Dynamic self-adaptive cognitive and mental health intervention system and method for middle-aged and elderly people

Abstract

The invention belongs to the technical field of cognitive disorder rehabilitation, and particularly relates to a dynamic self-adaptive middle-aged and elderly cognitive and mental health intervention system which comprises a VR interaction module, a physiological data acquisition and processing module, a dynamic self-adaptive module and a cloud management platform. Interaction scenes of the VR interaction module comprise an occupational scene and a nostalgic scene; the physiological data acquisition and processing module acquires and processes multi-modal physiological data of a user in real time; the dynamic self-adaptive module is used for calculating a cognitive load index and an emotion pressure index based on physiological data, dynamically adjusting occupational task complexity through a fuzzy logic controller, triggering restalgic scene switching, and embedding a mild cognitive task in a restalgic scene to maintain training continuity; and the cloud management platform stores user historical data, and optimizes a scene recommendation strategy through a machine learning model. A personalized intervention strategy can be customized according to the actual situation of the user, and the user experience is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of cognitive impairment rehabilitation, and particularly relates to a dynamic adaptive middle-aged and elderly cognitive and mental health intervention system and method. Background Art

[0002] Currently, the intervention technologies for middle-aged and elderly cognitive and mental health face multiple core problems, which restrict their intervention effects and popularity. The main problems are as follows:

[0003] The intervention means are single and static. Most existing intervention systems independently apply occupational therapy or reminiscence therapy, lacking a coordination mechanism. For example, occupational training software only focuses on task completion, such as kitchen simulation, ignoring the impact of emotional stress on cognitive load, resulting in a 30% increase in the user's anxiety value and a 40% abandonment rate during training. In addition, traditional VR rehabilitation systems use a fixed difficulty task library and cannot dynamically adjust according to the user's real-time state. Research shows that when the task difficulty does not match the user's ability, the cognitive training efficiency drops by 50%, and the improvement of the Mini-Mental State Examination (MMSE) score is less than 2 points per month.

[0004] There is a lag in physiological signal monitoring and feedback. Existing physiological signal acquisition devices only collect single-modal signals, such as independent eye trackers or smart bracelets, lacking the fusion analysis of multi-source data, such as eye movement, galvanic skin response, and electromyogram. Therefore, the cognitive-emotional interaction state cannot be accurately quantified. A certain clinical study shows that the error in predicting stress levels relying solely on heart rate data exceeds ±35%. At the same time, traditional systems use offline data processing, such as synchronizing to the server at night, resulting in a delay of more than 10 minutes in adjusting intervention strategies and being unable to respond in a timely manner to the user's immediate needs, such as switching scenes during sudden anxiety.

[0005] The application of VR technology has limitations. Most systems use prefabricated general scenarios, such as standard kitchen models, lacking personalized adaptation. Surveys show that 80% of middle-aged and elderly users feel alienated from non-customized scenarios, and their training participation is less than 20 minutes per day. Existing solutions mostly rely on visual-auditory channels, ignoring multi-sensory collaborative stimuli such as tactile vibration, resulting in poor immersion, a user rating lower than 3.5 points (on a 5-point scale), and a 40% reduction in the activation efficiency of neuroplasticity.

[0006] Weak intelligent regulation ability. Traditional adaptive systems use threshold alarms, such as pausing training when pressure exceeds the limit, lacking a dynamic balance mechanism. For example, when users face both high cognitive load (Cognitive Load Index CLI > 70) and moderate stress (Emotional Stress Index ESI > 50) simultaneously, the system cannot automatically generate optimal intervention strategies, such as reducing task complexity while activating mild nostalgia elements. In addition, local devices lack knowledge sharing with the cloud and cannot optimize the model using group data. A community pilot project shows that the recommendation accuracy of the system not connected to the cloud decays to 65% after 3 months.

[0007] To address these technical problems, there is an urgent need for an intervention system that can dynamically and adaptively adjust training content, which will have great application prospects. Summary of the Invention

[0008] To solve the above technical problems, the present invention provides a dynamic adaptive middle-aged and elderly cognitive and mental health intervention system, which has the advantages of improving the intervention effect, enhancing the user experience, achieving personalized adaptation, and intelligent regulation.

[0009] To achieve the above invention purpose, the technical solution adopted by the present invention is: a dynamic adaptive middle-aged and elderly cognitive and mental health intervention system, including a VR interaction module, a physiological data acquisition and processing module, a dynamic adaptive module, and a cloud management platform; the VR interaction module is used for information interaction between the user and the VR scene, and the interaction scenes include a professional scene and a nostalgic scene; the physiological data acquisition and processing module collects and processes the user's multimodal physiological data in real time; the dynamic adaptive module calculates the cognitive load index and the emotional stress index based on the physiological data, dynamically adjusts the complexity of the professional task and triggers the switching of the nostalgic scene through a fuzzy logic controller, and embeds mild cognitive tasks in the nostalgic scene to maintain the continuity of training; the cloud management platform stores the user's historical data and optimizes the scene recommendation strategy through a machine learning model.

