. Attached Figure Description

[0022] Figure 1 This is a flowchart illustrating the state feature analysis method based on waking-time information and dream characteristics provided in a specific embodiment of the present invention. Detailed Implementation

[0023] To enable those skilled in the art to better understand the technical solutions of the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Based on the embodiments in this application, other similar embodiments obtained by those skilled in the art without creative effort should all fall within the scope of protection of this application. Furthermore, directional terms mentioned in the following embodiments, such as "up," "down," "left," and "right," are only for reference to the directions in the accompanying drawings; therefore, the directional terms used are for illustrative purposes and not for limiting the invention.

[0024] First, it should be noted that the technical solution involved in this invention is a component module of the DreamScore framework. DreamScore is a comprehensive artificial intelligence technology architecture proposed by the inventor for the systematic analysis, prediction, generation, and quantitative evaluation of dream narratives.

[0025] The present invention will be further described below with reference to the accompanying drawings and preferred embodiments.

[0026] Please see Figure 1 This invention provides a state feature analysis method based on waking-time information and dream characteristics, including: S1. Obtain multi-source lucidity information related to the target individual, and convert the lucidity information into structured data that can be processed by a computer.

[0027] It should be further explained that this method provides a comprehensive analytical framework that integrates an individual's waking experiences with dream content during sleep, used to assist in assessing mental state. The entire process is executed by a dedicated application on a backend server or terminal device. In step S1, the multi-source waking information is mainly obtained through user-initiated text logs, automatically collected mobile device sensor data, wearable device data, and user-authorized social media message summaries. For example, a user recording a text log such as "I had an argument with a colleague today, felt anxious and frustrated, and worked late tonight" through a mobile application is one type of waking information. The obtained raw information needs to be cleaned (e.g., removing irrelevant characters and correcting typos) and standardized, ultimately converted into structured feature vectors or tensors for subsequent calculations.

[0028] S2. Obtain the dream text data of the target individual, and perform feature analysis on the dream text data to extract multi-dimensional dream features.

[0029] It should be further explained that in step S2, the dream text data comes from the dream content reported by the user after waking up via voice, text, or structured questionnaire. For example, a user reports, "I dreamed that I was running in a maze, unable to find the exit, and felt very panicked." Natural language processing is performed on this text data to extract deep features, rather than simple keywords. The specific parsing process includes word segmentation, part-of-speech tagging, and dependency parsing, and based on this, specific feature extraction is performed. The specific design includes: (1) Emotion feature extraction: The overall emotional valence (positive, negative, neutral) and intensity of dream text are analyzed using a pre-trained sentiment analysis model (such as a BERT-based model). At the same time, the density and change sequence of emotional words in the text can be analyzed.

[0030] (2) Topic feature extraction: The Latent Dirichlet Distribution (LDA) topic model is used to train several topics (e.g., “chasing and running away”, “falling”, “flying”, “exam”, “searching”) from a large amount of dream text corpus. Then, the probability distribution of the current dream text on each topic is calculated to form the topic feature vector.

[0031] (3) Narrative structure complexity feature extraction: By identifying events (usually represented by verb phrases) and temporal connectors in the dream text, an event chain is constructed. Narrative structure complexity ( This can be achieved by calculating the number of events in the event chain. ), the number of transitional relationships between events ( (such as turning points introduced by "but" or "suddenly") and the number of sub-narrative lines ( This is used for comprehensive evaluation. An example calculation formula is: ,in For the number of events, The number of adversative relationships. For the number of sub-narrative lines, , , As a weighting coefficient, in a preferred embodiment, the values ​​can be 0.5, 0.3, and 0.2 respectively. This weighting setting is based on a subjective evaluation experiment on the impact on narrative coherence, emphasizing the number of core events.

[0032] (4) Semantic coherence feature extraction: The semantic coherence between adjacent dream description sentences is evaluated by calculating the cosine similarity of sentence embeddings (such as using the Sentence-BERT model) and the average coherence score is obtained.

[0033] (5) Image element feature extraction: Based on the pre-constructed dream image dictionary (containing common images such as "water", "fire", "snake", "flying", "tooth loss" and their common psychological mappings), the specific images appearing in the dream text and their frequency are statistically analyzed.

[0034] S3. Perform feature fusion processing on the structured features corresponding to the lucidity information and the multidimensional dream features to generate a joint feature vector for representing the individual's overall state.

[0035] It should be further explained that in step S3, feature fusion aligns and combines waking-time structured features (e.g., daytime mood ratings, stress event encodings) with dream-time multidimensional features (e.g., dream emotion intensity, specific theme probability) in terms of feature dimensions. A basic approach is direct concatenation to form a high-dimensional joint feature vector. A more advanced approach can employ attention-based fusion, allowing the model to learn the interactive importance between waking-time features and dream-time features.

[0036] S4. Input the joint feature vector into the pre-trained state feature analysis model to calculate the state feature quantitative evaluation result; wherein, the state feature analysis model is trained based on the lucidity information features, dream features and corresponding labeled state data of the sample individual.

