End-stage patient psychological support scene generation and interaction system based on VR technology
By combining facial expressions, eye movements, and verbal text analysis, the VR psychological support scenario is dynamically adjusted, solving the problems of insufficient personalization and monotonous interaction methods in existing systems, and achieving personalized and autonomous psychological support effects.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HUNAN PROVINCIAL TUMOR HOSPITAL
- Filing Date
- 2026-04-08
- Publication Date
- 2026-07-03
AI Technical Summary
Existing VR psychological intervention systems lack personalized adaptation capabilities and cannot dynamically adjust content according to the patient's real-time emotional state. Traditional emotion analysis ignores multimodal data and has a single interaction method, which limits the patient's autonomy and immersion.
The scene acquisition module provides preset indicator scene data, the state acquisition module captures facial and eye state feedback images and language text, the feature extraction module extracts emotional and semantic features, the preference analysis module builds a preference analysis model, the framework definition module defines the dimensions of the AR simulation framework, the data filling module supplements historical data, and the scene construction module builds the interaction response module to realize immersive interaction.
It enables dynamic adjustment of the AR simulation framework based on the patient's real-time emotional state, providing a personalized and coherent psychological support experience, enhancing the patient's freedom of interaction, reducing reliance on professional psychologists, and improving the effectiveness of psychological support.
Smart Images
Figure CN122337511A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of scene simulation and interaction technology, specifically to a VR-based system for generating and interacting with psychological support scenarios for end-stage patients. Background Technology
[0002] While advancements in medical technology have extended the survival time of terminally ill patients, their psychological and emotional needs are often overlooked. Terminally ill patients frequently face negative emotions such as loneliness, anxiety, and depression. Traditional psychological interventions, such as counseling and medication, are limited by manpower, environment, and the patient's physical condition, making it difficult to provide continuous and personalized support. Virtual reality technology, with its immersive experience, demonstrates unique advantages in the medical field, alleviating patients' psychological stress by simulating natural landscapes and interactions with family and friends. Simultaneously, the development of artificial intelligence technology has made affective computing and semantic analysis possible, providing a technological foundation for dynamically understanding patient needs.
[0003] Existing VR psychological intervention systems mostly use fixed scene libraries, lacking personalized adaptation capabilities and unable to dynamically adjust content according to the patient's real-time emotional state, resulting in limited intervention effects; traditional emotion analysis mainly relies on a single modality, ignoring the synergistic effect of multimodal data, affecting the accuracy of preference identification; existing systems usually do not integrate patients' historical preference data, making it difficult to build a coherent personalized experience, and the interaction methods are monotonous, limiting the patient's autonomy and immersion. Summary of the Invention
[0004] (a) Technical problems to be solved
[0005] In view of the above-mentioned shortcomings of the existing technology, the present invention provides a VR-based system for generating and interacting with psychological support scenarios for terminal patients, which can effectively solve the problems of the existing technology.
[0006] (II) Technical Solution
[0007] To achieve the above objectives, the present invention provides the following technical solution:
[0008] This invention discloses a VR-based system for generating and interacting with psychological support scenarios for end-stage patients, comprising:
[0009] The scene acquisition module is used to provide users with several types of preset indicator scene data;
[0010] The status acquisition module is used to acquire facial and eye status feedback images when users experience data of a certain preset index scenario.
[0011] The feature extraction module is used to perform sentiment analysis on the status feedback image to extract sentiment features, and combine it with the user's corresponding language text to extract key semantic features;
[0012] The preference analysis module is used to build a preference analysis model through deep learning algorithms. The model takes sentiment features and semantic features as input and outputs a user preference coefficient for a preset indicator scenario.
[0013] The framework definition module is used to define the dimensions of the AR simulation framework based on the user's preference coefficients for several indicator scenarios;
[0014] The data filling module is used to call the scene repository and obtain the historical pre-stored scene data of the associated users, fill the historical pre-stored scene data into the corresponding dimensions of the AR simulation framework, and analyze the data completeness of each dimension. That is, the completeness is evaluated by calculating the proportion of the filled data items to the preset required data items of that dimension. If the data completeness of a certain dimension does not reach the preset threshold, supplementary tags are generated based on the specific missing content.
[0015] The scene construction module is used to construct a complete AR simulation framework after supplementing data based on the supplementary tag words, and to embed trigger commands for language and gesture interaction in the scene;
[0016] The interactive response module is used to simulate and demonstrate the constructed AR simulation framework, and dynamically demonstrate the corresponding scene content after the user triggers the language or gesture interaction command; the specific process is as follows:
[0017] By connecting to the constructed AR simulation framework through VR headset devices, the deployed virtual scenes are presented to users in an immersive first-person perspective.
