A system and method for correcting pen-holding and sitting posture based on high-entropy alloy hydrogel self-powered sensing

Abstract

The application discloses a pen-holding and sitting posture collaborative correction system and method based on high-entropy alloy hydrogel self-powered sensing, and relies on HEA / MXene / PVA self-powered hydrogel to prepare a passive pen-holding sensor, and realizes automatic networking of all-scene equipment in combination with a distributed soft bus; single-vision 2D-3D posture analysis, ST-GCN+RNN double-feature fusion deep learning, MLP pen-holding classification multi-algorithm, multi-source fusion determination of pen-holding and six types of abnormal sitting postures are fused; one-stage acousto-optic alignment prompting, two-stage seat + environment self-adaptive hierarchical correction are adopted, and sitting health reminding and personalized correction suggestion generation are attached. The existing device needs an external power supply, single acquisition, device isolation and correction solidification, the pen-holding recognition accuracy is 95.98%, passive maintenance-free, distributed linkage, humanized hierarchical correction, and the application is widely used for student home and classroom writing posture correction.

Description

Technical Field

[0001] This invention relates to the fields of intelligent human posture correction, flexible self-powered sensing, distributed Internet of Things networking, and deep learning computer vision technology. In particular, it relates to a pen grip and sitting posture collaborative correction system and method based on high-entropy alloy hydrogel self-powered sensing, which is suitable for correcting pen grip errors and real-time monitoring and correction of prolonged sitting posture for primary and secondary school students. Background Technology

[0002] Currently, posture and pen grip correction products on the market fall into three categories, each with significant shortcomings: 1. Mechanical rigid orthodontic products (orthodontic chairs, fixed pen grips): physically restrict the human body, have poor comfort, and the bad posture will be restored as soon as the device is removed. They cannot be personalized to users of different heights and body types. 2. Wearable sensor wristwatches for posture correction: These rely on ultrasonic ranging to detect the distance between the head and chest, and determine posture based on only two points. The detection dimension is limited, and it cannot identify sitting postures such as hunchback and side leaning. The false judgment rate is high, and the device needs to be charged frequently, which limits its battery life. 3. Monocular vision posture correction device: It relies solely on image recognition, lacks auxiliary verification of seat pressure and human physiological parameters, is greatly affected by light and occlusion, has no ability to detect pen grip synchronously, and the devices are independent of each other and cannot be linked with external devices such as tables, chairs, lights, and air conditioners to adapt to the environment.

[0003] Most existing deep learning posture recognition solutions are based on single-feature image modeling, which easily loses temporal features. Pen grip recognition often uses external pressure sensors, which require external power supply and complicated wiring. Existing flexible hydrogel sensing solutions cannot simultaneously achieve self-powered generation, high sensitivity, water loss resistance and long-term use. They cannot simultaneously achieve passive pen grip acquisition and multi-dimensional synchronous monitoring of posture, and lack a distributed device linkage and hierarchical correction architecture.

[0004] In summary, existing technologies lack an integrated solution that combines self-powered pen grip sensing, multi-source data fusion for posture recognition, distributed full-device linkage, and two-level hierarchical correction. Summary of the Invention

[0005] This invention aims to overcome the technical shortcomings of existing orthodontic devices, such as the need for external power supply for sensors, limited data acquisition dimensions, isolated devices lacking collaboration, fixed correction modes, and low recognition accuracy. It provides a collaborative correction system and control method for pen-holding and sitting posture based on high-entropy alloy MXene composite hydrogel self-powered sensors, distributed soft bus networking, and multi-model fusion AI recognition. This system achieves passive data acquisition, intelligent multi-device linkage, dual-mode correction of graded prompts and passive adjustment, and automatic generation of personalized correction plans. Therefore, this invention adopts the following technical solution: Hardware networking solution: Relying on a distributed soft bus for automatic discovery and authentication networking, the networking hardware includes: a high-entropy alloy (HEA) hydrogel self-powered pen grip (passive TENG sensor), a smart physiological cushion, a monocular camera, a screen speaker, an electric smart seat, an air conditioner / desk lamp environment controller, and a touch interactive screen; the main control terminal manages all devices uniformly without the need for manual pairing.