[0010] Preferably, the VR interaction module includes a professional scene interaction unit, a nostalgic scene interaction unit, and an immersive dialogue unit; the professional scene interaction unit constructs professional tasks based on a task decomposition algorithm and uses a real-time error correction mechanism for task training;

[0011] The nostalgic scene interaction unit generates a nostalgic scene through a generative adversarial network according to the user's historical daily data; according to the magnitude of the emotional stress index, dynamically adjusts the scene elements through an emotional intensity regulator, and the scene elements include the caring color tone and the background music rhythm;

[0012] The nostalgia scene interaction unit also embeds trigger objects in the nostalgia scene, associates each trigger object with a specific memory segment. When the user interacts with the trigger object, the associated memory segment is played, and the user's interaction sequence data and emotional state data during this process are obtained to train a long short-term memory network to predict the optimal emotional support timing.

[0013] The immersive dialogue unit uses natural language processing technology to guide the user to orally describe autobiographical and memoir content, and real-time records the user's voice data and converts it into structured text, so as to embed the user's autobiographical keywords and memory fragments in the nostalgia scene unit to enhance the scene relevance.

[0014] Preferably, the physiological data acquisition and processing module includes a physiological data acquisition unit, an eye movement analysis unit, and a skin conductance signal processing unit; the physiological data acquisition unit real-time monitors the user's fixation point trajectory, pupil diameter change, skin conductance, upper limb movement accuracy and coordination; the eye movement analysis unit extracts a fixation heat map through a convolutional neural network to identify the attention dispersion area, and when the heat map deviates from the task key area by more than 30%, an attention correction prompt is triggered; the skin conductance signal processing unit uses wavelet transform to remove motion artifacts and extracts the sympathetic nerve activation signal in the frequency band of 0.05 - 0.15Hz.

[0015] Preferably, the dynamic adaptive module includes a calculation unit, a dual-threshold mechanism adjustment unit, and a cross-scene continuity adjustment unit; the calculation unit calculates the cognitive load index and emotional stress index based on the user's physiological data;

[0016] The dual-threshold mechanism adjustment unit respectively determines whether the cognitive load index and the emotional stress index exceed the set thresholds: when the cognitive load index exceeds the cognitive load set threshold, the complexity of the occupational scene task is reduced; when the emotional stress index exceeds the emotional stress set threshold, the occupational scene is switched to the nostalgia scene;

[0017] The cross-scene continuity adjustment unit embeds mild cognitive tasks in the nostalgia scene to provide emotional support while continuing cognitive training.

[0018] Preferably, the cloud management platform includes a database, an intervention strategy optimization unit, and a query unit. The database is used to store the user's historical data, autobiographical text, and memoir audio and generate a personalized intervention report. The intervention strategy optimization unit takes the user's MMSE score, task completion efficiency, physiological signal time series characteristics, and autobiographical keyword emotional analysis results as inputs and the intervention strategy as the output to train a machine learning model to optimize the scene recommendation strategy; the query unit supports retrieving the user's memoir content by timeline and / or theme classification.

[0019] Correspondingly, a dynamic adaptive cognitive and mental health intervention method for the middle-aged and elderly includes the following steps:

[0020] Step 101: Decompose complex occupational tasks into multiple atomic action steps based on a task decomposition algorithm, associate each atomic action step with a cognitive training goal, and use a real-time error correction mechanism for occupational task training;

[0021] Step 102: Collect and process the user's physiological data in real time. The physiological data includes gaze point trajectory and pupil diameter change, galvanic skin signal, and electromyogram signal;

[0022] Step 103: Calculate the user's cognitive load index and emotional stress index based on the user's physiological data, and determine whether the cognitive load index and emotional stress index exceed the set thresholds. When the cognitive load index exceeds the cognitive load set threshold, reduce the complexity of the occupational scenario tasks and continue training; when the emotional stress index exceeds the emotional stress set threshold, switch the occupational scenario to a nostalgic scenario;

[0023] Step 104: Embed mild cognitive tasks in the nostalgic scenario, dynamically adjust the scenario elements through an emotional intensity regulator according to the user's physiological state; embed trigger objects in the nostalgic scenario, associate the trigger objects with specific memory fragments, and when the user interacts with the trigger objects, play the associated memory fragments, and train a long short-term memory network based on the user's interaction sequence data and emotional state data of the trigger objects to predict the optimal emotional support timing;

[0024] Step 105: At the end of the training, store the user's historical intervention data, autobiographical text, and memoir audio, and generate a personalized intervention report; train a machine learning model based on the user's MMSE score, task completion efficiency, physiological signal time series characteristics, autobiographical keyword sentiment analysis results, and intervention strategies to optimize the scenario recommendation strategy.