[0037] It should be further explained that in step S4, the state feature analysis model is a supervised machine learning model. Training this model requires a large amount of sample data, with each sample including: the waking-time information features of the sample individual and dream features as input features ( ), and status labels (such as "mild anxiety", "increased risk of depression", "stable state") given by professional assessors (such as mental health counselors) based on standard scales (such as the PHQ-9 depression scale, GAD-7 anxiety scale) or comprehensive diagnoses. The model learns the mapping function. This establishes the correlation between fused features and state assessment results. In the application phase, the joint feature vector of the target individual is input into this trained model to obtain the corresponding state feature quantification assessment results.

[0038] S5. Output the quantitative evaluation results of the state characteristics.

[0039] It should be further noted that in step S5, the output results can be presented to the user or professional physician in the form of a visual report, score, grade or text description.

[0040] Understandably, this method is the first to systematically calculate and integrate objective / subjective information from the waking period with dream content characteristics, breaking through the limitations of single data source analysis. Dreams, as a window into the subconscious, have an intrinsic connection between their content and an individual's waking psychological state, but traditional assessment methods often overlook this dimension. This method, by quantitatively analyzing dream characteristics and combining them with waking information, can more comprehensively and earlier capture clues to psychological state fluctuations that individuals might otherwise overlook. It provides a novel, multi-dimensional data fusion perspective for mental state assessment, improving the comprehensiveness of the assessment and its potential early warning capabilities.

[0041] Specifically, the quantitative assessment results of the state characteristics include at least one of the following: state characteristic score, state characteristic vector, state change trend index, risk assessment level information, or summary of life status recommendations; wherein, the state characteristic score is obtained by fusing the calculation results of each dimension sub-feature in the joint feature vector; the state change trend index is obtained by comparing the difference of the joint feature vector or the change of the state characteristic score between the current time period and the historical time period.

[0042] It should be further explained that this step specifies and quantifies the analysis results, enhancing their practicality and interpretability. The specific design includes: (1) State characteristic score ): is a scalar value, for example, ranging from 0 to 100. It is obtained by analyzing the joint feature vector. The weighted calculation yields the following results: ,in For the weight vector, For bias terms, For activation functions (such as the Sigmoid function, which maps the result to the 0-1 interval and then amplifies it). Weights This can be learned during model training, reflecting the contribution of different feature dimensions to the overall state. A high score may indicate a more positive and stable state, while a low score suggests possible stress or emotional distress.

[0043] (2) State feature vector: a multidimensional vector (e.g.) This contains more dimensions of information than a single rating and can be used for more detailed profile analysis.

[0044] (3) Indicators of state change trends ( ): By comparing the current time State characteristic score Compared to a previous time window (e.g., one week ago) The score is calculated from, for example . A consistently negative value may indicate a downward trend in the state. The degree of change in state structure can also be measured by calculating the cosine similarity or Euclidean distance between joint feature vectors.

[0045] (4) Risk assessment level information: scoring based on status characteristics or trend indicators Risk levels are categorized. For example, the definition is: "Low risk" For the "attention zone", It is designated as a "high-risk area". Or, when When the value is below the threshold of -10 for several consecutive periods, the "Trend Warning" level is triggered.

[0046] (5) Summary of life status suggestions: Based on the analysis results and the pre-set rule base or generative model, the text suggestions are automatically output. For example, if the theme of "being chased" is highly probable in the dream features and there are many stressful events during the day, and the assessment is a high-pressure state, it may be suggested that "Recent stress has been detected. It is recommended to try mindfulness breathing exercises and arrange appropriate leisure activities."

[0047] Understandably, quantifying the analysis results into specific scores, vectors, trend indicators, risk levels, or summary recommendations makes the assessment results intuitive, easy to understand, and traceable. Users or professionals can quickly grasp the overall situation, identify changing trends, clarify risk levels, and obtain concrete action guidance, greatly enhancing the practical value and user experience of the technical solution and facilitating long-term monitoring and intervention.

[0048] Specifically, the lucidity information includes at least one of the following: individual experience narrative text, emotional state record text, daily life behavior log, social relationship and interpersonal interaction record, information exposure or media exposure content; step S1 further includes: semantic embedding, topic modeling, and emotional polarity and intensity analysis of text-based lucidity information, and event encoding and frequency statistics of behavioral record information, in order to generate structured features corresponding to the lucidity information.

[0049] It should be further explained that this step defines in detail the multivariate composition of lucid information and its structured processing methods, specifically including: (1) Individual experience narrative text: refers to the description of important events of the day recorded by users in diary form, such as "I completed the project presentation today. Although I was nervous, the result was well received." In processing, semantic embedding is first performed, using a pre-trained language model (such as BERT) to convert the text into fixed-dimensional semantic vectors. Simultaneously, event extraction can be performed to identify and encode core verb phrases (such as "complete the report" and "receive praise").

[0050] (2) Emotional state record text: refers to the emotional words or sentences directly recorded by the user, such as "feeling happy" or "somewhat tired and irritable". When processing, sentiment analysis tools are used to identify the emotional polarity (positive, negative, neutral) and intensity score (e.g., from 0 to 1). Emotional diversity (the number of different emotional types recorded in a day) can also be further analyzed.