[0018] The system continuously captures user voice through a microphone array and captures user gestures through VR controllers or visual sensors.
[0019] The captured interaction commands are matched with preset trigger commands in real time. If the match is successful, the AR simulation framework is immediately driven to execute the corresponding content changes, plot progression or feedback information presentation, so as to realize the dynamic interaction between the user and the psychological support scene.
[0020] Furthermore, the scene acquisition module stores several categories of preset indicator scene data, including natural landscape, family and friends interaction, life review, palliative care education, and culture. Based on the user's preliminary demographic information or the initial assessment opinions of medical staff, the scene acquisition module selects a suitable subset of preset indicator scenes from the preset scene library and pushes them to the user sequentially or randomly through VR devices.
[0021] Furthermore, the process of extracting emotional features in the feature extraction module is as follows:
[0022] Receive facial and eye state feedback image sequences from the state acquisition module, and preprocess the image sequences, including face detection and alignment, illumination normalization, and keyframe extraction.
[0023] The processed image is input into a pre-trained facial emotion recognition convolutional neural network, which outputs a basic emotion probability distribution vector based on a facial action coding system.
[0024] By using a pre-defined eye region analysis sub-network, eye features are extracted from the image, and the relative change curves of blink frequency, average fixation duration, pupil diameter, and saccade path divergence are obtained per unit time.
[0025] The basic emotional probability distribution vector is fused and dimensionality reduced with the quantized eye dynamic feature vector to generate an emotional feature vector that characterizes the user's instantaneous emotional tendency and attention state.
[0026] Furthermore, the extraction process of key semantic features in the feature extraction module is as follows:
[0027] Synchronously receive language text data generated by users during or after experiencing preset indicator scenarios; preprocess the text data, including word segmentation, stop word removal, and word form restoration;
[0028] A bidirectional encoder model based on an attention mechanism is used to perform deep semantic encoding on the preprocessed text to obtain the context-related vector representation of each word. Based on this representation, a keyword extraction algorithm is used to identify noun entities with high information content and adjectives with strong emotions as primary semantic features.
[0029] Implicit themes are extracted from the entire text to obtain several core themes and their distribution weights as macro-semantic features;
[0030] The primary semantic features and macro semantic features are structurally integrated to generate key semantic feature vectors that reflect users' focus, emotional evaluation, and potential psychological needs. A timestamp synchronization mechanism is established to ensure that the emotional feature vectors and key semantic feature vectors correspond to the same scene segment or feedback stage of the user experience in the time dimension, thereby providing spatiotemporally consistent multimodal data pairs for subsequent preference analysis.
[0031] Furthermore, in the preference analysis module, during the construction and training phase, the preference analysis model collects feature data and subsequent explicit feedback scores of several end-stage patients experiencing different preset scenarios as training samples; the emotional feature vector and semantic feature vector are fused and input into the initial neural network; with the goal of minimizing the error between the predicted preference coefficient output by the model and the actual feedback score, the network weights are iteratively optimized through the backpropagation algorithm until the model converges.
[0032] Furthermore, the formula for calculating the user's preference coefficient for preset indicator scenarios in the preference analysis module is as follows:
[0033] ;
[0034] In the formula, This represents the user's preference coefficient for the current preset indicator scenario, with a value range of [0,1]. The larger the value, the higher the degree of preference. Represents the activation function. The total number of emotional characteristics Trainable weight coefficients representing emotional features Representing the One emotional characteristic, Represents the total number of semantic features. Trainable weight coefficients representing semantic features Representing the A semantic feature, This represents the model bias term.
[0035] Furthermore, the framework definition module normalizes the user's preference coefficients for multiple indicator scenarios and maps them to a multi-dimensional preference space. Based on the clustering distribution of the preference coefficients in the space, it defines a dominant preference dimension including scenario theme tendency, interaction complexity preference, audiovisual stimulus intensity, and emotional tone. Based on the dominant preference dimension, it constructs a multi-dimensional AR simulation framework template with quantifiable parameters.
[0036] Furthermore, the historical pre-stored scene data in the data filling module includes images, audio, video, 3D models, and text records of users' past preferences, and the tag words include dimension identifiers, descriptions of missing content, and suggested data formats.
