A postpartum pelvic floor muscle function evaluation and intelligent rehabilitation training device, method and system based on a flexible pressure sensor array

Abstract

The application discloses a postpartum pelvic floor muscle function evaluation and intelligent rehabilitation training device based on a flexible pressure sensing array and belongs to the field of medical instruments.The device comprises a flexible cushion body, a pressure sensing array, a multi-mode sensing module, a main control module, a communication module and a power module; the sensing array is integrated with a micro-capacitance type pressure sensing unit in the middle layer of the cushion, the spatial resolution is less than or equal to 5 mm, and the sensing array is divided into four functional subareas of a perineal body area, bilateral ischial tuberosity areas, a coccyx / sacrum area and a medial thigh area according to the anatomical structure of the female pelvic floor; the multi-mode sensing module comprises surface electromyography electrodes, a six-axis inertial measurement unit and a temperature compensation unit; the main control module is internally provided with an edge computing unit, and executes pressure field reconstruction, multi-muscle group compensation discrimination, five-dimensional digital image construction and self-adaptive rehabilitation training scheme generation algorithms.The application realizes fine collection of pelvic floor muscle contraction pressure distribution and real-time discrimination of multi-muscle group compensation behavior through non-invasive high-density pressure field reconstruction.

Description

Technical Field

[0001] This invention belongs to the field of medical device and rehabilitation engineering technology, specifically relating to a postpartum pelvic floor muscle function assessment and intelligent rehabilitation training device, method and system based on a flexible pressure sensor array. Background Technology

[0002] Pelvic floor dysfunction (PFD) is a common condition among postpartum women. Epidemiological data shows that approximately 30%-50% of multiparous women experience varying degrees of pelvic floor muscle weakness, manifesting as stress urinary incontinence, pelvic organ prolapse, and sexual dysfunction. The golden window for pelvic floor muscle function recovery is 42 days to 6 months postpartum. Timely and accurate assessment and scientific rehabilitation training are crucial for preventing long-term pelvic floor dysfunction.

[0003] Currently, the mainstream technical solutions for pelvic floor muscle function assessment and rehabilitation training fall into two main categories: invasive and non-invasive. Invasive solutions use vaginal pressure probes or electromyography probes as the core sensing elements, in conjunction with biofeedback therapy devices. Although this type of solution has high measurement accuracy, it has the following inherent defects: (1) The invasive operation results in a high psychological threshold for users, and those whose wounds have not healed within 42 days postpartum cannot use it; (2) The probe can only provide single-point or linear multi-point pressure data within the cavity, and cannot reflect the spatial distribution characteristics of pelvic floor muscle contraction; (3) The usage rate in home settings is low, which seriously restricts the continuity and compliance of rehabilitation training.

[0004] Non-invasive solutions have received widespread attention in recent years. In the prior art, Chinese invention patent application CN116036476A discloses a pelvic floor muscle training system based on a pressure-sensitive cushion, which arranges three piezoelectric thin film sensors in the cushion to determine the pelvic floor muscle contraction state by detecting changes in sitting pressure. However, this solution has the following technical bottlenecks: (1) The number of sensors is too small (only 3), the spatial resolution is seriously insufficient, and it is impossible to construct the pressure distribution field of pelvic floor muscle contraction; (2) It is impossible to distinguish between the correct contraction of the pelvic floor muscles and the compensatory contraction of the abdominal muscles, gluteal muscles or adductor muscles of the thighs, while clinical studies have shown that about 70% of women cannot contract their pelvic floor muscles correctly, and the incidence of compensatory behavior is extremely high; (3) It lacks a multimodal data fusion mechanism, and single-point pressure measurement is easily affected by interference factors such as sitting posture adjustment and changes in weight distribution, resulting in a low signal-to-noise ratio.