[0025] The present invention has the following beneficial effects:

[0026] 1. The intervention system of the present application includes a VR interaction module, a physiological data acquisition and processing module, a dynamic adaptive module, and a cloud management platform. The VR interaction module integrates occupational scenarios and nostalgic scenarios, combines multi-sensory feedback and real-time physiological data monitoring, and dynamically adjusts the complexity of occupational tasks, triggers the switching of nostalgic scenarios, and embeds mild cognitive tasks in the nostalgic scenario to maintain the continuity of training based on the user's physiological data, realizing dynamic adaptive cognitive and mental health intervention, solving the problems of single, static traditional intervention means and lagging physiological signal monitoring, being able to customize personalized intervention strategies according to the actual situation of users, providing the user experience, and having the advantages of improving the intervention effect, enhancing the user experience, realizing personalized adaptation, and intelligent regulation.

[0027] 2. Compared with traditional intervention systems, the vocational scenario interaction implemented based on the task decomposition algorithm in this application decomposes complex life skills into atomic action steps, associates each atomic action step with a cognitive training goal, and can dynamically adjust the weight coefficients of the atomic action steps according to the actual capabilities of the user, enabling each atomic step to be optimized for the user's current cognitive level. In this way, the user can gradually improve their overall cognitive and operational capabilities while gradually completing each small step. By setting up a progressive real-time error correction mechanism, it can help users promptly correct operation errors and avoid training interruptions and frustration caused by incorrect operations. It can significantly improve the cognitive and operational capabilities of middle-aged and elderly users, and enhance the training effect and user participation. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 is a structural diagram of the middle-aged and elderly cognitive and mental health intervention system of the present invention;

[0029] Figure 2 is a flowchart of the middle-aged and elderly cognitive and mental health intervention method of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0030] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. If not specifically specified, the technical means used in the embodiments are conventional means well known to those skilled in the art.

[0031] Such as Figure 1As shown, a dynamic adaptive cognitive and mental health intervention system for the middle-aged and elderly includes a VR interaction module, a physiological data acquisition and processing module, a dynamic adaptive module, and a cloud management platform. The VR interaction module is used for information interaction between the user and the virtual reality scene, integrated with a professional scene library and a nostalgic scene library, and supports multi-sensory feedback. The physiological data acquisition and processing module real-time collects and processes the user's multi-modal physiological data, specifically a device combination for synchronously obtaining eye movement trajectories, skin conductivity, and electromyogram signals, which can be implemented using high-sampling rate sensors and wireless transmission modules, and is used for real-time monitoring of cognitive load and emotional fluctuations. The dynamic adaptive module calculates the cognitive load index CLI and the emotional stress index ESI in real-time based on physiological data, dynamically adjusts the complexity of professional tasks and triggers nostalgic scene switching through a fuzzy logic controller, and embeds mild cognitive tasks in the nostalgic scene to maintain training continuity. Specifically, a real-time signal processing chip and a decision tree algorithm can be used to dynamically balance the task difficulty and emotional support intensity according to the user's state. The cloud management platform is used to store the user's historical data and generate personalized intervention reports, and optimizes the scene recommendation strategy through a machine learning model. Specifically, a distributed storage architecture and a neural network training framework can be used to long-term track the user's progress and optimize the intervention strategy.

[0032] Through this technical solution, the problem of user alienation caused by prefabricated scenes is solved, and the immersive experience is enhanced through a personalized virtual environment; the defect that the static scene cannot adapt to emotional changes is overcome, and the dynamic matching of scene elements and physiological states is realized; the triggering logic of memory activation is optimized, and the interference of mechanical prompts on the user experience is avoided, thereby improving the intervention effect of nostalgia therapy.

[0033] Furthermore, the VR interaction module includes a professional scene interaction unit, a nostalgic scene interaction unit, and an immersive dialogue unit. The specific implementation methods are as follows:

[0034] The professional scene interaction unit constructs professional tasks based on the task decomposition algorithm and uses a real-time error correction mechanism for task training.

[0035] Specifically, the task decomposition algorithm decomposes complex professional tasks into multiple atomic action steps, and associates each atomic action step with a cognitive training goal. The task decomposition algorithm can dynamically adjust the weight coefficients of each atomic action step according to the user's ability, so that each atomic step can be optimized for the user's current cognitive level. In this way, the user can gradually improve their overall cognitive and operation ability while gradually completing each atomic step.

[0036] In some embodiments, the implementation of the task decomposition algorithm may include rule-based step decomposition and step optimization based on a specific machine learning model. For example, the rule-based method can pre-define the standard operation steps for each life skill and dynamically adjust the difficulty and order of the steps according to indicators such as the user's MMSE score. The machine learning-based method can collect a large amount of user operation data and train a specific machine learning model to predict the optimal step decomposition and adjustment strategy.

