Intelligent sitting posture correction method and system

Abstract

The present invention relates to the field of health monitoring and intelligent control technology, and in particular to an intelligent sitting posture correction method and system, which combines a piezoelectric sensor, an image recognition device and an infrared sensor to achieve comprehensive monitoring and data collection of a user's sitting posture; through data fusion and normalization processing, an adaptive neural network model is constructed to perform personalized sitting posture analysis; the system generates short-term and long-term sitting posture trends through multi-level time series analysis, and optimizes the prediction results by combining heterogeneous data cross-validation; based on the optimized prediction data, a multi-objective feedback strategy is designed to generate and apply tactile, visual and audio feedback; through real-time monitoring and dynamic control mechanisms, the system can adaptively adjust the feedback strategy, generate control optimization data, and achieve personalized sitting posture correction for users; the present invention effectively improves the accuracy of sitting posture detection and user experience, and helps users develop good sitting habits.

Description

Technical Field

[0001] The present invention relates to the technical field of health monitoring and intelligent control, and in particular to an intelligent sitting posture correction method and system. Background Art

[0002] Intelligent posture correction technology is of great significance in modern society, especially for people who need to sit for long periods of time to work or study. Correct sitting posture can help prevent back pain, spinal problems, and other posture-related health problems. By helping users maintain a correct sitting posture, posture correction technology can not only improve an individual's physical health, but also improve work efficiency and quality of life.

[0003] Existing sitting posture correction technologies are mainly divided into two categories: one is the fixation of physical devices, such as traditional sitting posture correction chairs, which have no electronic monitoring function and are difficult to adjust the sitting posture in real time; the other category (Chinese invention patent, publication number: CN115665199B, name: sitting posture correction system, method and smart seat) is to install electronic devices such as cameras and pressure sensors on the seats to detect and correct the user's sitting posture. However, these methods have significant defects. First, there is a risk of privacy leakage in camera recognition technology, which makes users unwilling to use it. Secondly, traditional pressure sensors can usually only detect pressure distribution but not the user's vital signs. Therefore, when heavy objects are placed on the chair, the sitting posture alarm may be mistakenly triggered, and there is a problem of gravity interference. In addition, these systems cannot accurately detect changes in the user's vital signs, resulting in the inability to provide accurate sitting posture correction during actual use, and the user experience is poor. Summary of the invention

[0004] In view of the many problems existing in the above-mentioned prior arts, the present invention provides an intelligent sitting posture correction method and system. The present invention integrates a variety of sensor technologies and combines an adaptive neural network model to monitor and analyze the user's sitting posture data in real time and provide personalized correction suggestions; the system uses multi-level time series analysis to predict the user's posture change trend and improves the prediction accuracy through heterogeneous data cross-validation; finally, the system improves the accuracy of sitting posture correction and user experience by dynamically adjusting the feedback strategy, helping users to form and maintain healthy sitting habits.

[0005] The intelligent sitting posture correction method comprises the following steps:

[0006] Using piezoelectric sensors, dynamic flexible materials, multi-view image recognition devices, infrared depth sensors and sound wave reflection equipment, the user's sitting posture data is collected, and time synchronization, data integration, noise filtering and feature preprocessing are performed on the sitting posture data to generate preprocessed feature data;

[0007] Extracting posture angle feature data, pressure distribution feature data and dynamic change feature data from the preprocessed feature data, fusing and normalizing the extracted feature data, and using the normalized feature data to construct and complete the training of the adaptive neural network model;

[0008] Based on the normalized feature data, multi-level time series analysis is performed to generate short-term posture trend data, medium-term posture pattern data and long-term posture habit data, and future situation prediction data is generated in combination with the adaptive neural network model, and the prediction data is optimized through heterogeneous data cross-validation;

[0009] Design a multi-objective feedback strategy based on the optimization prediction data, generate feedback strategy data, generate and apply tactile feedback data, visual feedback data and audio feedback data, adjust the feedback strategy through real-time monitoring and dynamic control mechanism, and generate control optimization data;

[0010] The adjusted feedback strategy data is implemented, application feedback data is generated through a multimodal feedback mechanism, user response data is recorded, feedback strategy optimization and model parameter adjustment are performed based on feedback effect data and self-organizing mapping, and optimized model data is generated.

[0011] Preferably, the time synchronization step includes assigning an accurate timestamp to the sitting posture data collected by each sensor, unifying all data on a time axis through a timing alignment algorithm to ensure the consistency of data from different sensors at the same time point, and generating time synchronization data.

[0012] Preferably, the multi-level timing analysis step includes:

[0013] Use sliding window technology to analyze normalized feature data in a short period of time to identify rapidly changing posture features and generate short-term posture trend data;

[0014] The moving average method is used to perform mid-term analysis on the normalized feature data to identify the periodic change characteristics of the sitting posture and generate mid-term posture pattern data;

[0015] Exponential smoothing method is applied to perform long-term trend analysis on normalized feature data to evaluate the long-term stability of posture and generate long-term posture habit data.

[0016] Preferably, the calculation of the short-term posture trend data is based on the following steps:

[0017] For each time window, the short-term posture trend is determined by calculating the difference between the sitting posture data in the window and its average value. The calculation formula is:

[0018]

[0019] in, It is short-term attitude trend data; is the size of the sliding window; is the current sitting posture data; is the average value of the sitting posture data within the sliding window.

[0020] Preferably, the training step of the adaptive neural network model uses alternating least squares method for parameter optimization, which gradually adjusts the weight parameters of the model to minimize the prediction error, so that the model can be adaptively adjusted to adapt to the changes in sitting posture data of different users.

[0021] Preferably, the heterogeneous data cross-validation step includes comparing the future situation prediction data with data obtained through multi-view image recognition, infrared depth sensing and sound wave reflection, and improving the accuracy of the prediction data through consistency verification and error correction.

[0022] Preferably, the multi-objective feedback strategy step includes determining the feedback type and intensity through weight adjustment according to the user's personalized preferences and correction goals. The multi-objective feedback strategy optimizes the feedback method to maximize the user's acceptance and comfort through dynamic learning and adaptive adjustment of user preferences.

[0023] Preferably, the real-time monitoring and dynamic regulation mechanism includes: monitoring the user's posture changes and feedback effects in real time through sensors, and dynamically adjusting feedback parameters according to the collected real-time data to generate real-time regulation data to optimize the timeliness and effectiveness of feedback.