[0005] Furthermore, existing rehabilitation training programs generally employ monotonous repetitive Kegel exercises, lacking gamified feedback mechanisms and adaptive difficulty adjustment, resulting in poor user compliance and ineffective training. Therefore, there is an urgent need in this field for an integrated solution for postpartum pelvic floor muscle function assessment and rehabilitation training that combines non-invasive comfort, high spatial resolution, multi-muscle group compensation identification capabilities, and adaptive gamified feedback. Summary of the Invention

[0006] To address the aforementioned shortcomings of existing technologies, this invention provides a postpartum pelvic floor muscle function assessment and intelligent rehabilitation training device based on a flexible pressure sensor array. It aims to achieve refined acquisition of pelvic floor muscle contraction pressure distribution through non-invasive high-density pressure field reconstruction technology, and to realize real-time identification of multi-muscle group compensatory behaviors through multimodal data fusion and deep learning algorithms. This allows for the construction of a digital profile of pelvic floor muscle function and the generation of an adaptive gamified rehabilitation training program, thereby solving the technical problems of poor compliance with invasive programs, insufficient resolution of existing non-invasive programs, and monotonous training.

[0007] To achieve the above-mentioned objectives, the present invention adopts the following technical solution: A postpartum pelvic floor muscle function assessment and intelligent rehabilitation training device based on a flexible pressure sensor array includes: a flexible cushion body, a high-density flexible pressure sensor array, a multimodal sensor module, a main control module, a communication module, and a power supply module.

[0008] The flexible seat cushion body adopts a three-layer composite structure. From top to bottom, it consists of: a medical-grade antibacterial silicone surface layer, which preferably has a micro-protrusion matrix on its surface to enhance skin contact stability, and is preferably detachable for easy cleaning and replacement; a flexible pressure sensor array middle layer, which is the core sensing layer of the entire device; and a high-resilience memory foam bottom layer, which provides uniform support and filters out high-frequency vibration noise. The flexible seat cushion body is preferably designed as a detachable cushion cover, which can be easily slipped onto chairs in postpartum care centers, dining chairs at home, or office chairs, enabling users to "test as soon as they sit down" without any discomfort, significantly reducing the barrier to use.

[0009] The high-density flexible pressure sensing array is integrated into the middle layer and is preferably manufactured using a fabric-integrated printed electronics process. The array comprises M×N micro-capacitive pressure sensing units, where M≥16 and N≥16, with a spatial resolution ≤5mm, capable of covering the entire ischial tuberosity-perineum-coccyx sensing area. Each sensing unit uses a PET or PI film as a flexible substrate, with printed silver paste as the electrode layer and a microstructured elastomer as the dielectric layer, ensuring a good linear relationship between capacitance change and pressure within the 0-200kPa pressure range. Compared to the low-resolution schemes of existing technologies that only deploy three sensors, the high-density array of this invention can capture subtle pressure distribution changes generated by pelvic floor muscle contraction, providing a high-fidelity raw data foundation for subsequent pressure field reconstruction and compensation identification.

[0010] Crucially, based on the anatomical structure of the female pelvic floor, this invention divides the high-density flexible pressure sensing array into four functional zones: Zone A is the perineal body zone, corresponding to the surface projection area of ​​the puborectalis and pubococcygeus muscles, which is the core area of ​​peak pelvic floor muscle contraction pressure; Zone B is the bilateral ischial tuberosity zone, used to detect sitting stability and compensatory tension of the gluteus maximus; Zone C is the coccyx / sacrum zone, used to detect the involvement of the posterior group of pelvic floor muscles (iliococcygeus muscle); and Zone D is the inner thigh zone, used to identify compensatory behavior of the adductor muscles. Through the above-mentioned zone mapping design, this invention achieves independent perception of each functional sub-region of the pelvic floor muscles in a non-invasive cushion, providing a physical basis for spatial resolution for the compensation identification algorithm.

[0011] The multimodal sensing module includes flexible surface electromyography (EMG) electrodes, a six-axis inertial measurement unit (IMU), and a temperature compensation unit. The flexible surface EMG electrodes are positioned on both sides of area A (corresponding to the 3 o'clock and 9 o'clock positions on the human body) to collect superficial perineal EMG signals, which are then cross-validated with pressure signals to address the core challenge of determining whether pressure changes truly originate from pelvic floor muscle contraction. The six-axis IMU is embedded in the front of the seat cushion to detect the forward tilt angle of the torso and the pelvic tilt posture, thereby identifying spurious pressure changes caused by posture adjustments. The temperature compensation unit, located at the edge of the array, uses an NTC thermistor to collect ambient temperature reference values ​​and eliminate the influence of temperature drift on the capacitive sensor during signal preprocessing.