[0037] Specifically, the real-time error correction mechanism triggers progressive prompts when the user makes an operation error through gesture recognition and object interaction detection during intervention. The progressive prompts can be visual highlighting, voice guidance, and action demonstration in sequence. When the user makes an error during the operation, the system first prompts the user to pay attention to the error location through visual highlighting; if the user fails to correct the error, the system will further provide operation guidance through voice guidance; if the user still fails to complete the correct operation, the system will finally demonstrate the correct operation steps intuitively through action demonstration.

[0038] In some embodiments, the implementation of the real-time error correction mechanism may include computer vision-based gesture recognition and sensor-based object interaction detection. For example, gesture recognition can use a depth camera to capture the user's hand movements and perform action recognition through a convolutional neural network. Object interaction detection can install sensors on the object to monitor in real time whether the user's operation meets the expectations.

[0039] If the occupational task is decomposed into n atomic action steps, the calculation formula for the occupational task complexity F is as follows:

[0040]

[0041] where w i is the weight coefficient for dynamically adjusting the i-th atomic action step according to the user's ability, and w1 + w2… + w i … + w n = 1. If the user is more capable in a certain aspect, the corresponding step weight may be reduced; if the user is less capable in a certain aspect, the corresponding step weight may be increased, so as to flexibly adjust the occupational task complexity. The weight coefficient can be dynamically adjusted according to the user's unique cognitive ability. For example, for a user with good memory but poor fine hand movements, the steps involving hand operations are assigned a higher weight coefficient to highlight the training focus. In the initial stage of training, reduce the weight coefficient of complex steps to let the user establish information; after the ability is improved, increase the weight coefficient of complex steps, reduce the weight coefficient of simple steps, and can also increase the task concurrency factor to stimulate cognitive function.

[0042] f iIt is the complexity of the i-th atomization action step, which is related to the operation fineness and knowledge and skill requirements of this step. Assign values to different complexities, such as 1 for simple, 2 for relatively simple, 3 for medium, 4 for relatively complex, and 5 for complex.

[0043] x i It is the multi-task concurrency factor of the i-th atomization action step, indicating the degree to which this step is carried out simultaneously with other steps and interacts with them during execution. If it is a single task, the multi-task concurrency factor is 1; if two tasks are carried out simultaneously, the multi-task concurrency factor is 2.

[0044] This embodiment explains the occupational scenario interaction process taking cooking skills as an example. First, the task of cooking is decomposed into four processes: ingredient preparation, tool preparation, ingredient pre-treatment, and cooking. Then, ingredient preparation is further divided into identifying the required ingredients, confirming the freshness and quantity of the ingredients. Associating the identification of the required ingredients with visual recognition and concentration ability, and the ability to accurately screen information; associating the confirmation of the freshness and quantity of the ingredients with the ability of detailed observation; tool preparation is divided into selecting suitable cooking utensils and checking the usability of the cooking utensils. Associating the selection of suitable cooking utensils with the ability to accurately screen information and the ability of hand movement coordination, and associating the checking of the usability of the cooking utensils with the safety awareness ability; ingredient pre-treatment is divided into washing the ingredients and cutting the ingredients. Associating the washing of the ingredients with the ability of fine hand movement control, and associating the cutting of the ingredients with the hand-eye coordination ability; cooking is divided into heating the pan with cold oil, adding the vegetables separately, and stir-frying them together. Associating heating the pan with cold oil with the ability of external perception, associating adding the vegetables separately with the ability to observe external dynamic changes, and associating stir-frying them together with the ability of step sequence memory and the aesthetic cognition ability of cooking works.

[0045] When the user makes an operation error during the execution of the above atomization action steps, the intervention system first highlights the relevant correct operation area or operation tool visually to prompt the user about the error; if the user does not respond in time or still does not correct it, the intervention system will start voice guidance to elaborate on the correct operation method in detail; if the user still makes mistakes after the voice guidance, the intervention system will conduct action demonstrations, showing the correct action process through virtual avatars or videos to help the user correct the mistakes in time and ensure the accuracy and effectiveness of occupational task training. This progressive prompt can effectively help the user correct mistakes in time during the training process and improve the training effect.

[0046] The nostalgic scene interaction unit generates a high-fidelity virtual nostalgic scene according to the user's historical photos, audio and video records, and occupational files through a generative adversarial network (GAN); the emotional intensity regulator dynamically adjusts the scene elements according to the emotional stress index ESI. The scene elements include lighting tones, background music rhythms, etc. For example, if the emotional stress index ESI is relatively high, the lighting is adjusted to warm tones and a soothing background music is selected; otherwise, the lighting is adjusted to normal or cold tones and a lively background music is selected.

[0047] In some embodiments, based on the architecture of the adversarial generative network, a conditional input branch is added, and the user's emotional tags, recall keywords, and psychological data are input into the generator and discriminator as conditional information to strengthen the response of the scenario to the user's emotions. For example, when the user is in a sad mood, by inputting the recall keywords, the generator can adjust the nostalgic scene to warm yellow, highlight the characteristics of the keywords, play soft background music, and the discriminator determines whether the generated nostalgic scene meets the requirements based on these conditions.