[0006] Data acquisition scheme: The pen grip relies on the deformation and triboelectricity generated by the CoCrFeMnNi high-entropy alloy / MXene / PVA dual-network hydrogel to collect the five-finger grip deformation signal without external power supply; the seat cushion collects sitting pressure, heart rate, and respiration; a monocular camera collects human body video frames in real time, and the three types of data are uploaded to the main control for preprocessing simultaneously.

[0007] Identification scheme: Pen grip: The electrical features of pen grip are fed into a two-layer hidden layer MLP neural network to achieve accurate classification of 8 types of pen grip postures; Sitting posture: HRNet extracts skeletal key points, ST-GCN extracts spatiotemporal features of the skeleton, and RNN extracts temporal features of the skeletal angles. The dual-feature weighted fusion model is used to distinguish 6 types of abnormal sitting postures. The trunk deflection angle is calculated by converting monocular 2D-3D joint coordinates to assist in verification. The posture is determined by combining image, pressure and physiological multi-source data.

[0008] Two-level graded correction: Level 1 (Proactive Prompt): The screen displays a standard frame for alignment, and the error location is announced via voice, allowing the user to correct it independently; Level 2 (Passive Adaptation): Autonomously adjusts seat parameters, indoor temperature and humidity, and ambient light to guide users to adjust their sitting posture from an environmental perspective; Data maintenance: Automatic reminders for prolonged sitting (over 45 minutes), regular statistics on the frequency of poor posture, AI-generated personalized correction suggestions, and support for users to retrieve historical posture data as needed.

[0009] Specifically, this invention provides a pen grip and sitting posture collaborative correction system based on high-entropy alloy hydrogel self-powered sensing, comprising: a distributed master control terminal; a pen grip self-powered sensing sleeve made of HEA / MXene / PVA composite hydrogel, used to generate triboelectric signals when the fingers are squeezed and deformed; an intelligent seat cushion sensing module for collecting human sitting pressure, heart rate, and respiratory physiological data; a monocular vision acquisition device for acquiring continuous frame image data of the human body; and execution and feedback devices, including at least an audio-visual output device, an intelligent seat adjustment device, and an environmental control device. All of the above devices are communicatively connected to the distributed master control terminal. The distributed master control terminal is configured to: collect attribute parameters of the above-mentioned devices, create user profiles of static physical characteristics data (height, weight, shoulder width, arm length), filter target network devices based on user usage scenarios (writing / reading) needs, automatically discover and authenticate the network through a soft bus, establish communication links between the terminal and multiple target devices, and build an integrated pen grip and sitting posture correction system; preprocess and identify triboelectric signals, physiological signals, and image data based on deep learning algorithms to obtain a multi-dimensional feature dataset, and determine whether the user's pen grip posture and sitting posture meet the standards; generate graded correction instructions when an anomaly is detected, and send them to the execution and feedback devices.

[0010] Preferably, the self-powered pen grip sensing sleeve is based on a HEA-TENG tribological nanosensor made of CoCrFeMnNi high-entropy alloy microparticles doped with MXene / PVA dual-network conductive hydrogel. It requires no external power supply; the sensor generates an induced electrical signal through finger compression deformation, with a response time ≤50ms and a power density of 220.18mW·m³. -2 Furthermore, its output performance remains stable after 6000 cycles of compression and it can be thermo-remodeled and reused.

[0011] Preferably, when an anomaly is detected, the distributed master control terminal executes a graded correction instruction, including: a first-level correction mode: controlling the audio-visual output device to play a standard pen-holding and sitting posture reference frame, and simultaneously broadcasting voice prompts for the error location; a second-level correction mode: when the anomaly is not eliminated after the first-level correction mode has been running for a preset time, controlling the intelligent seat adjustment device to automatically adjust the seat height and tilt angle, and controlling the environmental control device to adjust the indoor temperature, humidity or ambient light.

[0012] Preferably, in step 4, the recognition process executed by the distributed master control terminal includes pen grip recognition and sitting posture recognition, simultaneously determining whether the pen grip posture and human sitting posture meet the standards. Sitting posture recognition involves using Faster-RCNN+HRNet to extract the spatiotemporal sequence of key skeletal points from the image data, and inputting physiological signals into a sitting posture recognition model fused with ST-GCN+RNN; pen grip recognition involves inputting triboelectric signals into an MLP pen grip classification model.