[0024] Preferably, the feedback effect data is calculated by the following steps:

[0025] In each feedback cycle, the relative sum of squares of the error between the user response data and the predicted response data is calculated using the following formula:

[0026]

[0027] in, To provide feedback effect data; is the number of response data points within the feedback cycle; is the feedback response value; is the predicted feedback value.

[0028] A system for implementing the intelligent sitting posture correction method, comprising:

[0029] A multimodal data acquisition module, including a piezoelectric sensor, a dynamic flexible material, a multi-view image recognition device, an infrared depth sensor, and an acoustic wave reflection device, for collecting the user's sitting posture data;

[0030] A data processing module, used for performing time synchronization, data integration, noise filtering and feature preprocessing on the collected sitting posture data to generate preprocessed feature data;

[0031] A feature extraction and fusion module, used to extract posture angle feature data, pressure distribution feature data and dynamic change feature data from the pre-processed feature data, and to fuse and normalize the extracted feature data;

[0032] An adaptive neural network training module, used to construct and train an adaptive neural network model using normalized feature data;

[0033] A time series analysis and prediction module is used to perform multi-level time series analysis based on normalized feature data, generate short-term posture trend data, medium-term posture pattern data, and long-term posture habit data, and generate future situation prediction data in combination with the adaptive neural network model, and optimize the prediction data through heterogeneous data cross-validation;

[0034] Feedback generation and regulation module, which is used to design multi-objective feedback strategies based on optimization prediction data, generate feedback strategy data, and generate tactile feedback data, visual feedback data, and audio feedback data, and adjust the feedback strategy through real-time monitoring and dynamic regulation mechanism;

[0035] A multimodal feedback mechanism module, used to implement the adjusted feedback strategy data, generate application feedback data, and record user response data;

[0036] The feedback optimization and model adjustment module is used to optimize the feedback strategy and adjust the model parameters based on the feedback effect data and self-organizing map to generate optimized model data.

[0037] Compared with the prior art, the advantages and beneficial effects of the present invention are:

[0038] The present invention realizes comprehensive monitoring of the user's sitting posture through high-precision piezoelectric sensors and dynamic flexible materials, significantly improving the accuracy of data and the response speed of the system;

[0039] The present invention realizes the collection of multi-dimensional posture data through the integration of a multi-view image recognition device and an infrared depth sensor, reduces false alarms caused by heavy objects, and enhances the accuracy of sitting posture judgment;

[0040] The present invention realizes the automatic generation of personalized sitting posture correction scheme for users through adaptive neural network model training and multi-level time series analysis, thereby improving the intelligence level of the system and user experience;

[0041] The present invention realizes real-time adjustment of personalized feedback strategy through a multimodal feedback mechanism, combining tactile, visual and audio feedback, and improves user acceptance and feedback effectiveness. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is a schematic diagram of the process of the present invention;

[0043] Figure 2 It is a schematic diagram of data collection and preprocessing in the present invention;

[0044] Figure 3 It is a schematic diagram of the adaptive neural network training process in the present invention;

[0045] Figure 4 It is a schematic diagram of the multi-objective feedback strategy flow in the present invention;

[0046] Figure 5 It is a structural block diagram of the system of the present invention. DETAILED DESCRIPTION

[0047] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present disclosure. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is obvious that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0048] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise", "include", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0049] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0050] like Figure 1-Figure 2 As shown, the intelligent sitting posture correction method comprises the following steps:

[0051] Using piezoelectric sensors, dynamic flexible materials, multi-view image recognition devices, infrared depth sensors and sound wave reflection equipment, the user's sitting posture data is collected, and time synchronization, data integration, noise filtering and feature preprocessing are performed on the sitting posture data to generate preprocessed feature data;

[0052] Preferably, the time synchronization step includes assigning an accurate timestamp to the sitting posture data collected by each sensor, unifying all data on a time axis through a timing alignment algorithm to ensure the consistency of data from different sensors at the same time point, and generating time synchronization data.

[0053] In the present invention, data acquisition is the basic link of the entire system. Through the combined use of piezoelectric sensors, dynamic flexible materials, multi-view image recognition devices, infrared depth sensors and sound wave reflection equipment, the system can comprehensively and accurately obtain the user's sitting posture data. The piezoelectric sensor is used to detect the pressure changes at different positions on the seat, and can reflect the user's sitting center of gravity and posture stability in real time. The dynamic flexible material is combined with the piezoelectric sensor to provide a response to the deformation of the seat surface and capture subtle changes in sitting posture. The multi-view image recognition device captures the user's image from multiple angles and combines it with computer vision algorithms to identify the user's posture profile and dynamic changes. The infrared depth sensor is used to obtain the depth information between the user and the surrounding environment, and accurately describe the three-dimensional spatial distribution of the sitting posture. The sound wave reflection device further refines the sitting posture data by analyzing the changes caused by the reflection of sound waves on the user's body surface.

[0054] In the present invention, dynamic flexible materials are important components for detecting and responding to changes in sitting posture. Dynamic flexible materials generally refer to materials that can change their electrical or mechanical properties according to external pressure or deformation, which allows them to be used to monitor and provide feedback on the user's sitting posture in real time. The main function of these materials in the system is to work with piezoelectric sensors to provide more detailed sitting posture data collection.

[0055] Common dynamic flexible materials include:

[0056] Polyvinylidene fluoride (PVDF) film, PVDF is a piezoelectric material that generates an electric charge when subjected to mechanical stress. It has good flexibility and stability. It can be used to make piezoelectric sensors for detecting pressure changes on seats.

[0057] Carbon nanotube (CNTs) composite materials, carbon nanotubes have excellent conductivity and mechanical strength. Combining them with flexible matrix materials can form highly sensitive sensors. They are used to make flexible pressure sensors and strain sensors, which are suitable for monitoring changes in sitting posture.

[0058] Conductive polymers (such as polypyrrole, polyaniline) are polymers that can change their conductivity under pressure or deformation. They have good flexibility and processability. They are used to make flexible resistive sensors suitable for detecting subtle posture changes.

[0059] Conductive silver nanowire (AgNW) film, silver nanowire film has good transparency and conductivity, suitable for making flexible and transparent sensors. It is used to make tactile sensors and pressure sensors for seat surface monitoring.

[0060] Graphene-based materials, graphene has excellent electrical properties and mechanical strength. Graphene films will produce obvious electrical changes under deformation. It is used to manufacture highly sensitive pressure and strain sensors.