[0048] Meanwhile, the nostalgic scene interaction unit also embeds trigger objects at reasonable positions in the nostalgic scene. The trigger objects include an old-fashioned typewriter, a workbench, etc. Each trigger object is associated with a specific memory segment. When the user interacts with the trigger object, a video or image of the associated memory segment is played, and the interaction sequence data between the user and the trigger object is used as the input, and the corresponding emotional state (such as the emotional stress index) during the interaction is used as the output to train a long short-term memory network (LSTM) to learn the relationship between different interaction sequences and changes in the user's emotional state and predict the optimal emotional support timing.

[0049] The immersive dialogue unit uses natural language processing technology to guide the user to orally describe their autobiography and memoir content, real-time records the user's voice data and converts it into structured text, so as to embed the user's autobiography keywords and recall fragments in the nostalgic scene unit to enhance the scene relevance. The immersive dialogue unit refers to a dialogue system based on natural language processing technology, and specifically, language recognition and text generation algorithms can be used to guide the user to conduct an autobiographical narrative and extract key memory fragments.

[0050] In some embodiments, the occupational scenarios include kitchen operation simulation, shopping simulation, handicraft making simulation, etc. The nostalgic scenarios include historical work environments, historical family environments, historical activity parks, historical activity supermarkets, etc. The VR interaction module supports multi-sensory feedback, including 3D visual rendering, spatial audio, and tactile vibration. By synchronously stimulating the auditory and tactile channels through the spatial audio and vibration devices, it is used to enhance the scene realism and user immersion. The hardware devices of the VR interaction module include a VR headset device, a spatial audio device, and a tactile vibration device. In some embodiments, the VR headset device uses a binocular 4K OLED display screen with a refresh rate ≥ 90Hz and is equipped with an Inside-out positioning system.

[0051] Furthermore, the physiological data acquisition and processing module includes a physiological data acquisition unit, an eye movement analysis unit, a galvanic skin signal processing unit, and an electromyogram signal processing unit. The specific implementation is as follows:

[0052] The physiological data acquisition unit includes a high-precision eye tracker, a galvanic skin response sensor (GSR), a surface electromyography sensor (sEMG), or integrates the GSR and sEMG sensors to form a wireless bioelectric sensor group. The high-precision eye tracker is used to monitor the user's fixation point trajectory and pupil diameter changes in real time. The GSR sensor is used to collect the user's skin conductivity to quantify the stress level. The sEMG sensor is used to detect the user's upper limb movement accuracy and coordination.

[0053] In some embodiments, the sampling rate of the eye tracker is ≥250Hz and the accuracy is ±0.5°. The wireless bioelectric sensor group is used to collect skin conductivity and electromyography signals. The wireless bioelectric sensor group supports Bluetooth 5.0 low-power transmission, ensuring the stability and low latency of data transmission and enabling efficient acquisition of skin signals and electromyography signals.

[0054] The eye movement analysis unit receives the user's fixation point trajectory and pupil diameter change data, processes these data using a convolutional neural network (CNN) to generate a fixation heat map, and identifies the user's attention distraction area. When the recognition result shows that the heat map deviates from the task critical area by more than 30%, the intervention system automatically triggers an attention correction prompt to help the user refocus attention.

[0055] The GSR signal processing unit receives the user's GSR signal, uses the wavelet transform algorithm to remove motion artifacts from these data, and extracts the sympathetic nerve activation signal in the frequency band of 0.05 - 0.15Hz. Through the changes in these signals, the intervention system can calculate the user's stress level in real time and dynamically adjust the intervention strategy according to the changes in the stress level.

[0056] The electromyography signal processing unit receives the user's electromyography signal data, first uses a band-stop filter to remove power frequency interference, and then performs normalization processing.

[0057] Furthermore, the dynamic adaptive module includes a calculation unit, a dual-threshold mechanism adjustment unit, and a cross-scene continuity adjustment unit. The specific implementation is as follows:

[0058] The calculation unit calculates the user's cognitive load index CLI and emotional stress index ESI during the training process based on multimodal physiological data.

[0059] (1) Calculate the cognitive load index

[0060] Data preprocessing: Remove outliers from the fixation point trajectory data collected by the high-precision eye tracker, collect pupil diameter change data at a certain frequency and remove outliers. Normalize the GSR signals collected by the GSR sensor to remove environmental interference. Remove power frequency interference (such as 50Hz) from the electromyography signals collected by the surface electromyography sensor, and then perform normalization processing.

[0061] Physiological index feature extraction: Eye movement index extraction includes the fixation point trajectory and the change in pupil diameter. The extraction of the fixation point trajectory specifically calculates the number of fixation point transfers per unit time and the proportion of the residence time of the fixation point in the key area. The extraction of the change in pupil diameter specifically calculates the change rate of the pupil diameter during the training process relative to the initial pupil diameter. The galvanic skin response (GSR) index extraction specifically calculates the change in skin conductivity during the training process relative to the initial average skin conductivity. The electromyogram (EMG) index extraction specifically calculates the root mean square value of the EMG signals of the muscles related to upper limb movements and calculates the correlation between the EMG signals of different muscles.