[0037] Furthermore, the operation of the scene construction module includes the following steps:
[0038] Step 71: Based on the parameters and data of each dimension of the fully populated AR simulation framework, call the 3D rendering engine to generate or assemble the corresponding virtual scene from the resource library in real time;
[0039] Step 72: Predefine a set of language keywords and gesture patterns that can trigger changes in scene content;
[0040] Step 73: Logically bind language keywords and gesture action patterns to specific objects, events, or plot branches in the scene, and set trigger response logic.
[0041] Furthermore, the scene acquisition module and the state acquisition module are interconnected via a wireless network; the state acquisition module and the feature extraction module are interconnected via a wireless network; the preference analysis module, the scene acquisition module, the feature extraction module, and the framework definition module are interconnected via a wireless network; the data filling module, the framework definition module, and the scene construction module are interconnected via a wireless network; and the scene construction module and the interaction response module are interconnected via a wireless network.
[0042] (III) Beneficial Effects
[0043] Compared with the known prior art, the technical solution provided by this invention has the following beneficial effects:
[0044] 1. By capturing patients' facial expressions, eye movements, and spoken text in real time, and combining this with deep learning models to analyze emotional and semantic features, the system dynamically calculates scene preference coefficients to ensure that recommended content highly matches the patient's current psychological state. It can adaptively adjust the dimensions of the AR simulation framework and supplement missing content based on historical data to form a coherent personalized experience, thereby improving the effectiveness of psychological support.
[0045] 2. By embedding natural language and gesture interaction commands into the generated AR scene, patients can directly control the scene evolution through voice commands or body movements, breaking through the limitations of traditional VR systems that rely on preset operations, and providing a low-threshold and high-freedom interaction method. This not only reduces the burden of use for terminal patients, but also enhances the psychological comfort effect by increasing participation.
[0046] 3. Through automated sentiment analysis and scene generation, the system reduces the continuous reliance on professional psychologists, alleviates the shortage of medical resources, and supports long-term tracking of changes in patient preferences by historical data storage and integrity verification functions. This provides medical staff with objective assessment data and assists in developing more scientific nursing plans. Attached Figure Description
[0047] Figure 1 This is a schematic diagram of the overall framework of the present invention;
[0048] Figure 2 This is a flowchart illustrating the operation of the scene construction module in this invention.
[0049] The numbers in the diagram represent: 1. Scene acquisition module; 2. State acquisition module; 3. Feature extraction module; 4. Preference analysis module; 5. Framework definition module; 6. Data filling module; 7. Scene construction module; 8. Interaction response module. Detailed Implementation
[0050] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.
[0051] The present invention will be further described below with reference to embodiments.
[0052] Example 1
[0053] This embodiment presents a VR-based system for generating and interacting with psychological support scenarios for end-stage patients, such as... Figures 1-2 As shown, it includes:
[0054] Scene acquisition module 1 is used to provide users with several types of preset indicator scene data. Scene acquisition module 1 stores several categories of preset indicator scene data, including natural landscape, family and friends interaction, life review, palliative care education and culture. Based on the user's preliminary demographic information or the initial assessment opinions of medical staff, scene acquisition module 1 selects a suitable subset of preset indicator scenes from the preset scene library and pushes them to the user sequentially or randomly through VR devices.
[0055] The status acquisition module 2 is used to acquire facial and eye status feedback images when users experience a certain preset indicator scenario data.
[0056] Feature extraction module 3 is used to perform sentiment analysis on the status feedback image to extract sentiment features, and combine it with the user's corresponding language text to extract key semantic features;
[0057] The process of extracting emotional features is as follows:
[0058] Receive facial and eye state feedback image sequences from state acquisition module 2, and preprocess the image sequences, including face detection and alignment, illumination normalization, and keyframe extraction;
[0059] The processed image is input into a pre-trained facial emotion recognition convolutional neural network, which outputs a basic emotion probability distribution vector based on a facial action coding system.
[0060] By using a pre-defined eye region analysis sub-network, eye features are extracted from the image, and the relative change curves of blink frequency, average fixation duration, pupil diameter, and saccade path divergence are obtained per unit time.
[0061] The basic emotional probability distribution vector is fused and dimensionality reduced with the quantized eye dynamic feature vector to generate an emotional feature vector that represents the user's instantaneous emotional tendency and attention state.