[0061] Liquid metal conductive materials (such as gallium alloys) have good conductivity and deformability, and can maintain conductivity changes when deformed. They are used to make stretchable flexible sensors, which are suitable for dynamic detection.

[0062] The working principle of dynamic flexible materials is based on their ability to respond to physical stimuli such as pressure, stretching or bending. When these materials are subjected to external forces, they deform, thereby changing their internal electrical properties such as resistance, capacitance or conductance. These changes can be sensed and converted into electrical signals, which the system uses to monitor the user's sitting posture changes in real time.

[0063] Common practical applications of dynamic flexible materials include:

[0064] (1) Graphene-based flexible sensors

[0065] Graphene is a material with excellent conductivity and flexibility. By depositing a graphene layer on a flexible substrate, a graphene-based flexible sensor can be made. This sensor can detect very small changes in pressure and is suitable for high-precision sitting posture monitoring. For example, when a user moves in a seat, the graphene sensor can capture changes in pressure distribution and transmit data to the system for analysis.

[0066] (2) Flexible piezoelectric film

[0067] Flexible piezoelectric films such as polyvinylidene fluoride (PVDF) films are able to generate electric charge when subjected to mechanical stress. These films are integrated into the seat surface and can sense changes in the user's sitting posture and feed the sitting posture information back to the system for processing. Due to its flexibility and thinness, this film does not affect the comfort of the seat.

[0068] (3) Flexible electronic fabrics

[0069] Flexible electronic fabrics are textiles that have conductive fibers or sensor elements embedded in them, giving the fabric itself a sensing function. This material can be used to cover the surface of a seat. When the user's sitting posture changes, the sensor elements in the fabric sense the pressure and deformation changes and provide real-time sitting posture data. The softness of this fabric is very suitable for long-term monitoring of user sitting posture.

[0070] By using dynamic flexible materials, the present invention can provide more comprehensive and accurate sitting posture data. Combined with other types of sensors, these materials can enhance the system's ability to perceive changes in the user's sitting posture, and improve the accuracy and real-time nature of sitting posture monitoring. Ultimately, this high-precision data collection provides the system with a basis for optimizing feedback and improving the user's sitting posture, helping users to form and maintain healthy sitting habits.

[0071] After data collection, time synchronization is a key step to ensure data consistency and accuracy. The data collected by each sensor is given a precise timestamp, and a timing alignment algorithm is used to unify all data on a time axis to form time-synchronized data. This data processing method ensures that at the same time point, the data from different sensors are consistent and can truly reflect the user's sitting posture. The time-synchronized data will then be integrated and noise filtered to remove environmental interference and sensor noise, and ultimately generate high-quality pre-processed feature data. The generation of this data provides a reliable basis for subsequent posture analysis and feedback.

[0072] For example, in actual implementation, by installing the above-mentioned sensing devices on office chairs, the system can monitor the user's sitting posture in real time. When the user has a bad sitting posture, such as leaning forward or backward for a long time, the piezoelectric sensor and dynamic flexible material can quickly detect the pressure changes of the seat cushion and backrest, and determine the user's posture angle in combination with the data of the image recognition device. The infrared depth sensor provides information on the distance between the user and the desk, helping to further identify whether the sitting posture meets ergonomic standards. The sound wave reflection device accurately captures the body's subtle posture adjustments by analyzing the time delay and amplitude changes of the sound waves. After time synchronization and preprocessing, these data provide complete posture characteristics, ensuring that the system can generate effective correction feedback in a short time.

[0073] like Figure 3 As shown, posture angle feature data, pressure distribution feature data and dynamic change feature data are extracted from the pre-processed feature data, the extracted feature data are fused and normalized, and the normalized feature data are used to construct and complete the training of the adaptive neural network model;

[0074] In the present invention, feature extraction and data fusion are the core links to realize sitting posture recognition and analysis. First, three key features are extracted from the preprocessed feature data: posture angle feature data, pressure distribution feature data and dynamic change feature data. The posture angle feature data is used to describe the user's sitting angle and posture direction in three-dimensional space, and help identify whether there are bad sitting postures such as leaning forward or leaning back. The pressure distribution feature data determines the center of gravity distribution and sitting posture stability of the user by analyzing the pressure changes in different areas of the seat. The dynamic change feature data evaluates the dynamic adjustment and stability of the sitting posture by capturing the user's posture changes at different time points.

[0075] The extracted feature data is fused and normalized to form unified normalized feature data. Fusion processing combines data from multiple sensors to generate a comprehensive description of sitting posture. Normalization processing converts feature data of different dimensions to the same scale to eliminate the impact of numerical differences and ensure data consistency in neural network training. This step ensures the quality and consistency of feature data and lays the foundation for subsequent neural network training.

[0076] The construction and training of the adaptive neural network model uses normalized feature data, and gradually optimizes the weight and structure of the model through a combination of supervised learning and unsupervised learning. The neural network can adaptively learn the user's sitting posture pattern and identify individual posture characteristics and habits. Through continuous feedback adjustment, the model can improve the recognition accuracy of sitting posture changes and achieve personalized sitting posture correction.

[0077] For example, in the actual application of the system, after a period of training, the neural network can accurately identify common bad sitting postures of a user at work, such as bending over the desk for a long time or leaning back too much. After collecting and processing the data, the system uses neural network analysis to determine the user's current posture and provide instant feedback, such as adjusting the seat tilt angle or providing voice reminders. This process not only helps users adjust their sitting posture in real time and improve their health habits, but also provides personalized health advice through long-term data accumulation.

[0078] Based on the normalized feature data, multi-level time series analysis is performed to generate short-term posture trend data, medium-term posture pattern data and long-term posture habit data, and future situation prediction data is generated in combination with the adaptive neural network model, and the prediction data is optimized through heterogeneous data cross-validation;

[0079] In the present invention, multi-level time series analysis is a key step to deeply analyze the user's sitting posture dynamics. Through the time series analysis of normalized feature data, the system can capture the short-term trend, medium-term pattern and long-term habits of the user's sitting posture. Short-term posture trend data is used to identify the user's immediate sitting posture changes and sudden posture adjustments by analyzing posture changes in a short period of time. For example, a user may have multiple posture adjustments during the working hours of a day, and these adjustments can be quickly identified through short-term trend analysis, thereby providing immediate posture adjustment suggestions.