[0051] (3) Daily life behavior log: This includes activities manually recorded through a mobile application or automatically inferred by sensors, such as "sleep duration of 7 hours", "walking 8000 steps", and "screen time of 6 hours". During processing, various behaviors are coded (e.g., sleep is coded as A01 and exercise is coded as A02), and their frequency, duration and other numerical characteristics are statistically analyzed.

[0052] (4) Social Relationships and Interpersonal Interaction Records: This includes call records (frequency, duration, contact type), metadata of SMS / social application messages (sending frequency, number of interaction objects), or brief user descriptions, such as "I talked to my parents on the phone today and had dinner with friends." During processing, features such as the scale of the social network, interaction frequency, and proportion of intimate interactions are extracted.

[0053] (5) Information exposure or media exposure content: This refers to news headlines viewed by users, video type tags watched, etc. For example, browsing history may include "economic recession concerns" or "natural disaster reports". During processing, the content can be classified by theme and sentiment analysis can be performed to count the proportion of negative information encountered by users or the exposure of specific theme content.

[0054] For all text-based information, the general structured process includes: word segmentation, stop word removal, and part-of-speech tagging; generating text vectors using the Bag-of-Words model or TF-IDF; or directly generating sentence vectors using deep learning models (such as Sentence-BERT). For more structured data such as behavioral logs, numerical and normalization processes are performed directly, and finally, all features are concatenated into a structured feature vector of lucidity information. .

[0055] Understandably, this solution clarifies and broadens the data boundaries of "awake period information," encompassing multiple levels from subjective emotions to objective behaviors, from personal activities to social interactions, and from active recording to passive exposure. By deeply structuring this heterogeneous information and transforming it into a unified and computable feature representation, a solid data foundation is laid for subsequent deep integration with dream features, ensuring the richness and multidimensionality of the input information for the state assessment model.

[0056] Specifically, the multidimensional dream features include at least two of the following extracted from the dream text data: emotional features, theme features, narrative structure complexity features, semantic coherence features, and imagery element features; wherein, the emotional features include emotional valence and emotional intensity sequence, the theme features are extracted through a theme model, and the narrative structure complexity features are calculated based on the completeness of the event chain and the number of turning points.

[0057] It should be further explained that this step provides a multi-dimensional and in-depth depiction of dream characteristics, going beyond a simple content summary. Specific design elements include: (1) Emotional characteristics: Not only is the overall emotional tone (valence) of the dream analyzed, but also the dynamic changes in emotion are considered. For example, an emotional intensity sequence can be extracted from a sequence of sentences describing a dream. ,in Representing the The emotional intensity value of a sentence or passage. Emotional volatility ( This can be measured by calculating the standard deviation of the sequence. in Let i be the emotional intensity of the i-th segment. For average emotional intensity, The number of segments. High price combined with high volatility may reflect inner contradictions.

[0058] (2) Topic Features: Using topic models such as LDA, K topics (e.g., K=20) are learned unsupervised from a large corpus of dream texts. Each topic is represented by a distribution of terms with high probabilities. For a specific dream text, its probability vector distribution on the K topics is inferred. This vector represents the thematic feature. This captures the macroscopic patterns of dream content.

[0059] (3) Narrative structure complexity characteristics: This characteristic reflects the logic and organization of the dream story. Specific calculations may include: a) Event Chain Completeness: Identify specific events (actions) mentioned in the dream and check if they form a chain with a cause, process, and result (which may be missing). Completeness Score ( The rating can be determined based on the number of event stages identified.

[0060] b) Number of turning points ( ): The number of words or situations that indicate a sudden change in the dream narrative.

[0061] c) Narrative structure complexity The above factors can be taken into account: in, Score the completeness of the event chain. The number of turning points , The weights can be set to 0.6 and 0.4 in a preferred embodiment, emphasizing that the integrity of the event chain has a greater impact on the overall structure. An unusually complex or fragmented structure may correspond to a specific state of mind.

[0062] (4) Semantic coherence features: Calculate the semantic similarity between adjacent sentences or fragments in the dream text, and calculate the average coherence score. Low coherence may reflect scattered thinking.

[0063] (5) Imagery element features: Based on common dream images associated with specific psychological states in psychological research (such as "water" which may be related to emotions, and "falling" which may be related to a sense of loss of control), an imagery dictionary is constructed. Features are extracted as binary existence vectors or frequency vectors of specific images appearing in the dream text.

[0064] Understandably, by extracting multi-dimensional features such as emotion, theme, structure, coherence, and imagery, this solution achieves a comprehensive quantitative analysis of dream content, encompassing emotion, content, form, and symbolic meaning. This deep feature extraction transforms dreams from vague subjective experiences into a set of calculable and analyzable objective indicators, enabling precise correlation analysis with waking information and significantly enhancing the depth and psychological basis of state assessment.