[0013] Preferably, the pen grip recognition includes: inputting triboelectric signal data into a dual-layer hidden layer MLP neural network to achieve accurate classification of 8 types of pen grip postures. The sitting posture recognition includes: HRNet extracting skeletal key points, ST-GCN extracting spatiotemporal features of the skeleton, and RNN extracting temporal features of the skeletal angles; weighted fusion of the two features to model and distinguish 6 types of abnormal sitting postures; supplemented by monocular 2D-3D joint coordinate conversion to calculate trunk deflection angle for auxiliary verification; and comprehensive determination of body posture based on image, pressure, and physiological multi-source data.

[0014] Preferably, the sitting posture recognition model is constructed based on the following method: Abnormal sitting postures are categorized into six types: hunchback, leaning back, leaning to the left, leaning to the right, crossing the left leg, and crossing the right leg. Videos of sitting postures in multiple scenarios are collected, and the frames are uniformly cut to 80 frames. Annotated samples are used to construct a dataset. Faster-RCNN is used to select human targets, and HRNet extracts 17 key points of the human skeleton, generating a spatiotemporal sequence of key points. Vectors of the spine and left and right lower legs are selected to calculate the frame-by-frame skeleton angles, constructing temporal features of the skeleton angles. A multi-scale spatiotemporal graph convolutional ST-GCN model is built to extract spatial-temporal features of the skeleton. An RNN recurrent network is built to extract local skeleton angle features. The dual features are weighted and fused with a weight coefficient of 0.55, and the sitting posture classification result is output via Softmax.

[0015] Preferably, it also includes a human-computer interaction terminal and a data operation and maintenance device. The data operation and maintenance device is used to periodically summarize abnormal data, generate and push targeted suggestions for improving pen grip and sitting posture based on the model, and receive user interaction commands sent by the human-computer interaction terminal to retrieve historical posture data as needed.

[0016] This invention provides a method for coordinated correction of pen grip and sitting posture based on high-entropy alloy hydrogel self-powered sensing, comprising the following steps: Step 1, a distributed master control terminal establishes a communication link with the pen grip self-powered sensing sleeve, the smart seat cushion sensing module, the monocular vision acquisition device, and the execution and feedback device through soft bus authentication; Step 2, the distributed master control terminal synchronously collects finger deformation triboelectric signals, human sitting posture pressure and heart rate and respiratory physiological data, and continuous frame image data of the human body; Step 3, the distributed master control terminal preprocesses and identifies the triboelectric signals, physiological signals, and image data based on deep learning algorithms to determine whether the user's pen grip posture and sitting posture meet the standards; Step 4, when an abnormal posture is determined, the first-level audio-visual prompt correction mode is activated first; if the abnormality is not eliminated after the first-level correction mode continues for a preset time, the second-level passive adaptation correction mode is activated to automatically adjust the smart seat parameters and environmental parameters.

[0017] The present invention provides an electronic device, including a memory and a processor. The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of the method described above.

[0018] The present invention provides a computer storage medium storing computer-executable instructions thereon, which, when executed by a processor, implement the steps of the method described above.

[0019] Technical effect Compared with existing technologies, the pen grip and sitting posture collaborative correction system and method based on high-entropy alloy hydrogel self-powered sensing provided by the present invention has the following technical effects: Self-powered advantages: The pen grip sensor uses HEA-MXene hydrogel triboelectric nanogenerator, which generates electricity automatically when the finger is pressed and deformed. It requires no battery or charging. The hydrogel has excellent water retention and can be used for a long time in a dry environment at room temperature. It can be reshaped and recycled repeatedly, reducing the cost of use. Distributed collaboration advantages: Based on automatic networking via a soft bus, tables, chairs, lights, cameras, and pen-holding devices are interconnected, breaking down device silos and achieving passive correction through environmental and seating linkage, distinguishing it from single-reminder products; Multimodal high-precision recognition: It integrates four-dimensional data of image skeleton time sequence, skeleton angle, seat cushion physiology, and pen grip electrical data, achieving a sitting posture recognition accuracy of 96.67% and a pen grip recognition accuracy of 95.98%, overcoming the defects of misjudgment due to single vision / single distance measurement. Layered and humanized correction: First, prompt the individual to make corrections independently; if ineffective, then passively adjust the environment and seat. This approach does not forcibly restrict the body and balances comfort and correction effectiveness. Additional health features: Sedentary timer reminder, real-time monitoring of user's heart rate and breathing changes, taking into account posture correction and eye and physical and mental health management. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the overall architecture of the pen grip and sitting posture collaborative correction system based on high-entropy alloy hydrogel self-powered sensing in an embodiment of the present invention.