[0080] Mid-term posture pattern data identifies periodic changes in the user's sitting posture through analysis over a longer time scale, such as the user's tendency to lean back to rest every afternoon. The recognition of this pattern helps the system provide more targeted correction suggestions to help users adjust bad habits. Long-term posture habit data evaluates the user's sitting stability and long-term health effects by analyzing longer-term sitting data. This can reveal the sitting habits formed by users over a period of time and be used to formulate long-term health recommendations.

[0081] By combining the adaptive neural network model, the system uses these time series analysis data to generate future posture prediction data. These prediction data can help the system identify the user's possible bad sitting posture trends in advance and provide corresponding preventive measures. For example, the system may predict that the user will have a bad posture due to fatigue after working at a desk for a long time in the afternoon. The system can send a reminder before the expected time to prevent this from happening.

[0082] In addition, the system optimizes the prediction data through cross-validation of heterogeneous data. Heterogeneous data sources include different sensor data and external data sources. By cross-validating these data, the system can improve the accuracy and reliability of the prediction data. For example, by comparing the posture prediction data with the infrared depth sensing data, it can more accurately confirm whether the prediction results are consistent with the actual posture changes.

[0083] Preferably, the multi-level timing analysis step includes:

[0084] Use sliding window technology to analyze normalized feature data in a short period of time to identify rapidly changing posture features and generate short-term posture trend data;

[0085] The moving average method is used to perform mid-term analysis on the normalized feature data to identify the periodic change characteristics of the sitting posture and generate mid-term posture pattern data;

[0086] Exponential smoothing method is applied to perform long-term trend analysis on normalized feature data to evaluate the long-term stability of posture and generate long-term posture habit data.

[0087] In the present invention, multi-level time series analysis is a key processing step for comprehensively analyzing the dynamic characteristics of user sitting posture. This process involves three main analysis techniques: sliding window technique, moving average method and exponential smoothing method. Each technique has its own focus, and combined with each other, it can provide short-term, medium-term and long-term analysis of user sitting posture.

[0088] Sliding window technology is used to analyze normalized feature data in a short period of time. The principle is to divide the time series data into multiple short time periods (i.e., windows) and analyze the data changes in each window. This method can quickly capture the changing characteristics of posture and identify posture fluctuations in a short period of time. For example, the user's sitting posture adjustment within an hour, such as leaning forward, leaning back, or twisting, can be detected in time through the sliding window technology, thereby generating short-term posture trend data. Such trend data can provide instant feedback information to the system, helping users to adjust bad sitting postures in a shorter period of time.

[0089] The moving average method is used for medium-term analysis. It calculates the average value of data over a period of time to identify periodic changes in sitting posture. This method can smooth out short-term random fluctuations and highlight the medium-term trend of the data. For example, a user may habitually change his sitting posture at a certain time of the day. This periodic change can be identified by the moving average method to generate medium-term posture pattern data. By analyzing this pattern, the system can provide users with more targeted sitting posture adjustment suggestions and optimize users' daily sitting habits.

[0090] Exponential smoothing is used for long-term trend analysis, focusing on evaluating the long-term stability of posture. This method gives higher weight to the latest data and gradually reduces the impact of historical data, thereby better reflecting long-term trends. For example, the system can identify the user's habit of sitting in a tilted position that has gradually formed over the past few months. These long-term posture habit data can help the system provide long-term health advice and remind users to adjust long-term bad postures in time to avoid potential health risks.

[0091] Through the combination of these technologies, the system realizes a comprehensive analysis of the dynamic characteristics of the user's sitting posture. The sliding window technology ensures sensitive detection of short-term changes, the moving average law provides a reliable analysis of medium-term patterns, and the exponential smoothing law reveals long-term trends and stability. The overall effect significantly improves the depth and breadth of sitting posture analysis, providing users with a full range of sitting posture monitoring and adjustment solutions. The implementation of this method is reflected in the system's ability to promptly remind users to adjust their sitting posture while working, provide personalized health advice, and help users improve their long-term sitting habits.

[0092] Preferably, the calculation of the short-term posture trend data is based on the following steps:

[0093] For each time window, the short-term posture trend is determined by calculating the difference between the sitting posture data in the window and its average value. The calculation formula is:

[0094]

[0095] in, It is short-term attitude trend data; is the size of the sliding window; is the current sitting posture data; is the average value of the sitting posture data within the sliding window.

[0096] In the present invention, the calculation of short-term posture trend data is achieved by analyzing the changes in sitting posture data in a short period of time. This process uses sliding window technology to quickly identify the dynamic changes of the user's sitting posture and provide a basis for immediate feedback and correction. Sliding window technology is a time series analysis method that divides continuous data into multiple windows of fixed length and performs local analysis in each window.

[0097] Specifically, the system segments the preprocessed sitting posture data into multiple sliding windows, each of which contains a set of continuous sitting posture data points. For each time window, the difference between each sitting posture data point in the window and the average value of the window data is calculated, and the average value of these differences is the short-term posture trend.

[0098] The principle of this formula is to evaluate the volatility of sitting posture by measuring the deviation of each sitting posture data point from the average. A large difference means that the user has undergone significant posture changes during the period, which can reflect the instability of sitting posture or rapid adjustment.

[0099] For example, in practical applications, suppose the system detects that a user frequently leans forward to look at the computer screen during work. Through sliding window technology, the system can monitor this leaning trend in real time and calculate the increase in short-term posture trend data. Based on this trend, the system can issue a reminder in time to suggest that the user adjust his posture to avoid fatigue or discomfort caused by long-term poor sitting posture.

[0100] Through the method of the present invention, the intelligent sitting posture correction system can efficiently identify the user's sitting posture change trend in a short time and provide immediate correction suggestions. The effect is to help users maintain good sitting habits and prevent discomfort or health problems caused by frequent posture changes in a short period of time. This real-time posture monitoring and feedback mechanism is the basis for the system to provide personalized health management services, ensuring that users can maintain the best sitting posture in daily work.

[0101] Preferably, the training step of the adaptive neural network model uses alternating least squares method for parameter optimization, which gradually adjusts the weight parameters of the model to minimize the prediction error, so that the model can be adaptively adjusted to adapt to the changes in sitting posture data of different users.

[0102] In the present invention, the adaptive neural network model is one of the core components, which is used to identify and predict the user's sitting posture pattern. In order to improve the adaptability of the model to different users, the system uses the alternating least squares method for model training and parameter optimization. The alternating least squares method is an optimization algorithm specifically used to solve the optimization process of minimizing errors in linear regression problems. Its core is to reduce the prediction error by iteratively adjusting the model parameters.