[0065] Specifically, the method further includes: acquiring multiple sets of lucidity information and corresponding dream text data of the target individual in multiple consecutive time periods; in step S3, generating corresponding joint feature vectors for each time period and arranging them in chronological order to form a joint feature vector sequence; step S4 includes: inputting the joint feature vector sequence into the state feature analysis model, analyzing the temporal evolution pattern of state features, and outputting a state feature analysis report reflecting long-term change trends.

[0066] It should be further explained that this step introduces a time series analysis dimension, expanding single-point analysis into longitudinal tracking, which is crucial for status monitoring and trend prediction.

[0067] In practice, the system continuously collects users' daily / weekly waking-time information and dream reports. Assuming a "day" as the basic time period, data is collected continuously for T days (e.g., T=30). For each day... Perform steps S1 to S3 to generate a joint feature vector for the current day. .

[0068] Arranging these vectors in chronological order yields a time series: This sequence depicts the trajectory of changes in the user's overall state characteristics over time.

[0069] In step S4, the state feature analysis model needs to be able to process time-series data. One implementation is to keep the base model (such as the model used to output daily scores) unchanged, generating a daily state score each day. Then these rating sequences Perform time series analysis, such as calculating moving averages and detecting trend lines (using linear regression to fit the slope). ), or identify periodic patterns. If A significantly negative value indicates a trend of deterioration.

[0070] A more advanced implementation uses models specifically designed for sequence data (such as Recurrent Neural Networks (RNNs), Long Short-Term Memory Networks (LSTMs), and Transformer encoders) as the core component of the state feature analysis model. This model directly integrates the joint feature vector sequence. As input, the model learns the temporal dependencies and outputs a summary analysis of the entire sequence. For example, the model might output: ① Trend judgment: such as "the situation has been stable in the past week" or "it has shown a slow downward trend in the past month".

[0071] ②Key inflection point detection: Identify the dates on which the state undergoes a significant change.

[0072] ③ Periodic reports: Summarize the weekly status fluctuation patterns (e.g., "Weekend status is generally better than weekday status").

[0073] The final analysis report can integrate daily scores, trend charts, inflection point alerts, and text summaries to provide users with a dynamic and comprehensive view of the situation.

[0074] Understandably, by introducing multi-time-series analysis, this solution can be upgraded from static snapshot assessment to dynamic process monitoring. This is invaluable for identifying slowly evolving state changes, providing early warnings of potential risks, and evaluating the effectiveness of interventions (such as psychological counseling and lifestyle modifications). It enables this technological solution to serve long-term mental health management, rather than just a single state check.

[0075] Specifically, the multiple time periods include cycles in units of days, weeks, or months; the state feature analysis model is a recurrent neural network, a long short-term memory network, or a temporal convolutional network, used to capture long-term dependencies and periodic patterns in the joint feature vector sequence and predict future state feature trends.

[0076] It should be further explained that this step clarifies the specific period selection and applicable high-level model architecture for time series analysis. The specific design includes: (1) Period selection: a) Using a "day" cycle: Best suited for high-frequency monitoring, capable of capturing fine-grained daily fluctuations, but requires high user compliance with daily recording. Daily data forms the basis for building longer-term trends.

[0077] b) Using a "week" cycle: Aggregate a week's worth of data (e.g., 7 daily feature vectors) to generate a weekly feature vector (e.g., take the mean, maximum, and minimum values). This helps smooth out random fluctuations during the day and highlights weekly rhythmic patterns of change (e.g., the weekday-weekend effect), and is a common choice for balancing details and trends.

[0078] c) Using a "monthly" cycle: Further aggregate weekly or daily data to observe longer-term, slow-moving trends, suitable for quarterly or annual reviews.

[0079] In a preferred embodiment, a two-level analysis of "daily-weekly" can be adopted: basic analysis is performed daily to generate daily indicators, while a trend analysis report based on weekly aggregated data is generated weekly.

[0080] (2) Temporal Model Selection and Prediction: The state feature analysis model can employ a Long Short-Term Memory (LSTM) network. LSTM units utilize their gating mechanism (input gate)... Forgotten Gate Output gate and cell state This approach can effectively learn long-term dependencies in a sequence. Given a joint feature vector sequence of length T as input... LSTM processes recursively step by step, and finally the hidden state at the last time step. Alternatively, aggregation of all hidden states (such as averaging) can encode information about the entire sequence for trend classification or regression prediction.

[0081] Specifically, the sequence can be input into an LSTM network to obtain the hidden state at the last time step. Then, through a fully connected layer and an activation function, it outputs a trend category (such as "rising", "falling", "stable") or a state score prediction for a future time period (such as the following week). .

[0082] Furthermore, predict future state scores The model training objective is to minimize the mean squared error (MSE) between the predicted and actual values ​​(if any). This predictive capability enables the solution to be forward-looking, providing early warnings before a significant decline in user status is anticipated.