[0062] The process of extracting key semantic features is as follows:
[0063] Synchronously receive language text data generated by users during or after experiencing preset indicator scenarios; preprocess the text data, including word segmentation, stop word removal, and word form restoration;
[0064] A bidirectional encoder model based on an attention mechanism is used to perform deep semantic encoding on the preprocessed text to obtain the context-related vector representation of each word. Based on this representation, a keyword extraction algorithm is used to identify noun entities with high information content and adjectives with strong emotions as primary semantic features.
[0065] Implicit themes are extracted from the entire text to obtain several core themes and their distribution weights as macro-semantic features;
[0066] The primary semantic features and macro semantic features are structurally integrated to generate key semantic feature vectors that reflect users' focus, emotional evaluation and potential psychological needs. A timestamp synchronization mechanism is established to ensure that the emotional feature vectors and key semantic feature vectors correspond to the same scene segment or feedback stage of the user experience in the time dimension, thereby providing spatiotemporally consistent multimodal data pairs for subsequent preference analysis.
[0067] By deeply integrating and finely quantifying the deep semantics of non-verbal physiological responses and linguistic text, a spatiotemporally aligned user state representation is constructed. This overcomes the one-sidedness and superficiality of voice sentiment analysis or text sentiment analysis alone. It can capture the complex, implicit, and potentially contradictory psychological states and unspoken needs of terminally ill patients, providing a reliable and rich basis for decision-making in generating highly personalized psychological support scenarios.
[0068] Preference analysis module 4 is used to build a preference analysis model through deep learning algorithms. The model takes sentiment features and semantic features as input and outputs the user's preference coefficient for preset indicator scenarios.
[0069] The framework definition module 5 is used to define the dimensions of the AR simulation framework based on the user's preference coefficients for several indicator scenarios. The framework definition module 5 normalizes the user's preference coefficients for multiple indicator scenarios and maps them to a multi-dimensional preference space. Based on the clustering distribution of the preference coefficients in the space, it defines the dominant preference dimensions, including scene theme tendency, interaction complexity preference, audiovisual stimulus intensity, and emotional tone. Based on the dominant preference dimensions, a multi-dimensional AR simulation framework template with quantifiable parameters is constructed.
[0070] The data population module 6 is used to call the scene repository and obtain the historical pre-stored scene data of the associated users. It populates the corresponding dimensions of the AR simulation framework with the historical pre-stored scene data and analyzes the data completeness of each dimension. That is, it evaluates the completeness by calculating the proportion of the filled data items to the preset required data items of that dimension. If the data completeness of a certain dimension does not reach the preset threshold, it generates supplementary tags based on the specific missing content. The historical pre-stored scene data includes images, audio, video, 3D models and text records of users' past preferences. The tags include dimension identifiers, descriptions of missing content and suggested data formats.
[0071] Scene construction module 7 is used to build a complete AR simulation framework after supplementing data based on supplementary tags, and to embed trigger commands for language and gesture interaction into the scene; the operation of scene construction module 7 includes the following steps:
[0072] Step 71: Based on the parameters and data of each dimension of the fully populated AR simulation framework, call the 3D rendering engine to generate or assemble the corresponding virtual scene from the resource library in real time;
[0073] Step 72: Predefine a set of language keywords and gesture patterns that can trigger changes in scene content;
[0074] Step 73: Logically bind language keywords and gesture action patterns to specific objects, events, or plot branches in the scene, and set trigger response logic.
[0075] Interactive response module 8 is used to simulate and demonstrate the constructed AR simulation framework, and dynamically demonstrate the corresponding scene content after the user triggers a language or gesture interaction command; the specific process is as follows:
[0076] By connecting to the constructed AR simulation framework through VR headset devices, the deployed virtual scenes are presented to users in an immersive first-person perspective.
[0077] The system continuously captures user voice through a microphone array and captures user gestures through VR controllers or visual sensors.
[0078] The captured interaction commands are matched with preset trigger commands in real time. If the match is successful, the AR simulation framework is immediately driven to execute the corresponding content changes, plot progression or feedback information presentation, so as to realize the dynamic interaction between the user and the psychological support scene.
[0079] Scene acquisition module 1 and status acquisition module 2 are connected via a wireless network. Status acquisition module 2 and feature extraction module 3 are connected via a wireless network. Preference analysis module 4, scene acquisition module 1, feature extraction module 3 and framework definition module 5 are connected via a wireless network. Data filling module 6, framework definition module 5 and scene construction module 7 are connected via a wireless network. Scene construction module 7 and interaction response module 8 are connected via a wireless network.