[0103] In the specific implementation, the system first performs preliminary training of the neural network through normalized feature data to establish a basic model. Then, the alternating least squares method is applied to further optimize the weight parameters of the model. The principle of this method is to fix one parameter set and optimize another parameter set, and gradually converge to a local optimal solution through alternating optimization. This process can effectively reduce the error of model prediction and make the model fit the user's actual sitting posture data more accurately.

[0104] For example, when the system monitors the user's daily sitting posture, it may capture differences between individuals, such as different heights and body shapes of users in the same sitting posture will produce different pressure distribution and posture angles. Through alternating least squares, the neural network can adaptively adjust these parameters to better capture and understand the unique sitting posture characteristics of each user. After each data input, the system adjusts the weight of the network through an optimization algorithm to gradually reduce the prediction error, thereby improving the accuracy of the model.

[0105] This training method is effective in improving the prediction accuracy of the model and enhancing its adaptability, enabling it to provide personalized analysis and feedback on the sitting posture of different users. This means that no matter how the user changes, their sitting posture data can be effectively interpreted, and the system can provide accurate posture assessment and correction suggestions. Ultimately, this adaptive neural network model ensures the wide applicability and reliability of the intelligent sitting posture correction system among a diverse user group, providing users with more targeted health management services.

[0106] Preferably, the heterogeneous data cross-validation step includes comparing the future situation prediction data with data obtained through multi-view image recognition, infrared depth sensing and sound wave reflection, and improving the accuracy of the prediction data through consistency verification and error correction.

[0107] In the present invention, heterogeneous data cross-validation is an important step to ensure the accuracy of prediction data. This step compares the future situation prediction data with data from multiple sensors to perform consistency checks and error corrections, thereby improving the overall prediction accuracy. The system uses data obtained by multi-view image recognition, infrared depth sensing, and sound wave reflection technology as a reference benchmark and makes a detailed comparison with the future situation prediction data.

[0108] Multi-view image recognition provides two-dimensional and three-dimensional visual information of the user's sitting posture, which can intuitively reflect the user's posture changes. The infrared depth sensor provides accurate distance information between the user and the surrounding environment, which helps to identify the spatial position and angle of the sitting posture. The sound wave reflection technology captures subtle posture changes by measuring the reflection time and intensity of sound waves on the user's body surface. These data come from different sensors and belong to the category of heterogeneous data, each of which has different perception capabilities and error characteristics.

[0109] During the cross-validation process, the system first pre-processes the data from each sensor to ensure their consistency at the same time point. Then, the system compares the future situation prediction data with the data from these sensors for consistency check. The purpose of this step is to confirm the consistency between the prediction results and the actual observed data. For example, if the predicted data shows that the user tilted his body at a certain point in time, and the image recognition and depth sensing data also show similar trends, then this verifies the reliability of the predicted data.

[0110] If significant differences are found between the predicted data and the sensor data during the comparison process, the system will perform error correction. Error correction includes adjusting model parameters or correcting deviations in the predicted data to ensure that the final data reflects the user's actual sitting posture. This correction mechanism not only improves the accuracy of the system's predictions, but also enhances the system's ability to adapt to different environments and user conditions.

[0111] For example, in actual applications, the system may predict that the user tends to lean forward during the afternoon working hours. Through heterogeneous data cross-validation, image recognition and depth sensing data also show that the user tends to lean forward during this time period, which enhances the credibility of the prediction. Conversely, if the data of a certain sensor is inconsistent with the predicted result, the system will make adjustments through error correction.

[0112] like Figure 4 As shown, a multi-objective feedback strategy is designed according to the optimization prediction data, feedback strategy data is generated, tactile feedback data, visual feedback data and audio feedback data are generated and applied, the feedback strategy is adjusted through real-time monitoring and dynamic control mechanism, and control optimization data is generated;

[0113] In the present invention, the design of multi-objective feedback strategy is an important part of achieving personalized correction and optimizing user experience. Based on the optimized prediction data, the system designs a set of multi-objective feedback strategies and generates feedback strategy data. The goal of these strategies is to provide users with real-time posture correction feedback through multiple perception channels to help users maintain a good sitting posture.

[0114] The core of feedback strategy design is to determine the type and intensity of feedback based on the user's personalized needs and sitting posture. The system uses optimized prediction data, including the user's posture trends and habitual patterns, to dynamically adjust the feedback strategy. The generated feedback strategy data determines how to select and combine tactile, visual, and audio feedback to guide the user to adjust their sitting posture in the best way.

[0115] Haptic feedback reminds users through vibrations or slight changes in pressure. For example, micro vibrators are embedded in the back of a chair or seat cushion. When a bad sitting posture is detected, they will emit slight vibrations to prompt the user to correct the posture. Visual feedback can remind users through a display or LED light. The display can show posture correction suggestions, while the LED light can indicate the posture status through color changes. Audio feedback reminds users through voice prompts or sound effects. For example, when a user maintains a bad sitting posture for a long time, the system can play a voice to remind the user to adjust the posture.

[0116] Real-time monitoring and dynamic control mechanisms ensure that feedback strategies can be adjusted according to the user's immediate response. The system uses sensors to monitor the user's posture changes and feedback effects in real time, and records the user's response and adjustment speed. Based on these monitoring data, the system dynamically adjusts the feedback strategy and generates control optimization data. For example, if the user responds slowly to tactile feedback, the system may increase the intensity of visual or audio feedback to ensure that the user can adjust his sitting posture in time.

[0117] The implementation effect of this feedback mechanism is to improve the effectiveness of feedback and user acceptance through the joint application of multiple perception channels. Users can choose the feedback method that best suits them according to their own perception preferences and habits, thereby improving the initiative and comfort of sitting posture adjustment.

[0118] In actual applications, for example, if a user is used to sitting for a long time in the office, the system can detect his sitting posture trend and select appropriate feedback methods, using a vibration reminder to remind him when he leans too much, or using voice to remind him to stand up and move around when he stays still for a long time. This personalized feedback not only improves the user's correction efficiency, but also enhances the user's comfort experience and helps users develop good sitting habits.

[0119] Preferably, the multi-objective feedback strategy step includes determining the feedback type and intensity through weight adjustment according to the user's personalized preferences and correction goals. The multi-objective feedback strategy optimizes the feedback method to maximize the user's acceptance and comfort through dynamic learning and adaptive adjustment of user preferences.