[0012] The main control module includes a signal preprocessing unit, an edge computing unit, and a storage unit, serving as the data processing and decision-making center for the entire device. The edge computing unit is configured to execute four core algorithms: (a) Pressure field reconstruction algorithm: The original capacitance values ​​of the array are mapped to a two-dimensional pressure distribution map using a lookup table method or fitting formula, and the following spatiotemporal features are extracted: peak pressure coordinate P_max (to locate the center point of contraction pressure), pressure asymmetry index ASI (to assess bilateral symmetry), and pressure gradient field. P (reflects the ability of muscle groups to coordinate contraction), pressure-time integral (PTI) (assessing type I slow muscle endurance), and pressure rise rate dP / dt (assessing type II fast muscle explosive power).

[0013] (b) Multi-muscle group compensation discrimination algorithm: This is one of the core technical contributions of this invention. The algorithm is based on a spatiotemporal convolutional neural network (ST-CNN) to perform multimodal fusion analysis on the temporal sequence of pressure distribution (sampling rate ≥ 50Hz, window 2 seconds), IMU posture data, and surface electromyography time-frequency features. The ST-CNN includes a spatial branch (a two-dimensional convolutional neural network extracts the spatial pattern of single-frame pressure distribution, such as pressure center shift and multi-peak distribution), a temporal branch (LSTM or Transformer encoder performs temporal modeling of the dynamic pressure change process), and a fusion layer (a cross-attention mechanism fuses multimodal features). The network outputs five classification results: correct contraction, Valsalva maneuver (breath-holding), abdominal muscle compensation, gluteal muscle compensation, and thigh adductor muscle compensation. This algorithm achieves intelligent discrimination of pelvic floor muscle contraction quality in a non-invasive cushion, and its classification accuracy has been experimentally verified to reach over 90%, far exceeding the simple threshold judgment method of existing technologies.

[0014] (c) Pelvic Floor Muscle Function Digital Profile Construction Algorithm: Based on the spatiotemporal features output by the pressure field reconstruction algorithm, a digital profile is constructed comprising five dimensions: muscle strength level, muscle endurance, explosive power, symmetry, and coordination. Each dimension is quantified using a percentage system, and a comprehensive functional index (PFI) is further generated. The muscle strength level dimension uses the maximum voluntary contraction (MVC) pressure value as a benchmark against the modified Oxford Scale (MOS); the muscle endurance dimension uses the PTI decay curve of a 60-second sustained contraction as an evaluation indicator; the explosive power dimension uses the mean dP / dt and coefficient of variation of 5 rapid contractions as evaluation indicators; the symmetry dimension uses the pressure ratio of the left and right A zones and dynamic consistency as evaluation indicators; and the coordination dimension uses the contraction-relaxation transition time and the frequency of compensation as evaluation indicators. The PFI is a weighted average of the five dimension scores, with weights pre-set based on clinical evidence-based medicine data.

[0015] (d) Adaptive Gamified Rehabilitation Training Program Generation Algorithm: Training programs are automatically matched based on the predominant defect type in the digital profile. When Type I muscle endurance defects are predominant, a step-by-step continuous contraction program is generated, with a pressure target of 30%-50% of MVC and a duration increasing step-by-step from 5 seconds to 60 seconds. When Type II muscle explosive power defects are predominant, a rapid response training program is generated, triggering the user to reach the target pressure and quickly relax within 1 second using random visual cues. When the frequency of compensatory behaviors exceeds a preset threshold, an isolation training mode is first entered, using real-time voice guidance and visual feedback to inhibit the involvement of incorrect muscle groups. During the training execution phase, pressure signals are mapped to game control variables, such as contraction strength controlling the virtual object's ascent height, duration controlling flight distance, and pressure symmetry controlling left-right balance. The game difficulty is dynamically adjusted according to the user's real-time performance using a scoring update formula to ensure the user is always in the optimal learning zone.

[0016] The communication module supports Bluetooth Low Energy (BLE) and Wi-Fi dual-mode communication for transmitting digital profiles, training data, and device status information to mobile terminals or cloud servers. The mobile terminal (smartphone or tablet) is configured to receive and display the digital profile, execute a gamified training interface, and receive personalized training plan adjustment instructions pushed from the cloud. The cloud server is configured to store historical assessment data, execute remote training plan optimization algorithms, and push high-risk user warning information to nursing terminals, achieving remote closed-loop management of the entire postpartum care center-family cycle.