[0083] Understandably, clearly defining the period selection and introducing time series models such as LSTM provide a solid technical path for the time series analysis capabilities of this solution. Period selection offers flexibility, adapting to different monitoring needs. The application of models such as LSTM enables the system to automatically learn complex time series patterns of state changes and even make predictions. This elevates state assessment and intervention from passive response to proactive early warning, greatly enhancing the practicality and forward-looking value of the solution.

[0084] Specifically, the lucidity information also includes biosignal information corresponding to the target individual, the biosignal information including at least one of electroencephalogram (EEG), electrocardiogram (ECG), or electrodermal signal; the method further includes: preprocessing the biosignal information to extract at least one of its time-domain statistical features, frequency-domain power spectrum features, or nonlinear dynamic features, and using the extracted biosignal features as part of the structured features corresponding to the lucidity information to participate in the feature fusion processing in step S3.

[0085] It should be further explained that this step further expands the scope of information during the lucid period, incorporating objective physiological signals and achieving a combination of "subjective reporting" and "objective physiological indicators".

[0086] Furthermore, biosignals are obtained through wearable devices (such as smart bracelets, EEG headbands, and heart patches) during specific time periods or through continuous monitoring. Specific designs include: (1) Electroencephalogram (EEG) signals: Preprocessing includes filtering (e.g., 0.5-45Hz bandpass filtering for noise reduction), removal of electrooculogram artifacts, and segmentation. Feature extraction includes: a) Time-domain characteristics: such as signal mean, variance, peak-to-peak value.

[0087] b) Frequency domain characteristics: Calculate the power spectral density (PSD) or relative power (power in a certain frequency band / total power) for each classic frequency band (δ wave: 1-4Hz, θ wave: 4-8Hz, α wave: 8-13Hz, β wave: 13-30Hz, γ wave: >30Hz). For example, the relative power of α wave is related to the relaxed state.

[0088] c) Nonlinear characteristics: such as approximate entropy and sample entropy, used to characterize the complexity of EEG signals.

[0089] (2) Electrocardiogram (ECG) signal: Preprocessing includes R-wave detection and heart rate variability (HRV) analysis. Feature extraction includes: a) Time-domain characteristics: such as mean heart rate, standard deviation of RR interval (SDNN), and root mean square of the difference between adjacent RR intervals (RMSSD). SDNN reflects overall autonomic nervous activity, while RMSSD is more correlated with parasympathetic nervous activity.

[0090] b) Frequency domain characteristics: The low-frequency power (LF: 0.04-0.15Hz), high-frequency power (HF: 0.15-0.4Hz), and their ratio (LF / HF) of HRV were obtained through spectral analysis. HF is related to parasympathetic nerve activity, and LF / HF can reflect the balance between the sympathetic and parasympathetic nervous systems.

[0091] (3) Electrodermal conductance signal (EDA / GSR): Reflects skin conductance level (SCL) and skin conductance response (SCR). Features include: baseline SCL, number of SCR occurrences, and total SCR amplitude, which are closely related to emotional arousal.

[0092] After extracting the aforementioned biosignal features, they are normalized together with features from text, behavioral logs, etc., and then concatenated to form a more comprehensive structured feature vector of lucid information. For example, a feature vector might contain: [daytime average heart rate, SDNN of HRV, alpha wave relative power, number of skin conductance responses, daytime emotional text score, ...].

[0093] Understandably, incorporating biosignal features such as electroencephalography (EEG), electrocardiography (ECG), and electrodermal conductance (EDC) provides objective, physiological evidence for state assessment. Changes in psychological state are often accompanied by alterations in the autonomic nervous system and central nervous system activity. Combining these physiological indicators with subjective dream and narrative reports allows for multimodal verification and complementarity. For example, a user may subjectively report feeling "calm," but their HRV data may show increased sympathetic nerve activity; this contradiction itself could be a noteworthy signal. This fusion of subjective and objective data significantly enhances the objectivity, reliability, and depth of the assessment results.

[0094] Specifically, the feature fusion processing in step S3 includes: performing feature splicing, weighted splicing, or attention-based feature interaction fusion on the structured features corresponding to the waking period information and the multidimensional dream features; wherein, the weights in the weighted splicing are set based on feature importance analysis or domain knowledge, and the attention-based feature interaction is used to dynamically calculate the correlation strength between different waking period information features and different dream features.

[0095] It should be further explained that this step elaborates on various technical means for effectively integrating two types of heterogeneous information (awake state and dream state), and the specific design includes: (1) Feature Concatenation: The most direct method. Assume the feature vector for the lucid period is... The dream feature vector is Then the joint eigenvector This method preserves all the original information but does not take into account the interactions between features.

[0096] (2) Weighted splicing: Before splicing, weights are assigned to different feature dimensions to reflect their prior importance. ,in , It is a weight scalar or a diagonal weight matrix. The weights can be determined based on feature selection methods (such as ranking features based on model importance) or set by domain knowledge (for example, recent daytime events are considered more important than long-term dream features and can be assigned higher weights).

[0097] (3) Feature interaction based on attention mechanism: This is a more advanced fusion method. The core idea is to enable the model to dynamically learn which wakefulness features are more relevant to which dream features in a specific context. Specific designs include: a) First, and Each vector is mapped to the same space through a linear transformation to obtain the query vector. Key vector Sum value vector Typically, it can be Considered a query , Consider as key Sum .