[0021] Figure 2 This is a schematic diagram illustrating the structure and signal acquisition principle of the self-powered hydrogel sensor for pen gripping in an embodiment of the present invention. Figure 3 This is a structural diagram of the TENG hydrogel according to an embodiment of the present invention; Figure 4 This is a diagram of the deep learning model architecture for sitting posture dual-feature fusion in an embodiment of the present invention; Figure 5 The diagram shows the timing flowchart, neural network diagram, confusion matrix, and accuracy result diagram of the system hierarchical correction control in this embodiment of the invention. Detailed Implementation

[0022] To make the technical means, creative features, objectives and effects of the present invention easy to understand, the following embodiments, in conjunction with the accompanying drawings, specifically illustrate a pen grip and sitting posture collaborative correction system and method based on high-entropy alloy hydrogel self-powered sensing.

[0023] This embodiment provides a pen grip and sitting posture collaborative correction system based on a high-entropy alloy hydrogel self-powered sensor.

[0024] Figure 1 This is a schematic diagram of the overall architecture of the pen grip and sitting posture collaborative correction system based on high-entropy alloy hydrogel self-powered sensing in this embodiment.

[0025] like Figure 1 As shown, the pen grip and sitting posture collaborative correction system based on high-entropy alloy hydrogel self-powered sensing in this embodiment includes a distributed main control terminal 10, a pen grip self-powered sensing sleeve 20, an intelligent seat cushion sensing module 30, a monocular vision acquisition device 40, an execution and feedback device 50, and a data operation and maintenance device 60. Each device is communicatively connected to the distributed main control terminal 10.

[0026] Figure 2 This is a schematic diagram of the structure and signal acquisition principle of the self-powered hydrogel sensor for pen gripping in this embodiment.

[0027] Figure 3 This is a structural diagram of the TENG hydrogel in this embodiment.

[0028] like Figure 2 , Figure 3 As shown, the self-powered pen grip sensor sleeve 20 is made of HEA / MXene / PVA composite hydrogel and is used to generate triboelectric signals when the finger is squeezed and deformed. The self-powered pen grip sensor sleeve 20 is based on a HEA-TENG triboelectric nanosensor made of CoCrFeMnNi high-entropy alloy microparticle-doped MXene / PVA dual-network conductive hydrogel. It requires no external power supply; the induced electrical signal is generated by finger squeezing and deformation, with a response time ≤50ms and a power density of 220.18mW·m³. -2 Furthermore, the output performance remains stable after 6000 compression cycles.

[0029] In this embodiment, the hydrogel is prepared as follows: Cellulose suspension preparation: Weigh 9.8 g zinc chloride + 0.6 g calcium chloride hexahydrate, add 3.6 mL deionized water, stir at 75 ℃ for 1 h, slowly add 0.2 g cellulose powder, and continue stirring until completely dissolved to obtain a transparent cellulose suspension for later use. Borax crosslinking solution preparation: Dissolve 0.4 g borax decahydrate in 9.6 mL deionized water, stir magnetically at 85 ℃ until completely dissolved, and prepare the crosslinking agent solution.

[0030] The preparation method of pure PVA base liquid is as follows: 0.4 g PVA particles were added to 3.6 mL of deionized water and stirred at 85 ℃ for 6 h until completely dissolved; 4 g of cellulose suspension was added and stirred at 85 ℃ for 20 min; 3 mL of borax solution was added and stirred at a constant temperature until homogeneous; and the mixture was allowed to cool naturally to obtain the PVA / cellulose matrix hydrogel stock solution.