[0017] This invention also provides a method for assessing postpartum pelvic floor muscle function based on a flexible pressure sensor array. Using the aforementioned device, the method comprises six steps: multimodal data acquisition, signal preprocessing, pressure field reconstruction, multi-muscle group compensation identification, construction of a digital profile of pelvic floor muscle function, and result output. The method acquires data while the user performs a standard contraction sequence, which includes holding a maximum voluntary contraction for 5 seconds, five rapid contraction-relaxation cycles, and a 60-second sustained moderate-intensity contraction. The main control module guides the user through these steps via voice and issues real-time correction prompts when the frequency of compensatory behavior exceeds a preset threshold.

[0018] The present invention also provides a training control method for the above-mentioned device, comprising six steps: digital profile reading, defect-dominant type determination, training program generation, gamified execution, training quality assessment, and program iteration, forming a closed-loop rehabilitation system of "assessment-training-reassessment".

[0019] The present invention also provides a postpartum pelvic floor muscle full-cycle remote rehabilitation management system, including the device, mobile terminal, cloud server and nursing terminal, which realizes personalized training program iteration and high-risk user early warning through cloud data aggregation and swarm intelligence algorithm.

[0020] Compared with the prior art, the present invention has the following significant advantages: (1) Non-invasive comfort: The high-density flexible pressure sensor array in the form of a cushion completely avoids the psychological and physiological discomfort caused by invasive probes. It can be used immediately after childbirth, which significantly improves the user compliance in the home setting. (2) High spatial resolution: The sensor unit array of ≥16×16 combined with the spatial resolution of ≤5mm realizes the fine reconstruction of the pelvic floor muscle contraction pressure distribution field in a non-invasive device, breaking through the measurement blind zone of the existing 3-sensor scheme. (3) Multi-muscle group compensation identification: Through the ST-CNN multimodal fusion algorithm, high-precision classification and identification of correct contraction and four common compensation behaviors were achieved, solving the long-standing problem of unclear signal source in non-invasive schemes; (4) Digital profiling and adaptive training: Five-dimensional digital profiling provides a comprehensive assessment system that surpasses traditional muscle strength grading. The adaptive gamified training program is precisely matched according to the individual's defect type and suppresses compensation through real-time feedback, which significantly improves training efficiency. (5) Full-cycle remote management: Through cloud data aggregation and swarm intelligence algorithms, a smooth connection between centralized care in the postpartum care center and home-based self-training is achieved. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the overall structure of the postpartum pelvic floor muscle function assessment and intelligent rehabilitation training device based on a flexible pressure sensor array as described in this invention. Figure 2 for Figure 1 A schematic cross-sectional view of the three-layer composite structure of the flexible seat cushion body in the device shown. Figure 3 for Figure 1 The schematic diagram of the functional partition planar layout of the high-density flexible pressure sensing array in the device shown illustrates four functional partitions: A (perineal body region), B (bilateral ischial tuberosities region), C (coccyx / sacrum region), and D (inner thigh region). Figure 4 This is a system architecture block diagram of the postpartum pelvic floor muscle full-cycle remote rehabilitation management system described in this invention; Figure 5 This is a schematic diagram of the spatiotemporal convolutional neural network structure of the multi-muscle group compensation discrimination algorithm described in this invention; Figure 6 A radar diagram illustrating the five-dimensional digital profile of pelvic floor muscle function as described in this invention. Figure 7 This is a flowchart of the postpartum pelvic floor muscle function assessment method described in this invention; Figure 8 This is a flowchart of the training control method described in this invention.

[0022] Explanation of markings in the diagram: 100 Device; 110 Flexible seat cushion body; 111 Medical-grade antibacterial silicone surface layer; 111a Micro-bump matrix; 112 Flexible pressure sensor array middle layer; 113 High-resilience memory foam bottom layer; 114 U-shaped pressure relief notch; 120 High-density flexible pressure sensor array; 121 Area A (perineal body area); 122 Area B (bilateral ischial tuberosities area); 123 Area C (coccyx / sacrum area); 124 Area D (inner thigh area); 130 Multimodal sensing module; 131 Flexible surface electromyography electrode; 132 Six-axis inertial measurement unit; 133 Temperature compensation unit; 140 Main control module; 141 Signal preprocessing unit; 142 Edge computing unit; 143 Storage unit; 150 Communication module; 160 Power supply module; 200 Mobile terminal; 300 Cloud server; 400 Nursing terminal. Detailed Implementation

[0023] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. These embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention. Those skilled in the art can make various modifications and substitutions to the present invention without departing from the spirit and scope of the present invention, and all such modifications and substitutions should fall within the scope of protection of the present invention.