[0120] In the present invention, the multi-objective feedback strategy is designed to provide users with the best posture correction experience through personalization and adaptive adjustment. This strategy uses a weight adjustment method to determine the type and intensity of feedback based on the user's personalized preferences and specific correction goals, thereby achieving efficient sitting posture correction.

[0121] In principle, the multi-objective feedback strategy dynamically adjusts the feedback mechanism to meet the user's personalized needs by analyzing the user's historical responses and real-time data. The system first uses a period of use to understand the user's acceptance of different feedback methods. For example, the user may be sensitive to tactile feedback but slow to respond to audio feedback. After collecting this data, the system uses a weight adjustment method to assign different weights to each feedback method to highlight the feedback type that the user prefers more, while weakening or adjusting the intensity of other feedback.

[0122] The core of this strategy lies in the ability to adaptively adjust. The system not only relies on the initial user preference data, but also continuously updates and optimizes the feedback strategy through a dynamic learning mechanism. Over time, the system will adjust the weight and intensity of various types of feedback in real time based on the user's actual adjustment effect and acceptance. For example, when the system finds that the user adjusts faster under the tactile feedback prompt, it can increase the weight of tactile feedback and reduce the intensity of visual or audio feedback to ensure that the user is always within a comfortable acceptance range.

[0123] For example, during the implementation of the system, when a user used the intelligent sitting posture correction system, the system recorded the user's posture adjustment speed and comfort under visual and audio feedback. After a period of data collection, the system found that users were more receptive to visual feedback and the adjustment effect was obvious. Therefore, the system automatically adjusted the feedback strategy, increasing the frequency and significance of visual prompts, while reducing the intensity of audio prompts, so that users can feel a more comfortable prompt rhythm during use.

[0124] This feedback strategy is effective. Through the multi-target feedback strategy, users can obtain correction suggestions that are more in line with their personal needs, reducing correction failures caused by not adapting to the feedback method. At the same time, the adaptive adjustment capability ensures that the system can continuously optimize the feedback strategy as user habits change, improving the user's long-term usage experience and health management effect.

[0125] Preferably, the real-time monitoring and dynamic regulation mechanism includes: monitoring the user's posture changes and feedback effects in real time through sensors, and dynamically adjusting feedback parameters according to the collected real-time data to generate real-time regulation data to optimize the timeliness and effectiveness of feedback.

[0126] In the present invention, the real-time monitoring and dynamic control mechanism is the key part to ensure the accuracy and timeliness of feedback. This mechanism monitors the user's posture changes and feedback effects in real time through sensors, and dynamically adjusts the feedback parameters to optimize the effectiveness of the feedback.

[0127] In principle, the mechanism relies on the coordinated work of multiple sensors in the system. These sensors include piezoelectric sensors, multi-view image recognition devices, infrared depth sensors, and sound wave reflection devices, which can fully capture the user's sitting posture data. These sensors provide data to the system in real time, enabling the system to continuously monitor the user's sitting posture changes and adjustments. When the user's posture deviates from the ideal sitting posture, the system can quickly identify and record this change.

[0128] Based on the real-time data collected, the system automatically analyzes to determine the effectiveness of the current feedback strategy. If the feedback effect is not good, for example, the user fails to adjust the posture in time after receiving feedback, the system will dynamically adjust the feedback parameters based on the real-time data. This adjustment may include changing the type of feedback (such as changing from tactile feedback to audio feedback), or changing the intensity and frequency of the feedback to ensure that the user can perceive and respond in time.

[0129] The generated real-time control data provides the system with a basis for further optimizing the feedback strategy. This data not only reflects the user's posture changes, but also records the user's response speed and effect to different feedback methods. Through comprehensive analysis of these data, the system can identify the most effective feedback mode, thereby providing more accurate feedback in future interactions. For example, if the system finds that the user responds to visual feedback faster than other feedback methods, it can give priority to visual feedback when the user deviates from the ideal sitting posture again.

[0130] For example, in the implementation of the system, when the user gradually starts to lean forward while working, resulting in a bad sitting posture, the sensor in the system detects this change and immediately reminds the user through tactile feedback (such as seat vibration). However, the monitoring data shows that the user did not adjust the posture in time. Based on the real-time control data, the system adjusts the feedback method to audio prompts and reduces the frequency of tactile feedback. This dynamic adjustment allows the user to respond quickly and return to a healthy sitting posture.

[0131] The effect of this real-time monitoring and dynamic control mechanism is that, by flexibly adjusting the feedback strategy, the system can always provide the best posture correction suggestions to ensure that users can quickly adjust to the correct sitting posture. Ultimately, this mechanism not only improves the efficiency of sitting posture correction and user experience, but also enhances the system's adaptability to different user needs, helping users develop and maintain good sitting habits and prevent health problems caused by bad sitting posture.

[0132] The adjusted feedback strategy data is implemented, application feedback data is generated through a multimodal feedback mechanism, user response data is recorded, feedback strategy optimization and model parameter adjustment are performed based on feedback effect data and self-organizing mapping, and optimized model data is generated.

[0133] In the present invention, implementing the adjusted feedback strategy is an important step to ensure the effectiveness of posture correction. Through the multimodal feedback mechanism, the system can provide diversified feedback according to the personalized needs of the user, which includes various forms such as touch, vision and audio. Tactile feedback may remind the user through vibration or pressure changes in the seat, visual feedback can convey information through prompts or color changes on the display, and audio feedback reminds the user to make adjustments through voice prompts or sounds. This multimodal feedback mechanism can effectively improve user acceptance and correction efficiency.

[0134] After implementing these feedback strategies, the system will record the user's response data to further analyze the user's response to different feedback methods. These response data include not only whether the user adjusts the sitting posture in time, but also information such as the adjustment speed and acceptance. The recording and analysis of these data can provide important basis for the system to optimize the feedback strategy.

[0135] Feedback effect data is obtained by calculating the error between user response data and predicted response data. Based on these feedback effect data, the system uses adaptive learning algorithms such as self-organizing maps (SOM) to optimize feedback strategies and adjust model parameters. Self-organizing maps are an unsupervised learning algorithm that can help the system automatically adjust feedback strategies and model parameters by identifying potential patterns in data to improve the adaptability and accuracy of the system.

[0136] For example, in the actual application of the system, suppose a user responds well to tactile feedback but is slow to respond to audio feedback. By recording the user's response data, the system finds that the user quickly adjusts his sitting posture after receiving tactile feedback, while audio feedback does not bring significant results. Based on these feedback effect data, the system uses self-organizing mapping to optimize the feedback strategy, reduce the frequency of audio feedback, increase the intensity of tactile feedback, and adjust the relevant parameters of the neural network model to better meet the user's personalized needs.