[0031] The preparation method of HEA / MXene / PVA composite hydrogel (optimal ratio) is as follows: The pretreated MXene stock solution was diluted to 10 mg / mL and added to the PVA base solution at 130 wt% of the matrix solution mass. The mixture was continuously stirred at 85 ℃ to achieve uniform dispersion. 6 wt% CoCrFeMnNi HEA micro powder was added to the mixture and stirred continuously to make the HEA particles uniformly embedded in the MXene conductive network. The mixture was allowed to stand naturally at room temperature to solidify, thus obtaining a dual-network HEA / MXene / PVA conductive hydrogel.

[0032] A 2 mm thick silicone mold was used to create a 0.5 cm × 1 cm rectangular groove. The composite hydrogel slurry was injected into the groove and allowed to cure naturally at room temperature. The size of a single sensor piece was 0.5 cm × 1 cm × 2 mm. Copper leads were attached to both sides of the gel as positive and negative electrodes and encapsulated in a silicone pen sleeve to create a self-powered HMP-TENG pen grip sensor.

[0033] Furthermore, the preparation method of the pen grip sensor sleeve in this embodiment is as follows: A double-network conductive hydrogel was prepared using a PVA+cellulose substrate, doped with 6wt% CoCrFeMnNi high-entropy alloy micron particles and MXene nanosheets, and cross-linked with borax. A 0.5cm×1cm rectangular sensor sheet was cut and embedded into a silicone pen grip groove to fabricate a pen-grip type HMP-TENG sensor. The 3cm×3cm gel device achieved a power density of 220.18mW·m³. -2 The deformation response is 50ms, and the performance does not degrade after 6000 cycles. The sensor leads are connected to the Arduino signal acquisition board and connected to the distributed soft bus via WiFi.

[0034] The intelligent seat cushion sensing module 30 is used to collect physiological data on human sitting posture pressure, heart rate, and respiration; the monocular vision acquisition device 40 is used to collect continuous frame image data of the human body. In this embodiment, the main control is an embedded development board; the intelligent seat cushion integrates pressure sensing and heart rate and respiration acquisition modules; a monocular USB camera, an LCD screen, and an active speaker are arranged on the desktop.

[0035] The execution and feedback device 50 includes an audio-visual output device, an intelligent seat adjustment device, an environmental control device, and a human-computer interaction terminal.

[0036] The data operation and maintenance equipment 60 is used to periodically summarize abnormal data, generate targeted suggestions for improving pen grip and sitting posture based on the model and push them, and receive user interaction commands sent by the human-computer interaction terminal to retrieve historical posture data as needed.

[0037] In this embodiment, the distributed master control terminal 10 is configured to: collect the attribute parameters of the aforementioned devices, automatically discover and authenticate the network via a soft bus, and establish a communication link between the terminal and multiple target devices. Furthermore, it preprocesses and identifies triboelectric signals, physiological signals, and image data based on deep learning algorithms to determine whether the user's pen-holding posture and sitting posture meet the standards.

[0038] In this embodiment, the recognition process executed by the distributed master control terminal 10 includes pen grip recognition and sitting posture recognition, and simultaneously determines whether the pen grip posture and human sitting posture meet the standards.

[0039] Figure 4 This is a diagram of the deep learning model architecture for sitting posture dual-feature fusion in this embodiment.

[0040] like Figure 4 As shown, in this embodiment, the sitting posture recognition uses Faster-RCNN+HRNet to extract the spatiotemporal sequence of key skeletal points from the image data, and the physiological signal input is fused with the ST-GCN+RNN sitting posture recognition model. Specifically, the recognition process includes: HRNet extracts key skeletal points, ST-GCN extracts spatiotemporal features of the skeleton, and RNN extracts temporal features of the skeletal angles. The dual features are weighted and fused to form a model, distinguishing 6 types of abnormal sitting postures. The model is further assisted by monocular 2D-3D joint coordinate conversion to calculate the trunk deflection angle for verification, and the posture is determined by combining image, pressure, and physiological multi-source data.