[0024] Example 1: Device Hardware Structure and Evaluation Method Reference Figures 1 to 4 This embodiment provides a postpartum pelvic floor muscle function assessment and intelligent rehabilitation training device 100 based on a flexible pressure sensor array. The device 100 includes a flexible cushion body 110, a high-density flexible pressure sensor array 120, a multimodal sensor module 130, a main control module 140, a communication module 150, and a power supply module 160.

[0025] The flexible seat cushion body 110 adopts a three-layer composite structure (see reference). Figure 2The medical-grade antibacterial silicone surface layer 111 is 3mm thick and features a micro-bump matrix 111a on its upper surface. The bumps are 1mm high and 5mm apart, arranged in a hexagonal, dense pattern to enhance the friction coefficient between the skin and the cushion, reducing false pressure signals caused by body slippage during contraction. The medical-grade antibacterial silicone surface layer 111 is detachable and connected to the flexible pressure sensor array middle layer 112 via a zipper surrounding the edge, facilitating regular disassembly for cleaning and disinfection. The flexible pressure sensor array middle layer 112 serves as the support layer for the high-density flexible pressure sensor array 120, approximately 2mm thick, and uses a PET film substrate. The high-resilience memory foam bottom layer 113 is 20mm thick high-resilience memory foam with a density of 50kg / m³, providing uniform support and absorbing high-frequency vibrations from the environment. The flexible seat cushion body 110 has overall dimensions of 450mm (length) × 420mm (width) × 25mm (thickness) and is designed as a detachable seat cover that can be fitted onto the seat of a standard office chair or a postpartum care center chair. The rear center of the flexible seat cushion body 110 has a U-shaped pressure-relief notch 114 corresponding to the coccyx area, used to release localized pressure in the coccyx region; the front edge of the flexible seat cushion body 110 has a wavy double groove corresponding to the thigh support areas on both sides.

[0026] The high-density flexible pressure sensing array 120 is integrated into the middle layer 112 of the flexible pressure sensing array. It adopts a 32×32 micro-capacitive pressure sensing unit array, totaling 1024 sensing units, with a spatial resolution of 4.5mm, covering a sensing area of ​​approximately 144mm×144mm, which is sufficient to cover the intertuberous distance (approximately 100-120mm) and perineal body area in women. Each sensing unit adopts a parallel plate capacitor structure: the upper electrode is printed silver paste, the lower electrode is printed silver paste, and the middle dielectric layer is a 200μm thick microstructured PDMS elastomer (with a pyramid-shaped microstructure on the surface). When subjected to pressure, the pyramid microstructure deforms, the electrode spacing decreases, and the capacitance value increases. Through pre-calibration, a capacitance-pressure mapping lookup table is established, ensuring that the pressure measurement error is ≤5% within the range of 0-200kPa. The array is connected to the main control module 140 via a flexible printed circuit (FPC), with a scanning frequency of 100Hz, satisfying the 50Hz biosignal sampling theorem.

[0027] Reference Figure 3The high-density flexible pressure sensor array 120 is divided into four functional zones based on the anatomical structure of the female pelvic floor. Zone A 121 is the perineal body zone, located at the geometric center of the array, covering an area of ​​approximately 30mm × 40mm. It corresponds to the surface projection area of ​​the puborectalis and pubococcygeus muscles and is the core area of ​​the pressure peak during pelvic floor muscle contraction. Zone B 122 is the bilateral ischial tuberosity zone, located on both sides of Zone A 121, each covering an area of ​​approximately 25mm × 30mm, used to detect sitting stability. When the user contracts the pelvic floor muscles, if the pressure in Zone B 122 increases abnormally, it suggests possible compensatory tension in the gluteus maximus. Area C123, located behind Area A121, is the coccyx / sacrum region, covering approximately 30mm x 35mm. It is used to detect the involvement of the posterior pelvic floor muscles (iliococcygeus). When the user performs the Valsalva maneuver (breath-holding and pressing down), the pressure in Area C123 increases significantly while the pressure in Area A121 is insufficient, which is a typical spatial characteristic of the Valsalva maneuver. Area D124, located on the inner thigh, is on the left and right sides below the front of the array, each covering approximately 20mm x 25mm. It is used to identify adductor muscle compensation. When the pressure in Area D124 increases synchronously during contraction while the pressure in Area A121 is insufficient, it indicates that the user is incorrectly using the adductors of the thigh instead of the pelvic floor muscles.