[0080] Compared with existing technologies, this embodiment achieves more accurate emotional state recognition and preference analysis by integrating multi-dimensional data of facial expressions, eye dynamics, and spoken text, combined with a deep learning model. It overcomes the limitations of traditional single feedback methods, such as questionnaires or simple observations. Based on the user's real-time emotional characteristics and historical data, it automatically constructs an AR simulation framework that meets the user's psychological needs and dynamically adjusts the scene content. Compared with static preset scenes or general templates, it improves the ability to personalize and adapt. By automatically verifying the completeness of dimensional data and generating supplementary tags, it ensures the comprehensiveness of scene construction, solves the problem of insufficient scene adaptation caused by data loss in traditional methods, and reduces the operational burden on medical staff.
[0081] Example 2
[0082] At other levels, this embodiment also provides another optimization mechanism based on embodiment 1, specifically a preference analysis model construction and training phase, including: collecting feature data and subsequent explicit feedback scores of several end-stage patients experiencing different preset scenarios as training samples; fusing emotional feature vectors and semantic feature vectors and inputting them into an initial neural network; with the goal of minimizing the error between the predicted preference coefficient output by the model and the actual feedback score, iteratively optimizing the network weights through the backpropagation algorithm until the model converges.
[0083] The formula for calculating the user's preference coefficient for preset indicator scenarios is:
[0084] ;
[0085] In the formula, This represents the user's preference coefficient for the current preset indicator scenario, with a value range of [0,1]. The larger the value, the higher the degree of preference. Represents the activation function. The total number of emotional characteristics Trainable weight coefficients representing emotional features Representing the One emotional characteristic, Represents the total number of semantic features. Trainable weight coefficients representing semantic features Representing the A semantic feature, This represents the model bias term, used to adjust the output baseline;
[0086] By combining emotional and semantic features, a more comprehensive model of user psychological state can be achieved, improving the accuracy of preference prediction. Feature contributions can be dynamically adjusted according to individual user differences and scenario characteristics, enhancing personalized adaptation capabilities.
[0087] In summary, this invention combines facial expressions, eye states, and language text analysis to dynamically extract users' emotional and semantic features. It uses a preference analysis model to calculate preference coefficients, ensuring that the scene content accurately matches the patient's psychological needs and improving the effectiveness of psychological intervention. Based on historical preference data and real-time feedback, it automatically adjusts the dimensions and content of the AR simulation framework and supplements missing data through an integrity verification mechanism to ensure the continuity and richness of the scene and avoid negative emotions caused to patients by repetitive or inappropriate content.
[0088] By using voice and gesture interaction, patients are given autonomy in the virtual environment, enhancing their sense of participation and control, alleviating the helplessness and anxiety commonly experienced in the terminal stage. Through a supplementary tagging mechanism, the system intelligently prompts medical staff to improve patients' personalized data, reducing the burden of manual assessment, and providing traceable preference analysis for subsequent treatment.
[0089] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A terminal stage patient psychological support scene generation and interaction system based on VR technology, characterized in that, include: The scene acquisition module is used to provide users with several types of preset indicator scene data; The status acquisition module is used to acquire facial and eye status feedback images when users experience data of a certain preset index scenario. The feature extraction module is used to perform sentiment analysis on the status feedback image to extract sentiment features, and combine it with the user's corresponding language text to extract key semantic features; The preference analysis module is used to build a preference analysis model through deep learning algorithms. The model takes sentiment features and semantic features as input and outputs a user preference coefficient for a preset indicator scenario. The framework definition module is used to define the dimensions of the AR simulation framework based on the user's preference coefficients for several indicator scenarios; The data filling module is used to call the scene repository and obtain the historical pre-stored scene data of the associated users, fill the historical pre-stored scene data into the corresponding dimensions of the AR simulation framework, analyze the data completeness of each dimension, and if the data completeness of a certain dimension does not reach the preset threshold, supplementary tags are generated based on the specific missing content. The scene construction module is used to construct a complete AR simulation framework after supplementing data based on the supplementary tag words, and to embed trigger commands for language and gesture interaction in the scene; The interactive response module is used to simulate and demonstrate the constructed AR simulation framework, and dynamically demonstrate the corresponding scene content after the user triggers the language or gesture interaction command. 2.The VR technology-based end-stage patient psychological support scene generation and interaction system according to claim 1, characterized in that, The scene acquisition module stores several categories of preset indicator scene data, including natural landscape, family and friends interaction, life review, palliative care education, and culture. Based on the user's preliminary demographic information or the initial assessment opinions of medical staff, the scene acquisition module selects a suitable subset of preset indicator scenes from the preset scene library and pushes them to the user sequentially or randomly through VR devices. 