[0137] The effect of this optimization mechanism based on feedback effect data and self-organizing mapping is that through continuous feedback adjustment and model optimization, the system can continuously improve the efficiency of sitting posture correction and user experience. Ultimately, by generating optimized model data, the system can not only provide more accurate sitting posture correction suggestions, but also help users develop healthy sitting habits in the long term and reduce health problems caused by poor sitting posture. This process not only improves the intelligence level and responsiveness of the system, but also provides effective support for users' health management.

[0138] Preferably, the feedback effect data is calculated by the following steps:

[0139] In each feedback cycle, the relative sum of squares of the error between the user response data and the predicted response data is calculated using the following formula:

[0140]

[0141] in, To provide feedback effect data; is the number of response data points within the feedback cycle; is the feedback response value; is the predicted feedback value.

[0142] In the present invention, the calculation of feedback effect data is an important step for evaluating the effectiveness of the feedback strategy. By calculating the relative error sum of squares between the user response data and the predicted response data, the system can quantify the user's response to the feedback, thereby optimizing and adjusting the feedback strategy. This step uses an error analysis method to ensure that the system can provide the most effective sitting posture correction suggestions.

[0143] In principle, feedback effectiveness data is calculated by comparing the difference between actual user responses and the ideal responses predicted by the system.

[0144] Through this error calculation, the system can evaluate the effectiveness of each feedback strategy in actual applications. For example, in a feedback cycle, the system may predict that the user will quickly adjust their sitting posture after receiving tactile feedback (predicted feedback value), but the actual user response shows a slower adjustment speed (feedback response value). This difference will be reflected as a larger error value in the calculation. The relative sum of squared errors provides a way to quantify this difference, allowing the system to identify which feedback methods are most effective and which ones need further optimization.

[0145] In an embodiment, suppose the system detects that after receiving multiple visual feedbacks, the user's sitting posture adjustment speed gradually increases, while the audio feedback fails to effectively prompt the user to adjust. In this case, the error analysis will show that the error value of the visual feedback is small, while the error value of the audio feedback is large. Based on this data, the system can increase the frequency of visual feedback in future feedback strategies, while reducing or optimizing audio feedback to ensure that the user can respond quickly and adjust the posture.

[0146] The effect of this error analysis is that it provides a data-driven approach to optimize feedback strategies. Through continuous iteration and adjustment, the system can improve user experience and correction results. Ultimately, by accurately quantifying the user's response to different feedback methods, the intelligent sitting posture correction system can continuously improve the accuracy and personalization of correction suggestions, help users form and maintain good sitting habits, and improve overall health.

[0147] like Figure 5 As shown, a system for implementing the intelligent sitting posture correction method comprises:

[0148] The multimodal data acquisition module, including piezoelectric sensors, dynamic flexible materials, multi-view image recognition devices, infrared depth sensors and acoustic wave reflection devices, is used to collect the user's sitting posture data; the piezoelectric sensors are installed at key positions of the seat to detect the pressure distribution of the user when sitting in real time. These sensors can sense subtle pressure changes and help identify the center of gravity offset and imbalance of the sitting posture. The dynamic flexible material fits the seat surface and responds to the deformation of the seat, providing a basis for capturing dynamic changes in posture. The multi-view image recognition device uses computer vision algorithms to obtain user posture images from different angles to ensure the comprehensiveness of posture monitoring. The infrared depth sensor provides accurate three-dimensional spatial data by capturing the changes in the distance between the user and the surrounding objects. The acoustic wave reflection device uses the changes in the reflection of sound waves on the body surface to refine the details of the sitting posture. These devices work together to provide the system with detailed and diverse raw data.

[0149] The data processing module is used to perform time synchronization, data integration, noise filtering and feature preprocessing on the collected sitting posture data to generate preprocessed feature data; the data processing module is responsible for the fine processing of the original data. It ensures that the data from different sensors are aligned on the same time axis through time synchronization, eliminating the errors caused by time differences. Data integration integrates multiple data sources into one to form a unified sitting posture description. Noise filtering technology is used to remove the inevitable noise in the collection process to ensure the accuracy of subsequent analysis. Feature preprocessing standardizes and formats the data for further analysis by the feature extraction module.

[0150] The feature extraction and fusion module is used to extract the posture angle feature data, pressure distribution feature data and dynamic change feature data from the preprocessed feature data, and fuse and normalize the extracted feature data; the feature extraction and fusion module is the core part of data analysis. This module extracts posture angle features, pressure distribution features and dynamic change features from the preprocessed data. The posture angle feature is used to describe the specific posture of the user's sitting posture in three-dimensional space, such as the angle of leaning forward, leaning back or twisting. The pressure distribution feature depicts the pressure distribution map of the user sitting on the chair, which helps to identify the center of gravity position and posture stability. The dynamic change feature identifies the dynamic adjustment mode of the sitting posture by analyzing the time series data. The fusion and normalization processing integrates different types of feature data together, eliminates the differences between different dimensions, and provides standardized input for the adaptive neural network training module.

[0151] The adaptive neural network training module is used to construct and train the adaptive neural network model using the normalized feature data; the adaptive neural network training module constructs and trains the neural network model based on these normalized feature data. The model gradually optimizes the weights and structures through supervised learning to accurately identify the user's sitting posture pattern. The adaptability of the model is reflected in its ability to adjust according to the characteristics of different users to improve the accuracy of recognition and prediction.

[0152] The timing analysis and prediction module is used to perform multi-level timing analysis based on normalized feature data, generate short-term posture trend data, medium-term posture pattern data and long-term posture habit data, and generate future situation prediction data in combination with the adaptive neural network model, and optimize the prediction data through heterogeneous data cross-validation; the timing analysis and prediction module performs multi-level timing analysis based on feature data to generate short-term posture trend, medium-term posture pattern and long-term posture habit data. These data are combined with the neural network model to generate future situation prediction. Heterogeneous data cross-validation further optimizes the accuracy of the prediction data to ensure that the prediction results are consistent with the user's actual sitting posture dynamics.