[0028] The multimodal sensing module 130 includes a flexible surface electromyography electrode 131, a six-axis inertial measurement unit 132, and a temperature compensation unit 133 (arrangement reference). Figure 1 The flexible surface electromyography (EMG) electrodes 131 consist of two dry electrodes, positioned on the left and right sides of area A 121 (corresponding to the 3 o'clock and 9 o'clock positions on the human body). The electrodes are 10 mm in diameter, spaced 40 mm apart, and made of conductive polymer material with a skin contact impedance ≤50 kΩ. The flexible surface EMG electrodes 131 are used to collect superficial perineal EMG signals at a sampling frequency of 1000 Hz. After bandpass filtering (10-500 Hz) and rectification smoothing, the root mean square (RMS) value is extracted as the EMG intensity index. The six-axis inertial measurement unit 132 (model MPU-6050) is embedded at the center of the front end of the seat cushion to detect the tilt angle of the seat cushion relative to the horizontal plane. When the tilt angle exceeds ±5°, the main control module 140 issues a posture correction prompt and removes abnormal posture data segments during the data analysis phase. The temperature compensation unit 133 consists of two NTC thermistors (10kΩ, B value 3950), which are respectively arranged in the upper left and lower right corners of the high-density flexible pressure sensor array 120 to collect ambient temperature. The main control module 140 performs real-time temperature compensation on the capacitance value of each channel according to the temperature-capacitance drift calibration curve, and the compensation accuracy ensures that the pressure measurement error caused by temperature drift is ≤2%.

[0029] The main control module 140 uses an STM32H743 microcontroller (480MHz clock speed, built-in FPU and DSP instruction set), integrating a signal preprocessing unit 141, an edge computing unit 142, and a storage unit 143. The signal preprocessing unit 141 performs multiplexing scanning, denoising filtering (moving average + wavelet threshold denoising), and temperature drift compensation on 1024-channel capacitor signals. The edge computing unit 142 runs the four core algorithms. The storage unit 143 is a 32GB eMMC cache used to buffer local evaluation data. The communication module 150 uses an nRF52840 chip, supporting BLE 5.0 and 2.4GHz Wi-Fi dual-mode communication. The power module 160 is a 5000mAh lithium polymer battery, supporting USB-C charging, and can operate continuously for 8 hours on a full charge.

[0030] Reference Figure 7 This embodiment also provides a method for assessing postpartum pelvic floor muscle function using the above-described device 100, the specific steps of which are as follows: Step S1, Multimodal Data Acquisition: The user sits upright on the device 100, maintaining a relaxed state. The main control module 140 guides the user to perform a standard contraction sequence via voice: (a) Maximum voluntary contraction and hold for 5 seconds – requiring the user to contract and hold the pelvic floor muscles with maximum effort; (b) Rapid contraction-relaxation cycle 5 times – requiring the user to complete a maximum contraction within 1 second after hearing the prompt and immediately relax; (c) 60 seconds of continuous moderate-intensity contraction – requiring the user to continuously contract with approximately 50% of maximum effort. During this process, the high-density flexible pressure sensor array 120 acquires pressure distribution time-series data at a sampling frequency of 100Hz, the flexible surface electromyography electrode 131 simultaneously acquires electromyographic signals, the six-axis inertial measurement unit 132 simultaneously acquires posture data, and the temperature compensation unit 133 simultaneously acquires ambient temperature.

[0031] Step S2, Signal Preprocessing: The signal preprocessing unit 141 performs moving average filtering (window length 5 points) and wavelet threshold denoising (db4 wavelet, 3-level decomposition, soft thresholding) on ​​the pressure data to eliminate high-frequency electronic noise; based on the temperature data collected by the temperature compensation unit 133, temperature compensation is performed on the capacitance value of each channel through a pre-calibrated temperature-capacitance drift lookup table; the electromyography signal is subjected to 10-500Hz bandpass filtering, full-wave rectification, and 100ms moving average smoothing; the IMU attitude data is subjected to Kalman filtering, and the accelerometer and gyroscope data are fused to output stable pitch and roll angles.