[0041] In this embodiment, the posture recognition model is constructed based on the following method: Abnormal sitting postures were categorized into six types: hunchback, leaning back, leaning to the left, leaning to the right, crossing the left leg, and crossing the right leg. Videos of sitting postures in multiple scenarios were collected, and the frames were uniformly cut to 80 frames. The samples were labeled to construct a dataset. Faster-RCNN selects human targets, HRNet extracts 17 human skeleton key points, generates a spatiotemporal sequence of key points, selects the spine and left and right lower leg vectors to calculate the frame-by-frame skeleton angles, and constructs the temporal features of skeleton angles. A multi-scale spatiotemporal graph convolutional ST-GCN model was constructed to extract spatial-temporal features of the skeleton. An RNN recurrent network was constructed to extract local skeletal angle features. The dual features were weighted and fused with a weight coefficient of 0.55, and the sitting posture classification results were output through Softmax.

[0042] The pen grip recognition process is as follows: triboelectric signals are input into the MLP pen grip classification model. Pen grip recognition includes: inputting triboelectric signal data into a two-layer hidden layer MLP neural network to achieve accurate classification of 8 types of pen grip postures. Figure 5 The diagram shows the timing flowchart, neural network diagram, confusion matrix, and accuracy result diagram of the system hierarchical correction control in this embodiment.

[0043] like Figure 5 As shown, when an anomaly is detected, the distributed master control terminal 10 generates a tiered correction instruction and sends it to the execution and feedback device 50. The correction instruction includes: Level 1 Correction Mode: Controls the audio-visual output device to play a standard pen-holding and sitting posture reference frame, and simultaneously broadcasts voice prompts for incorrect positions; Second-level correction mode: If the abnormality is not eliminated after the first-level correction mode has been running for a preset time (30s in this embodiment), the intelligent seat adjustment device is controlled to automatically adjust the seat height and tilt angle, and the environmental control device is controlled to adjust the indoor temperature, humidity or ambient light.

[0044] This embodiment also provides a method for coordinated correction of pen grip and sitting posture based on a high-entropy alloy hydrogel self-powered sensor, which works based on the above system and includes: In step S1, the distributed master control terminal 10 establishes a communication link with the pen-holding self-powered sensor sleeve 20, the smart cushion sensor module 30, the monocular vision acquisition device 40, the execution and feedback device 50, and the human-computer interaction terminal through soft bus authentication.

[0045] Step S2: The distributed master control terminal 10 synchronously collects finger deformation friction electrical signals, human sitting posture pressure and heart rate and respiratory physiological data, and continuous frame image data of the human body. The specific process is as follows: Step S2-1: Upon first use, the user's static data such as height, weight, shoulder width, and dominant hand are entered, and the system automatically generates the user information. In step S2-2, when the user writes, the pen grip sleeve bends and squeezes the hydrogel with the fingers, generating a differential voltage signal through friction. Different grip postures (wrist hook, pen hold, fingertip grip) correspond to unique electrical waveforms; the seat cushion collects the hip pressure distribution and heart rate in real time; the monocular camera captures upper body images at 25 frames per second.

[0046] Step S3: The distributed master control terminal 10 preprocesses and identifies triboelectric signals, physiological signals, and image data based on deep learning algorithms to determine whether the user's pen-holding posture and sitting posture meet the standards. The specific process is as follows: Step S3-1: The video is fed into Faster-RCNN to lock the human body region. HRNet outputs the coordinates of 17 human skeleton key points, splits the spine and left and right lower leg vectors, and calculates the angle between the skeleton and the coordinate axis frame by frame. ST-GCN extracts the spatiotemporal features of the key points, and RNN extracts the temporal features of the angle. The fusion coefficient α=0.55. After fusing the features, Softmax outputs the sitting posture classification (normal / hunchback / leaning back / leaning left and right / crossing legs). The pen-holding waveform is normalized and then fed into MLP (128 hidden layers + 64 hidden layers + 8 classification outputs) to complete the grip posture determination. Step S3-2, supplemented by monocular 2D joint → 3D torso conversion: extract the 2D coordinates of the four points of shoulder and hip from the monocular image, establish the torso spatial model, solve the camera pose using the PNP algorithm, use the Rodriguez transformation rotation matrix, and calculate the torso pitch and lateral tilt angles using Euler angles to supplement the basis for sitting posture determination.