[0098] b) Calculate attention score: ,in This refers to the dimension of the key vector. Attention weight matrix. Each element in the equation represents the strength of the correlation between a certain dream feature dimension and a certain lucidity feature dimension.

[0099] c) The attention-weighted waking-time feature representation obtained is then combined with the original dream features (e.g., splicing or adding) to form the final joint feature representation. .

[0100] This approach makes fusion no longer a simple superposition, but rather a targeted information interaction and filtering based on the specific content of each input (for example, a dream about "exams" may focus more on features related to "work stress" and "ability assessment" in the waking information), which theoretically can achieve stronger representational capabilities.

[0101] Understandably, the provision of various feature fusion strategies, ranging from simple to complex, makes the solution flexible and scalable. The introduction of weighted splicing and attention mechanisms, especially the attention mechanism, can simulate the subconscious focus of human experts on certain key associations during analysis (e.g., linking specific dream images with recent specific life events). This data-driven interactive fusion significantly improves the model's ability to capture complex lucid-dream associations, potentially uncovering deeper and more personalized state association patterns, which is a key technical aspect for improving assessment accuracy.

[0102] Specifically, the state feature analysis model in step S4 is an ensemble learning model based on a multilayer perceptron classifier, a support vector machine, a random forest, or a deep neural network model; the training objective of the model is to minimize the difference loss between its output state feature quantification evaluation result and the sample's true state label.

[0103] It should be further explained that this step clarifies the model types and their training paradigms that can be used for final state analysis, and the specific design includes: (1) Model type selection: a) Multilayer Perceptron (MLP): A basic feedforward neural network consisting of an input layer, several hidden layers, and an output layer. Suitable for joint feature vectors. As input, the output can be a classification result (such as a risk level) or a regression result (such as a state score). MLPs are able to learn non-linear relationships between features.

[0104] b) Support Vector Machine (SVM): Suitable for small sample data, it performs classification by finding the maximum margin hyperplane. For regression tasks, Support Vector Regression (SVR) can be used.

[0105] c) Random Forest: An ensemble learning algorithm that improves generalization and robustness by constructing multiple decision trees and combining their results (voting or averaging), and can provide feature importance assessment.

[0106] d) Deep neural network models: such as more complex fully connected networks, or Transformer encoders that incorporate attention mechanisms. Deep models have the strongest representation learning capabilities when there is sufficient data.

[0107] In a preferred embodiment, an MLP can be constructed that includes two fully connected hidden layers (e.g., a first layer with 128 neurons and a second layer with 64 neurons, using the ReLU activation function), and the output layer uses one neuron (using the Sigmoid activation function for regression scoring) or multiple neurons (using the Softmax activation function for classification).

[0108] (2) Model training: The training objective is to minimize the loss function. For regression tasks (predicting scores), the mean squared error loss (MSE) is commonly used: ,in For the sample size, Score the true state of the i-th sample. The score predicted by the model. For classification tasks (predicted ratings), cross-entropy loss is commonly used: ,in For the number of categories, It is a sample True category label (one-hot encoded) The model predicts the sample. Category The probability of loss. The training process iteratively updates the model parameters through backpropagation and an optimizer (such as Adam) to minimize the loss. .

[0109] (3) Model Validation and Evaluation: Use independent validation and test sets to evaluate model performance. For regression, R² score and mean absolute error (MAE) can be used; for classification, accuracy, precision, recall, F1 score, and other metrics can be used. It is essential to ensure that the model is not overfitting and has good generalization ability.

[0110] Understandably, clarifying the specific implementation and training methods of the model enables those skilled in the art to reproduce and train the state feature analysis model. Through supervised learning, the model can automatically learn a complex mapping function from fused features to state evaluation results from a large number of labeled "wake-dream-state" data pairs, avoiding the tediousness and subjectivity of manually formulating rules. The availability of multiple model options also provides flexibility for applications under different data scales and scenarios, ensuring the technical feasibility of the solution.

[0111] This invention provides another embodiment, which offers a state feature analysis system based on waking-time information and dream characteristics. The state feature analysis system based on waking-time information and dream characteristics includes: (1) The lucidity information processing module is used to obtain multi-source lucidity information of the target individual and perform structured processing to generate lucidity information features.

[0112] It should be further noted that this system implements the aforementioned methods in a modular manner, allowing deployment on cloud servers, local computers, or mobile terminals. The lucidity information processing module includes a data acquisition interface and a data processing unit. The data acquisition interface is responsible for collecting raw lucidity information data from multiple sources, including user terminal applications, wearable devices, and authorized third-party APIs (such as limited data interfaces on social media). The data processing unit performs data cleaning, format conversion, and various structured analyses described above (such as text embedding, sentiment analysis, behavioral coding, and physiological signal feature extraction), ultimately outputting a standardized lucidity information feature vector.

[0113] (2) Dream feature analysis module, used to obtain the dream text data of the target individual and perform feature analysis to extract multidimensional dream features.