[0153] The feedback generation and control module is used to design a multi-objective feedback strategy based on the optimized prediction data, generate feedback strategy data, and generate tactile feedback data, visual feedback data, and audio feedback data, and adjust the feedback strategy through real-time monitoring and dynamic control mechanisms; The feedback generation and control module designs a multi-objective feedback strategy based on the optimized prediction data and generates feedback strategy data. This module combines tactile feedback (such as chair vibration), visual feedback (such as display prompts), and audio feedback (such as voice reminders), and adjusts the feedback strategy through real-time monitoring and dynamic control mechanisms to ensure the timeliness and effectiveness of feedback.

[0154] The multimodal feedback mechanism module is used to implement the adjusted feedback strategy data, generate application feedback data, and record the user's response data; the multimodal feedback mechanism module implements the adjusted feedback strategy, generates application feedback data, and records the user's response. These feedback mechanisms work together to provide users with personalized sitting posture correction suggestions and improve the user's correction effect and satisfaction.

[0155] The feedback optimization and model adjustment module is used to optimize the feedback strategy and adjust the model parameters based on the feedback effect data and self-organizing map, and generate optimized model data. The feedback optimization and model adjustment module optimizes the strategy and adjusts the model parameters based on the feedback effect data and self-organizing map. Through continuous iteration, the optimized model data is generated to further improve the intelligence level and responsiveness of the system.

[0156] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware.

[0157] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. An intelligent sitting posture correction method, characterized in that: The following steps are involved: Using piezoelectric sensors, dynamic flexible materials, multi-view image recognition devices, infrared depth sensors and sound wave reflection equipment, the user's sitting posture data is collected, and time synchronization, data integration, noise filtering and feature preprocessing are performed on the sitting posture data to generate preprocessed feature data; Extracting posture angle feature data, pressure distribution feature data and dynamic change feature data from the preprocessed feature data, fusing and normalizing the extracted feature data, and using the normalized feature data to construct and complete the training of the adaptive neural network model; Based on the normalized feature data, multi-level time series analysis is performed to generate short-term posture trend data, medium-term posture pattern data and long-term posture habit data, and future situation prediction data is generated in combination with the adaptive neural network model, and the prediction data is optimized through heterogeneous data cross-validation; The multi-level time series analysis step includes: using a sliding window technique to analyze the normalized feature data in the short term to identify the rapidly changing posture features and generate short-term posture trend data; using a moving average method to perform a medium-term analysis on the normalized feature data to identify the periodic change features of the sitting posture and generate medium-term posture pattern data; using an exponential smoothing method to perform a long-term trend analysis on the normalized feature data to evaluate the long-term stability of the posture and generate long-term posture habit data; The training step of the adaptive neural network model uses the alternating least squares method to perform parameter optimization, which gradually adjusts the weight parameters of the model to minimize the prediction error, so that the model can be adaptively adjusted to adapt to the changes in the sitting posture data of different users; The heterogeneous data cross-validation step includes comparing the future situation prediction data with the data obtained through multi-view image recognition, infrared depth sensing and sound wave reflection, and improving the accuracy of the prediction data through consistency verification and error correction; Designing a multi-objective feedback strategy based on the optimized prediction data, generating feedback strategy data, generating and applying tactile feedback data, visual feedback data and audio feedback data, adjusting the feedback strategy through real-time monitoring and dynamic control mechanisms, and generating control optimization data; the multi-objective feedback strategy steps include determining the feedback type and intensity through weight adjustment according to the user's personalized preferences and correction goals, and the multi-objective feedback strategy optimizes the feedback method through dynamic learning and adaptive adjustment of user preferences to maximize user acceptance and comfort; The adjusted feedback strategy data is implemented, application feedback data is generated through a multimodal feedback mechanism, user response data is recorded, feedback strategy optimization and model parameter adjustment are performed based on feedback effect data and self-organizing mapping, and optimized model data is generated.

2. The intelligent sitting posture correction method according to claim 1, characterized in that: The time synchronization step includes assigning an accurate timestamp to the sitting posture data collected by each sensor, unifying all the data on a time axis through a timing alignment algorithm to ensure the consistency of data from different sensors at the same time point, and generating time synchronization data.

3. The intelligent sitting posture correction method according to claim 1, characterized in that: The calculation of the short-term posture trend data is based on the following steps: For each time window, the short-term posture trend is determined by calculating the difference between the sitting posture data in the window and its average value. The calculation formula is: in, It is short-term attitude trend data; is the size of the sliding window; is the current sitting posture data; is the average value of the sitting posture data within the sliding window.

4. The intelligent sitting posture correction method according to claim 1, characterized in that: The real-time monitoring and dynamic regulation mechanism includes: monitoring the user's posture changes and feedback effects in real time through sensors, and dynamically adjusting feedback parameters according to the collected real-time data to generate real-time regulation data to optimize the timeliness and effectiveness of feedback.

5. The intelligent sitting posture correction method according to claim 1, characterized in that: The feedback effect data is calculated by the following steps: In each feedback cycle, the relative error sum of squares between the user response data and the predicted response data is calculated using the following formula: in, To provide feedback effect data; is the number of response data points within the feedback cycle; is the feedback response value; is the predicted feedback value.

6. A system for implementing the intelligent sitting posture correction method according to any one of claims 1 to 5, characterized in that: include: A multimodal data acquisition module, including a piezoelectric sensor, a dynamic flexible material, a multi-view image recognition device, an infrared depth sensor, and an acoustic wave reflection device, for collecting the user's sitting posture data; A data processing module, used for performing time synchronization, data integration, noise filtering and feature preprocessing on the collected sitting posture data to generate preprocessed feature data; A feature extraction and fusion module, used to extract posture angle feature data, pressure distribution feature data and dynamic change feature data from the pre-processed feature data, and to fuse and normalize the extracted feature data; An adaptive neural network training module, used to construct and train an adaptive neural network model using normalized feature data; A time series analysis and prediction module is used to perform multi-level time series analysis based on normalized feature data, generate short-term posture trend data, medium-term posture pattern data, and long-term posture habit data, and generate future situation prediction data in combination with the adaptive neural network model, and optimize the prediction data through heterogeneous data cross-validation; Feedback generation and regulation module, which is used to design multi-objective feedback strategies based on optimization prediction data, generate feedback strategy data, and generate tactile feedback data, visual feedback data, and audio feedback data, and adjust the feedback strategy through real-time monitoring and dynamic regulation mechanism; A multimodal feedback mechanism module, used to implement the adjusted feedback strategy data, generate application feedback data, and record user response data; The feedback optimization and model adjustment module is used to optimize the feedback strategy and adjust the model parameters based on the feedback effect data and self-organizing map to generate optimized model data.

Images