[0047] Step S4: When an abnormal posture is detected, the first-level audio-visual prompt correction mode is activated first. If the abnormality is not eliminated after the first-level correction mode has been running for a preset duration, the second-level passive adaptation correction mode is activated to automatically adjust the smart seat parameters and environmental parameters. The specific process is as follows: Step S4-1 (Level 1 Audio-Visual Prompt): If an abnormal pen grip / sitting posture is detected, a standard human body / pen grip reference frame is overlaid on the screen in real time, and the speaker announces "Head too low, please raise your upper body; grip the pen with your wrist, please adjust your finger position"; real-time image replay, if 3 consecutive frames meet the standard, announce "Correct posture".

[0048] Step S4-2 (Secondary Passive Adjustment): If the Level 1 reminder is ineffective for 30 consecutive seconds, the main control sends an instruction: the smart seat adjusts the seat height and backrest tilt angle according to the user's static data; the air conditioner adjusts the temperature to 23~25℃ and the humidity to 45%; the desk lamp adjusts the desktop illuminance to 2000 lumens, optimizing the writing environment and guiding the user to correct their posture.

[0049] Step S4-3, Sedentary Life and Suggestion Output: After 45 minutes of sitting on the cushion, a pop-up window and voice reminder will prompt you to get up and move around; the frequency of poor posture is summarized daily, and AI analyzes the high-frequency error points to generate suggestions for "daily pen grip practice and sitting posture stretching", which will be displayed on the screen.

[0050] Step S5, Human-Computer Interaction Expansion: The user interacts with the voice command "View today's sitting posture data". The interaction module receives the command, the main control retrieves the entire day's abnormal records and physiological curves stored in the cloud, and displays the data visually on the screen, enabling personalized data queries. The above embodiments are preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention.

Claims

1. A pen grip and posture collaborative correction system based on high-entropy alloy hydrogel self-powered sensing, characterized in that, include: Distributed master control terminal; The pen grip self-powered sensor sleeve is made of high-entropy alloy / MXene / PVA composite hydrogel and is used to generate triboelectric signals when the finger is squeezed and deformed. The intelligent seat cushion sensing module is used to collect data on human sitting posture pressure, heart rate, and respiratory physiology. Monocular vision acquisition device, used to acquire continuous frame image data of the human body; The execution and feedback equipment includes at least sound and light output devices, intelligent seat adjustment devices, and environmental control devices. All of the above devices are communicatively connected to the distributed master control terminal. The distributed master control terminal is configured to: collect the attribute parameters of the above-mentioned devices, automatically discover and authenticate the network through the soft bus, and establish a communication link between the terminal and multiple target devices; preprocess and identify the triboelectric signal, the physiological signal and the image data based on the deep learning algorithm to determine whether the user's pen-holding posture and sitting posture meet the standards; generate graded correction instructions when an abnormality is determined, and send them to the execution and feedback device.

2. The pen grip and sitting posture collaborative correction system based on high-entropy alloy hydrogel self-powered sensing according to claim 1, characterized in that: The self-powered pen grip sensing sleeve is based on a high-entropy alloy-TENG tribological nanosensor made of CoCrFeMnNi high-entropy alloy microparticles doped with MXene / PVA dual-network conductive hydrogel. It requires no external power supply; the sensor generates an induced electrical signal through finger compression deformation, with a response time ≤50ms and a power density of 220.18mW·m⁻¹. -2 Furthermore, the output performance remains stable after 6000 compression cycles.

3. The pen grip and posture collaborative correction system based on high-entropy alloy hydrogel self-powered sensing according to claim 1, characterized in that: When an anomaly is detected, the distributed master control terminal executes a tiered correction instruction, including: Level 1 correction mode: Control the audio-visual output device to play a standard pen-holding and sitting posture reference frame, and simultaneously broadcast voice prompts for incorrect positions; Second-level correction mode: If the abnormality is not eliminated after the first-level correction mode has been running for a preset time, the intelligent seat adjustment device is controlled to automatically adjust the seat height and tilt angle, and the environmental control device is controlled to adjust the indoor temperature, humidity or ambient light.