[0032] Step S3, Pressure Field Reconstruction: The preprocessed 1024 capacitance values ​​are mapped to a two-dimensional pressure distribution map using a lookup table, with a spatial resolution of 4.5mm. The following spatiotemporal features are extracted: Peak pressure coordinate P_max—the pixel coordinates corresponding to the maximum pressure value in the pressure distribution map; Pressure asymmetry index ASI=|P_left-P_right| / (P_left+P_right), where P_left and P_right are the average pressures of the left and right halves of region A121, respectively; Pressure gradient field. P – The spatial gradient of the pressure distribution map is calculated using the Sobel operator, reflecting the rate at which pressure diffuses outward from the center; Pressure-Time Integral (PTI) – The pressure curve during the 60-second sustained contraction phase is integrated over time; Pressure Rise Rate (dP / dt) – The rate at which pressure rises from the baseline to the peak value is calculated for each of the five contractions during the rapid contraction phase, and the mean and coefficient of variation are taken.

[0033] Step S4, Multi-muscle Group Compensation Identification: Refer to Figure 5 The spatiotemporal features extracted in step S3, the time-frequency features of electromyography (EMG) processed in step S2 (converted to a time-frequency map via short-time Fourier transform), and the IMU pose features are input into a spatiotemporal convolutional neural network (ST-CNN). The network structure of the ST-CNN is as follows: the spatial branch uses a 3-layer 2D-CNN (3×3 convolutional kernels, 32, 64, and 128 channels respectively, followed by BatchNorm and ReLU activation after each layer), the input is a 32×32×T pressure distribution time series map (T=100, corresponding to a 2-second window), and the output is a 128-dimensional spatial feature vector; the temporal branch uses a 2-layer BiLSTM (128 hidden units), the input is a spatial feature sequence, and the output is a 256-dimensional temporal feature vector; the EMG branch uses a 1D-CNN (1×5 convolutional kernels, 16 and 32 channels) to extract EMG time-frequency features and outputs a 64-dimensional EMG feature vector; the pose branch uses a fully connected layer (inputting 6-dimensional IMU data and outputting a 32-dimensional pose feature vector). The fusion layer concatenates the four feature vectors into a 480-dimensional vector, and adaptively weights it using a cross-attention mechanism (Query from the spatial branch, Key / Value from the temporal branch). Finally, it passes through a fully connected layer (480→128→5) and Softmax to output five-class probabilities. Classification results include: correct contraction (probability threshold > 0.8), Valsalva maneuver, abdominal muscle compensation, gluteal muscle compensation, and adductor muscle compensation. When the classification result is a compensatory behavior, the compensation type is recorded and the corresponding data segment is labeled. In subsequent digital profile construction, the weight coefficient of this data segment is reduced (e.g., multiplied by 0.3) to eliminate interference from spurious data.

[0034] Step S5, Construction of a Digital Profile of Pelvic Floor Muscle Function: Based on the spatiotemporal features extracted in Step S3, calculate percentage scores for five dimensions: Muscle Strength Grade Score = 100 × (MVC stress value / mean MVC value for healthy women of the same age), upper limit 100; Muscle Endurance Score = 100 × (PTI_measured / PTI_reference), where the PTI reference value is the median PTI of healthy women of the same age for 60 seconds of continuous contraction; Explosive Power Score = 100 × [1 - (dP / dt coefficient of variation / 0.3)], the smaller the coefficient of variation, the higher the score; Symmetry Score = 100 × (1 - ASI), the closer the ASI is to 0, the higher the score; Coordination Score = 100 × [1 - (contraction-relaxation transition time / 2 seconds)] × [1 - (compensation frequency)]. Comprehensive Functional Index (PFI) = 0.25 × Muscle Strength Grade + 0.25 × Muscle Endurance + 0.20 × Explosive Power + 0.15 × Symmetry + 0.15 × Coordination. (Reference) Figure 6 The five-dimensional score is visualized on the mobile terminal 200 in the form of a radar chart.