[0062] Determine the weights of each index: The weight coefficients can be assigned to the eye movement index, GSR index, and EMG index respectively according to the expert scoring method.

[0063] The calculation formula of the cognitive load index CLI is as follows:

[0064] CLI = k1·L 眼动 +k2·L 皮电 +k3·L 肌电 ;

[0065] In the formula, k1, k2, and k3 represent the weight coefficients of the eye movement index, GSR index, and EMG index respectively, and k1 + k2 + k3 = 1; L 眼动 is the quantization value of the comprehensive eye movement characteristics. For example, the three characteristics of the number of fixation point transfers, the proportion of residence time in the key area, and the change rate of pupil diameter are obtained through weighted summation; L 皮电 is the quantization value of the change in skin conductivity; L 肌电 is the quantization value of the root mean square value and correlation of the EMG signals. The higher the calculated cognitive load index CLI value, the higher the current cognitive load of the user.

[0066] (2) Calculate the emotional stress index

[0067] The calculation formula of the emotional stress index ESI is as follows:

[0068]

[0069] In the formula, N is the number of peaks of the GSR signal; T is the time window; A is the average amplitude of the GSR signal.

[0070] The dual-threshold mechanism adjustment unit determines whether the cognitive load index and the emotional stress index exceed the set thresholds: When the cognitive load index exceeds the cognitive load set threshold, the complexity of the occupational scenario task is reduced; when the emotional stress index exceeds the emotional stress set threshold, the occupational scenario is switched to the nostalgic scenario. The complexity of the occupational scenario task includes the number of atomic action steps and the time limit for each step.

[0071] In some embodiments, when the cognitive load index exceeds 1, the intervention system will automatically reduce the number of steps of the occupational task and extend the time limit of the steps with a large difficulty coefficient. For example, it will automatically reduce the number of kitchen operation steps or extend the time limit of each step or extend the time limit of a certain step to reduce the user's cognitive load. This can effectively reduce the user's cognitive load and avoid the frustration and abandonment behaviors caused by excessive task difficulty. On the other hand, when the emotional stress index exceeds 2, the intervention system will interrupt the current occupational scenario and switch to a nostalgic scenario to help the user relieve stress through emotional support and memory activation. This dual-threshold mechanism adjustment ensures that the intervention system can dynamically adjust according to the user's real-time state and improve the intervention effect.

[0072] The cross-scenario continuity adjustment unit embeds mild cognitive tasks in the nostalgic scenario. The purpose is to continue cognitive training while providing emotional support to ensure that the emotional relaxation of the user in the nostalgic scenario does not interrupt the continuity of cognitive training. In some embodiments, some tasks related to the user's past experiences are set in the nostalgic scenario, such as sorting old photos, identifying old objects, etc. Through these mild cognitive tasks, the user can still perform cognitive training while relaxing.

[0073] In some embodiments, the hardware device of the dynamic adaptive module configures an edge computing gateway. The edge computing gateway integrates NVIDIA Jetson AGX Xavier and has powerful computing capabilities. It can run complex adaptive algorithms in real time to ensure that the system can quickly respond to the changes in the user's state. At the same time, the edge computing gateway is built-in with a security encryption module to protect the privacy of physiological data. Specifically, it can calculate the cognitive load index and the emotional stress index in real time and adjust the VR scene parameters in real time according to the calculation results. The VR scene parameters include the complexity of the occupational task and the triggering conditions for switching to the nostalgic scenario.

[0074] Furthermore, the cloud management platform is used to store the user's historical intervention data, autobiographical text, and memoir audio and generate a personalized intervention report. Using the user's MMSE score, task completion efficiency, physiological signal temporal characteristics, and the results of the emotional analysis of autobiographical keywords as inputs and the intervention strategy as the output, continuously train the machine learning model to optimize the scenario recommendation strategy. The physiological signal temporal characteristics can be the physiological signal characteristics corresponding to each atomic action step during the intervention process. At the same time, a child access interface is provided to support retrieving the user's memory content for nostalgic retention according to the timeline or theme classification.

[0075] Through the collaborative intervention of occupational scenarios and nostalgic scenarios, dynamic adaptive regulation, and personalized scenario generation, the present invention solves the problems of single data, static state, and lag in physiological signal monitoring in traditional intervention means. Experiments show that the intervention system significantly improves the MMSE scores of middle-aged and elderly users (with an average increase within 3 months), reduces the dropout rate during training from 40% to less than 10%, reduces the peak stress by 30%, and increases the scene acceptance and immersion scores by 50%. The intervention system has significant advantages in improving cognitive rehabilitation efficiency, relieving psychological stress, and enhancing user participation.