[0114] It should be further explained that the dream feature parsing module includes a dream text input interface and a feature parsing engine. The input interface receives dream text submitted by the user (speech must first be converted to text via speech recognition). The feature parsing engine integrates a natural language processing pipeline, including a word segmenter, sentiment analysis model, topic model (such as LDA), dependency parser, and dream imagery dictionary, etc., and sequentially calculates multi-dimensional features such as emotion, theme, structure, coherence, and imagery, outputting a dream feature vector.

[0115] (3) Feature fusion module, connected to the lucidity information processing module and the dream feature analysis module, is used to fuse the lucidity information features and the multidimensional dream features to generate a joint feature vector.

[0116] It should be further explained that the feature fusion module receives feature vectors from the first two modules. Depending on the configuration, it performs simple vector concatenation, weighted concatenation, or runs an attention fusion subnetwork. This module is responsible for unifying heterogeneous features into a single representation space and outputting a fused joint feature vector.

[0117] (4) State analysis module, connected to the feature fusion module, which integrates a pre-trained state feature analysis model to receive the joint feature vector and calculate and generate state feature quantification evaluation results.

[0118] It should be further explained that the state analysis module is the core computational unit of the system. It loads a pre-trained state feature analysis model (such as MLP, SVM, random forest, or neural network model). This module receives the joint feature vector, calls the model to perform forward propagation calculations, and generates quantitative evaluation results of state features (scores, vectors, ranks, etc.). This module may also contain a time series analysis submodule for processing sequence inputs and performing trend calculations.

[0119] (5) Result output module, connected to the state analysis module, used to output the state feature quantitative evaluation result.

[0120] It should be further explained that the results output module is responsible for outputting the results generated by the status analysis module to the user through a user interface (such as a mobile application front-end or a web interface) in the form of visual charts (trend charts, radar charts), score displays, level indicators (such as color codes: green / yellow / red), or natural language reports. Simultaneously, it can also generate structured data reports for further analysis by professionals.

[0121] Furthermore, the various modules communicate and transfer data with each other through predefined application programming interfaces (APIs) or internal data buses, working together to complete the entire process from data input to result output.

[0122] Understandably, this system concretizes the methodology into collaborative hardware and software modules, clearly defining the system's architecture and division of labor. This modular design facilitates system development, maintenance, and expansion. For example, the NLP model in the dream feature analysis module can be upgraded independently without affecting other modules. Clear module division also facilitates distributed deployment, such as deploying computationally intensive model inference in the cloud and placing data acquisition and simple preprocessing on user terminals. This system provides a complete, efficient, and engineerable solution for implementing the aforementioned methods.

[0123] In a preferred embodiment, this application also provides an electronic device, the electronic device comprising: The computer device includes a memory and a processor, wherein the memory stores computer-readable instructions that, when executed by the processor, implement the state feature analysis method based on waking information and dream characteristics. The computer device can be broadly categorized as a server, terminal, or any other electronic device with the necessary computing and / or processing capabilities. In one embodiment, the computer device may include a processor, memory, network interface, communication interface, etc., connected via a system bus. The processor of the computer device can be used to provide the necessary computing, processing, and / or control capabilities. The memory of the computer device may include a non-volatile storage medium and internal memory. The non-volatile storage medium may store an operating system, computer programs, etc. The internal memory can provide an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface and communication interface of the computer device can be used to connect and communicate with external devices via a network. When the computer program is executed by the processor, it performs the steps of the method of the present invention.

[0124] This invention can be implemented as a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, causes the steps of the methods of embodiments of the invention to be performed. In one embodiment, the computer program is distributed across multiple network-coupled computer devices or processors, such that the computer program is stored, accessed, and executed in a distributed manner by one or more computer devices or processors. A single method step / operation, or two or more method steps / operations, may be executed by a single computer device or processor or by two or more computer devices or processors. One or more method steps / operations may be executed by one or more computer devices or processors, and one or more other method steps / operations may be executed by one or more other computer devices or processors. One or more computer devices or processors may execute a single method step / operation, or execute two or more method steps / operations.

[0125] Those skilled in the art will understand that the method steps of this invention can be performed by a computer program instructing related hardware, such as a computer device or processor, to perform the steps of this invention when executed. Depending on the context, any references herein to memory, storage, databases, or other media may include non-volatile and / or volatile memory. Examples of non-volatile memory include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), flash memory, magnetic tape, floppy disk, magneto-optical data storage device, optical data storage device, hard disk, solid-state drive, etc. Examples of volatile memory include random access memory (RAM), external cache memory, etc.

[0126] The technical features described above can be combined arbitrarily. Although not all possible combinations of these technical features are described, any combination of these technical features should be considered to be covered by this specification, provided that such combination does not contain contradictions.