4. The pen grip and sitting posture collaborative correction system based on high-entropy alloy hydrogel self-powered sensing according to claim 1, characterized in that: The recognition process performed by the distributed master control terminal includes pen grip recognition and posture recognition, simultaneously determining whether the pen grip posture and human sitting posture meet the standards. The posture recognition is as follows: Faster-RCNN+HRNet is used to extract the spatiotemporal sequence of key points of the image skeleton in the image data, and physiological signals are input into the posture recognition model fused with ST-GCN+RNN; The pen grip recognition is achieved by using a triboelectric signal input MLP pen grip classification model.

5. The pen grip and posture collaborative correction system based on high-entropy alloy hydrogel self-powered sensing according to claim 4, characterized in that: The pen grip recognition includes: inputting the triboelectric signal data into a dual-layer hidden-layer MLP neural network to achieve accurate classification of 8 types of pen grip postures. The sitting posture recognition includes: HRNet extracts skeletal key points, ST-GCN extracts spatiotemporal features of the skeleton, RNN extracts temporal features of the skeletal angles, dual-feature weighted fusion modeling, distinguishes 6 types of abnormal sitting postures, and is supplemented by monocular 2D-3D joint coordinate conversion to calculate trunk deflection angle for auxiliary verification, and comprehensively judges body posture by combining image + pressure + physiological multi-source data.

6. The pen grip and posture collaborative correction system based on high-entropy alloy hydrogel self-powered sensing according to claim 4, characterized in that: The posture recognition model is constructed based on the following method: Abnormal sitting postures were categorized into six types: hunchback, leaning back, leaning to the left, leaning to the right, crossing the left leg, and crossing the right leg. Videos of sitting postures in multiple scenarios were collected, and the frames were uniformly cut to 80 frames. The samples were labeled to construct a dataset. Faster-RCNN is used to select human targets, HRNet extracts 17 human skeleton key points, generates a spatiotemporal sequence of key points, selects the spine and left and right lower leg vectors to calculate the frame-by-frame skeleton angles, and constructs the temporal features of skeleton angles. A multi-scale spatiotemporal graph convolutional ST-GCN model was constructed to extract spatial-temporal features of the skeleton. An RNN recurrent network was constructed to extract local skeletal angle features. The dual features were weighted and fused with a weight coefficient of 0.55, and the sitting posture classification results were output through Softmax.

7. The pen grip and posture collaborative correction system based on high-entropy alloy hydrogel self-powered sensing according to claim 1, characterized in that, Also includes: Human-computer interaction terminal; The data operation and maintenance equipment periodically summarizes abnormal data, generates and pushes targeted suggestions for improving pen grip and sitting posture based on the model, and receives user interaction commands sent by the human-computer interaction terminal to retrieve historical posture data as needed.

8. A method for coordinated correction of pen grip and sitting posture based on self-powered sensing of high-entropy alloy hydrogel, characterized in that, Includes the following steps: Step 1: The distributed master terminal establishes a communication link with the pen-holding self-powered sensor sleeve, the smart cushion sensor module, the monocular vision acquisition device, the execution and feedback device, and the human-computer interaction terminal through soft bus authentication. Step 2: The distributed master control terminal synchronously collects finger deformation friction electrical signals, human sitting posture pressure and heart rate and respiratory physiological data, and human continuous frame image data. Step 3: The distributed main control terminal preprocesses and identifies the triboelectric signal, the physiological signal and the image data based on deep learning algorithms to determine whether the user's pen-holding posture and sitting posture meet the standards. Step 4: When an abnormal posture is detected, the first-level audio-visual prompt correction mode is activated first. If the anomaly is not eliminated after the first-level correction mode has been running for a preset duration, the second-level passive adaptation correction mode will be activated to automatically adjust the smart seat parameters and environmental parameters.

9. An electronic device comprising a memory and a processor, characterized in that: The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the method as described in claim 8.

10. A computer-readable storage medium having computer-executable instructions stored thereon, characterized in that: When the computer-executable instructions are executed by the processor, they implement the steps of the method as described in claim 8.

Images