[0076] As Figure 2 shown, a dynamic adaptive cognitive and mental health intervention method for middle-aged and elderly people includes the following steps:

[0077] Step 101: Decompose complex occupational tasks into multiple atomic action steps based on a task decomposition algorithm, associate each atomic action step with a cognitive training goal, and use a real-time error correction mechanism to conduct occupational task training;

[0078] Step 102: Collect and process the physiological data of the user in real time, where the physiological data includes the gaze point trajectory and pupil diameter change, galvanic skin signal, and electromyogram signal;

[0079] Step 103: Calculate the cognitive load index and emotional stress index of the user based on the user's physiological data, and determine whether the cognitive load index and emotional stress index exceed the set thresholds. When the cognitive load index exceeds the cognitive load set threshold, reduce the complexity of the occupational scenario task and continue training; when the emotional stress index exceeds the emotional stress set threshold, switch the occupational scenario to a nostalgic scenario;

[0080] Step 104: During the interaction in the nostalgic scenario, embed mild cognitive tasks in the nostalgic scenario, dynamically adjust the scene elements through an emotional intensity regulator according to the user's physiological state; embed trigger objects in the nostalgic scenario, associate the trigger objects with specific memory fragments, and when the user interacts with the trigger objects, play the associated memory fragments, and train a long short-term memory network based on the interaction sequence data and emotional state data of the user with the trigger objects to predict the optimal emotional support timing;

[0081] Step 105: At the end of the training, store the user's historical intervention data, autobiographical text, and memoir audio, and generate a personalized intervention report; optimize the scene recommendation strategy according to the user's MMSE score, task completion efficiency, physiological signal time series characteristics, autobiographical keyword emotional analysis results, and intervention strategy training machine learning model.

[0082] The embodiments described above are only descriptions of the preferred embodiments of the present invention, and do not limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations, variations, modifications, and substitutions made by those of ordinary skill in the art to the technical solutions of the present invention shall fall within the protection scope determined by the claims of the present invention.

Claims

1. A dynamic adaptive cognitive and mental health intervention system for the middle-aged and elderly, characterized in that: It includes a VR interaction module, a physiological data acquisition and processing module, a dynamic adaptive module, and a cloud management platform; the VR interaction module is used for information interaction between the user and the VR scene, and the interaction scenes include a professional scene and a nostalgic scene; The physiological data acquisition and processing module acquires and processes the user's multi-modal physiological data in real time; The dynamic adaptive module calculates the cognitive load index and the emotional stress index based on the physiological data, dynamically adjusts the complexity of professional tasks and triggers the switching of nostalgic scenes through a fuzzy logic controller, and embeds mild cognitive tasks in the nostalgic scene to maintain the continuity of training; the cloud management platform stores the user's historical data and optimizes the scene recommendation strategy through a machine learning model.

2. The dynamic adaptive middle-aged and elderly cognitive and mental health intervention system according to claim 1, characterized in that: The VR interaction module includes a professional scene interaction unit, a nostalgic scene interaction unit, and an immersive dialogue unit; the professional scene interaction unit constructs professional tasks based on a task decomposition algorithm and uses a real-time error correction mechanism for task training; The nostalgic scene interaction unit generates a nostalgic scene through a generative adversarial network according to the user's historical daily data; according to the magnitude of the emotional stress index, it dynamically adjusts the scene elements through an emotional intensity regulator, and the scene elements include the care color tone and the background music rhythm; The immersive dialogue unit uses natural language processing technology to guide the user to orally describe the content of their autobiography and memoirs, real-time records the user's voice data and converts it into structured text, so as to embed the user's autobiographical keywords and memory fragments in the nostalgic scene unit to enhance the scene relevance.

3. The dynamic adaptive middle-aged and elderly cognitive and mental health intervention system according to claim 2, characterized in that: The nostalgic scene interaction unit also embeds trigger objects in the nostalgic scene, associates each trigger object with a specific memory fragment, and when the user interacts with the trigger object, the associated memory fragment is played, and the interaction sequence data and emotional state data of the user during this process are obtained to train a long short-term memory network to predict the optimal emotional support timing.

4. A dynamic adaptive cognitive and mental health intervention system for the middle-aged and elderly according to claim 1, characterized in that: The physiological data acquisition and processing module includes a physiological data acquisition unit, an eye movement analysis unit, and a galvanic skin signal processing unit; The physiological data acquisition unit real-time monitors the user's fixation point trajectory, pupil diameter change, skin conductivity, upper limb movement accuracy and coordination; the eye movement analysis unit extracts a fixation heat map through a convolutional neural network to identify the attention dispersion area, and when the heat map deviates from the task key area by more than 30%, it triggers an attention correction prompt; The galvanic skin signal processing unit uses wavelet transform to remove motion artifacts and extracts the sympathetic nerve activation signal in the frequency band of 0.05 - 0.15Hz.