[0127] The specific embodiments of the present invention described above do not constitute a limitation on the scope of protection of the present invention. Any other corresponding changes and modifications made in accordance with the technical concept of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A method for analyzing state characteristics based on waking-time information and dream features, characterized in that, include: S1. Obtain multi-source lucidity information related to the target individual, and convert the lucidity information into structured data that can be processed by a computer; S2. Obtain the dream text data of the target individual, and perform feature analysis on the dream text data to extract multi-dimensional dream features; S3. Perform feature fusion processing on the structured features corresponding to the lucidity information and the multidimensional dream features to generate a joint feature vector for representing the individual's overall state; S4. Input the joint feature vector into the pre-trained state feature analysis model to calculate the state feature quantitative evaluation result; wherein, the state feature analysis model is trained based on the waking-time information features, dream features and their corresponding labeled state data of the sample individual; S5. Output the quantitative evaluation results of the state characteristics.

2. The state feature analysis method based on waking-time information and dream characteristics according to claim 1, characterized in that, The quantitative assessment results of the state characteristics include at least one of the following: state characteristic score, state characteristic vector, state change trend index, risk assessment level information, or summary of life status recommendations; wherein, the state characteristic score is obtained by fusing the calculation results of each dimension sub-feature in the joint feature vector; the state change trend index is obtained by comparing the difference of the joint feature vector or the change of the state characteristic score between the current time period and the historical time period.

3. The state feature analysis method based on waking-time information and dream characteristics according to claim 1, characterized in that, The lucidity information includes at least one of the following: individual experience narrative texts, emotional state record texts, daily life behavior logs, social relationship and interpersonal interaction records, and information exposure or media exposure content; step S1 further includes: Semantic embedding, topic modeling, and emotion polarity and intensity analysis are performed on text-based lucidity information, and event encoding and frequency statistics are performed on behavioral record information to generate structured features corresponding to the lucidity information.

4. The state feature analysis method based on waking-time information and dream characteristics according to claim 1, characterized in that, The multidimensional dream features include at least two of the following extracted from the dream text data: emotional features, theme features, narrative structure complexity features, semantic coherence features, and imagery element features; wherein, the emotional features include emotional valence and emotional intensity sequence, the theme features are extracted through a theme model, and the narrative structure complexity features are calculated based on the completeness of the event chain and the number of turning points.

5. The state feature analysis method based on waking-time information and dream characteristics according to claim 1, characterized in that, The method further includes: Acquire multiple sets of lucidity information and corresponding dream text data of the target individual over multiple consecutive time periods; In step S3, a joint feature vector is generated for each time period, and the vectors are arranged in chronological order to form a sequence of joint feature vectors. Step S4 includes: inputting the joint feature vector sequence into the state feature analysis model, analyzing the temporal evolution pattern of the state features, and outputting a state feature analysis report reflecting the long-term trend of change.

6. The state feature analysis method based on waking-time information and dream characteristics according to claim 5, characterized in that, The multiple time periods include cycles in units of days, weeks, or months; the state feature analysis model is a recurrent neural network, a long short-term memory network, or a temporal convolutional network, used to capture long-term dependencies and periodic patterns in the joint feature vector sequence and predict future state feature trends.

7. The state feature analysis method based on waking-time information and dream characteristics according to claim 1, characterized in that, The lucidity information also includes biosignal information corresponding to the target individual, the biosignal information including at least one of electroencephalogram (EEG), electrocardiogram (ECG), or electrodermal signal; the method further includes: The biological signal information is preprocessed to extract at least one of its time-domain statistical features, frequency-domain power spectrum features, or nonlinear dynamic features. The extracted biological signal features are then used as part of the structured features corresponding to the awake period information and participate in the feature fusion processing in step S3.

8. The state feature analysis method based on waking-time information and dream characteristics according to claim 1, characterized in that, The feature fusion process in step S3 includes: The structured features corresponding to the lucidity information and the multidimensional dream features are subjected to feature splicing, weighted splicing, or feature interaction fusion based on attention mechanism; wherein, the weights in the weighted splicing are set based on feature importance analysis or domain knowledge, and the feature interaction based on attention mechanism is used to dynamically calculate the correlation strength between different lucidity information features and different dream features.

9. The state feature analysis method based on waking-time information and dream characteristics according to claim 1, characterized in that, The state feature analysis model in step S4 is an ensemble learning model based on a multilayer perceptron classifier, support vector machine, random forest, or deep neural network model; the training objective of the model is to minimize the difference loss between its output state feature quantification evaluation result and the true state label of the sample.

10. A state feature analysis system based on waking-time information and dream characteristics, characterized in that, include: The lucidity information processing module is used to acquire multi-source lucidity information of the target individual and perform structured processing to generate lucidity information features; The dream feature analysis module is used to acquire the dream text data of the target individual and perform feature analysis to extract multidimensional dream features; The feature fusion module, connected to the lucid information processing module and the dream feature parsing module, is used to fuse the lucid information features with the multidimensional dream features to generate a joint feature vector; The state analysis module is connected to the feature fusion module. It integrates a pre-trained state feature analysis model to receive the joint feature vector and calculate and generate a state feature quantitative evaluation result. The result output module is connected to the state analysis module and is used to output the quantitative evaluation results of the state characteristics.

Images