3.The VR technology based end-stage patient psychological support scene generation and interaction system according to claim 1, characterized in that, The process of extracting emotional features in the feature extraction module is as follows: Receive facial and eye state feedback image sequences from the state acquisition module and preprocess the image sequences; The processed image is input into a pre-trained facial emotion recognition convolutional neural network, which outputs a basic emotion probability distribution vector based on a facial action coding system. By using a pre-defined eye region analysis sub-network, eye features are extracted from the image, and the relative change curves of blink frequency, average fixation duration, pupil diameter, and saccade path divergence are obtained per unit time. The basic emotional probability distribution vector is fused and dimensionality reduced with the quantized eye dynamic feature vector to generate an emotional feature vector that represents the user's instantaneous emotional tendency and attention state. 4.The VR technology based end-stage patient psychological support scene generation and interaction system according to claim 1, wherein, The extraction process of key semantic features in the feature extraction module is as follows: Synchronously receive language and text data generated by users during or after experiencing preset indicator scenarios; preprocess the text data; A bidirectional encoder model based on an attention mechanism is used to perform deep semantic encoding on the preprocessed text to obtain the context-related vector representation of each word; based on this representation, a keyword extraction algorithm is used to identify noun entities and adjectives as primary semantic features. Implicit themes are extracted from the entire text to obtain several core themes and their distribution weights as macro-semantic features; By structurally integrating primary semantic features with macro-semantic features, key semantic feature vectors that reflect users' focus, emotional evaluation, and potential psychological needs are generated. 5.The VR technology based end-stage patient psychological support scene generation and interaction system according to claim 1, characterized in that, In the preference analysis module, during the construction and training phase, the preference analysis model collects feature data and subsequent explicit feedback scores from several end-stage patients experiencing different preset scenarios as training samples; the emotional feature vector and semantic feature vector are then fused and input into the initial neural network. With the goal of minimizing the error between the predicted preference coefficients output by the model and the actual feedback scores, the network weights are iteratively optimized through the backpropagation algorithm until the model converges. 6.The VR technology based end-stage patient psychological support scene generation and interaction system according to claim 1, wherein, The formula for calculating the user's preference coefficient for preset indicator scenarios in the preference analysis module is as follows: ； In the formula, represent the preference coefficient of the user to the current preset index scene, represent the activation function, represent the total number of emotional features, represent the trainable weight coefficient of the emotional feature, represent the first emotional feature, represent the total number of semantic features, represent the trainable weight coefficient of the semantic feature, represent the first semantic feature, represent the model bias term. 7.The VR technology based end-stage patient psychological support scene generation and interaction system according to claim 1, wherein, The framework definition module normalizes the user's preference coefficients for multiple indicator scenarios and maps them to a multi-dimensional preference space. Based on the clustering distribution of the preference coefficients in the space, it defines a dominant preference dimension including scenario theme tendency, interaction complexity preference, audiovisual stimulus intensity, and emotional tone. Based on the dominant preference dimension, it constructs a multi-dimensional AR simulation framework template with quantifiable parameters. 8.The VR technology based end-stage patient psychological support scene generation and interaction system according to claim 1, wherein, The historical pre-stored scene data in the data filling module includes images, audio, video, 3D models and text records of users' past preferences, and the tag words include dimension identifiers, descriptions of missing content and suggested data formats. 9.The VR technology based end-stage patient psychological support scene generation and interaction system according to claim 1, wherein, The operation of the scene construction module includes the following steps: Step 71: Based on the parameters and data of each dimension of the fully populated AR simulation framework, call the 3D rendering engine to generate or assemble the corresponding virtual scene from the resource library in real time; Step 72: Predefine a set of language keywords and gesture patterns that can trigger changes in scene content; Step 73: Logically bind language keywords and gesture action patterns to specific objects, events, or plot branches in the scene, and set trigger response logic.
10. The VR-based psychological support scene generation and interaction system for terminally ill patients according to claim 1, characterized in that, The scene acquisition module and the state acquisition module are interconnected via a wireless network. The state acquisition module and the feature extraction module are interconnected via a wireless network. The preference analysis module, the scene acquisition module, the feature extraction module, and the framework definition module are interconnected via a wireless network. The data filling module, the framework definition module, and the scene construction module are interconnected via a wireless network. The scene construction module and the interaction response module are interconnected via a wireless network.