5. The dynamic adaptive middle-aged and elderly cognitive and mental health intervention system according to claim 1, characterized in that: The dynamic adaptive module includes a calculation unit, a dual-threshold mechanism adjustment unit, and a cross-scene continuity adjustment unit; the calculation unit calculates the cognitive load index and the emotional stress index based on the user's physiological data; The dual-threshold mechanism adjustment unit respectively determines whether the cognitive load index and the emotional stress index exceed the set thresholds: when the cognitive load index exceeds the cognitive load set threshold, the complexity of the professional scene task is reduced; when the emotional stress index exceeds the emotional stress set threshold, the professional scene is switched to the nostalgic scene; The cross-scene continuity adjustment unit embeds mild cognitive tasks in the nostalgic scene to continue cognitive training while providing emotional support.

6. The dynamic adaptive middle-aged and elderly cognitive and mental health intervention system according to claim 5, characterized in that: The calculation of the cognitive load index based on the user's physiological data in the calculation unit includes: Extraction of physiological index features: The extraction of eye movement indexes includes the fixation point trajectory and the change in pupil diameter. The extraction of the fixation point trajectory is to calculate the number of fixation point transfers per unit time and the proportion of the residence time of the fixation point in the key area. The extraction of the change in pupil diameter is to calculate the change rate of the pupil diameter relative to the initial pupil diameter during the training process. The extraction of the galvanic skin index is to calculate the change in skin conductivity relative to the initial average skin conductivity during the training process. The extraction of the electromyogram index is to calculate the root mean square value of the electromyogram signals of the muscles related to the upper limb movements and calculate the correlation of the electromyogram signals of different muscles. Determination of the weights of various indexes: Weight coefficients are assigned to the eye movement index, the galvanic skin index, and the electromyogram index according to experience. The calculation formula of the cognitive load index CLI is as follows: CLI = k1·L 眼动 + k2·L 皮电 + k3·L 肌电 ; Wherein, k1, k2, and k3 respectively represent the weight coefficients of the eye movement index, the galvanic skin response index, and the electromyogram index, and k1 + k2 + k3 = 1; L 眼动 is the quantization value of the comprehensive eye movement feature, which is obtained by weighted summation of three features: the number of fixation point transfers, the proportion of the residence time in the key area, and the pupil diameter change rate; L 皮电 is the quantization value of the change amount of skin conductivity; L 肌电 is the quantization value of the root mean square value and correlation of the electromyogram signal.

7. A dynamic adaptive cognitive and mental health intervention system for the middle-aged and elderly according to claim 5, characterized in that: In the calculation unit, the calculation formula of the emotional stress index is as follows: In the formula, N is the number of peaks of the galvanic skin signal; T is the time window; A is the average amplitude of the galvanic skin signal.

8. A dynamic adaptive cognitive and mental health intervention system for the middle-aged and elderly according to claim 1, characterized in that: The cloud management platform includes a database, an intervention strategy optimization unit, and a query unit. The database is used to store the user's historical data, autobiographical text, and memoir audio and generate a personalized intervention report. The intervention strategy optimization unit takes the user's MMSE score, task completion efficiency, physiological signal time series characteristics, and the emotional analysis results of the autobiographical keywords as inputs and the intervention strategy as the output to train a machine learning model to optimize the scenario recommendation strategy. The query unit supports retrieving the user's memoir content by timeline and / or topic classification.

9. A dynamic adaptive cognitive and mental health intervention method for the middle-aged and elderly, characterized in that: It includes the following steps: Step 101: Decompose complex occupational tasks into multiple atomic action steps based on the task decomposition algorithm, associate each atomic action step with the cognitive training goal, and use a real-time error correction mechanism for occupational task training. Step 102: Real-time collect and process the user's physiological data, where the physiological data includes the fixation point trajectory and the change in pupil diameter, galvanic skin signals, and electromyogram signals. Step 103: Calculate the user's cognitive load index and emotional stress index based on the user's physiological data, and determine whether the cognitive load index and the emotional stress index exceed the set thresholds. When the cognitive load index exceeds the cognitive load set threshold, the complexity of the occupational scenario task is reduced and the training continues. When the emotional stress index exceeds the emotional stress set threshold, the occupational scenario is switched to the nostalgic scenario. Step 104: Mild cognitive tasks are embedded in the nostalgic scenario, and the scenario elements are dynamically adjusted through an emotional intensity regulator according to the user's physiological state. Trigger objects are embedded in the nostalgic scenario, and the trigger objects are associated with specific memory segments. When the user interacts with the trigger objects, the associated memory segments are played, and a long short-term memory network is trained based on the user's interaction sequence data and emotional state data to predict the optimal emotional support timing. After the training is completed, the user's historical intervention data is stored and a personalized intervention report is generated. A machine learning model is trained based on the user's MMSE score, task completion efficiency, physiological signal time series characteristics, the emotional analysis results of the autobiographical keywords, and the intervention strategy to optimize the scenario recommendation strategy.

Images