[0035] Step S6, Result Output: The digital profile, compensation identification results, and action segmentation markers are transmitted to the mobile terminal 200 via BLE and simultaneously uploaded to the cloud server 300 in structured JSON format. When PFI < 40, a high-risk classification prompt is output, and the cloud server 300 pushes a red warning message to the nursing terminal 400; when 40 ≤ PFI < 70, a medium-risk classification prompt is output, and a yellow prompt message and reinforcement training suggestions are pushed; when PFI ≥ 70, a low-risk classification prompt is output, and a green maintenance training suggestion is pushed. The above risk classification prompts are for medical staff reference, and the final diagnosis is made by medical staff.

[0036] Example 2: Training Control Method for the Device Reference Figure 8 This embodiment provides a training control method for the device 100 described in Embodiment 1, and the specific steps are as follows: Step T1, Digital Profile Reading: The main control module 140 reads the user's latest pelvic floor muscle function digital profile from the storage unit 143, including the five-dimensional score and the comprehensive functional index (PFI). Assume a user's digital profile is: muscle strength 55, muscle endurance 42, explosive power 68, symmetry 70, coordination 60, PFI = 55.

[0037] Step T2, Determination of Deficiency Dominance Type: Analyze the five-dimensional score. The lowest score is muscle endurance (42 points), which is determined to be a Type I muscle deficiency dominant type. If the lowest score is explosive power, it is determined to be a Type II muscle deficiency dominant type; if the lowest score is symmetry or coordination, it is determined to be a control deficiency dominant type.

[0038] Step T3, Training Program Generation: Based on the predominant defect type, an initial training program is matched from the preset training program library. For Type I muscle defect predominant type, a "stepped continuous contraction program" is matched. Initial parameters are: target pressure = 35% of the measured MVC (assuming the measured MVC corresponds to a pressure value of 40 kPa, then the target pressure is 14 kPa); the duration of the first level is 5 seconds, increasing by 3 seconds for each level passed, up to a maximum of 60 seconds; the total number of levels is 10, and the passing standard for each level is: the pressure is maintained within ±20% of the target pressure during continuous contraction, and the proportion of contractions judged as correct by the compensation discrimination algorithm is ≥80%.

[0039] Step T4, Gamified Execution: The game interface on the mobile terminal 200 displays a virtual hot air balloon. The user controls the balloon's ascent by contracting their pelvic floor muscles: the contraction force maps to the balloon's vertical ascent speed, the duration maps to the balloon's horizontal flight distance, and the pressure symmetry between the left and right A-zones maps to the balloon's left-right balance. The game background is a canyon scene, and the hot air balloon needs to pass through a series of circular archways of varying heights. The height of each archway corresponds to the target pressure range, and the width corresponds to the allowable pressure fluctuation range. The main control module 140 dynamically adjusts the game difficulty based on the user's real-time performance: if the user passes through the doorway perfectly three times in a row (pressure falls within the target range and compensation ratio <10%), the height of the next doorway is reduced by 5% (difficulty increased); if the user fails to pass through twice in a row, the height of the next doorway is increased by 10% (difficulty decreased), and voice guidance is inserted: "Please relax your abdomen and feel the upward lifting force of your pelvic floor muscles." The difficulty adjustment is executed according to the score update formula, where the user performance score R_new=R_old+K×(SE), where S is the actual performance (pass = 1, fail = 0), E is the expected performance probability (estimated based on the current difficulty and the user's historical data), and K=20 is the learning rate coefficient.

[0040] Step T5, Training Quality Assessment: During training, the edge computing unit 142 executes the compensation identification algorithm in real-time with a 2-second window. The frequency of compensation behavior in a single training session is statistically analyzed. For example, in a training session, if 2 out of 10 contractions are identified as abdominal muscle compensation, the compensation frequency is 20%, exceeding the preset threshold of 15%. At this point, the main control module 140 reduces the target pressure of the current level by 10% and outputs an isolation guidance prompt: the game interface switches to simplified mode, displaying only the pressure histogram of area A. The user needs to increase the pressure of area A to the target value under voice guidance, while keeping the pressure of area B below the baseline level. After the user successfully completes the isolation guidance 3 times consecutively, the full game mode is restored. When the Valsalva maneuver is detected, a red flashing and vibration prompts the user to stop the breath-holding and pressing behavior; when abdominal muscle compensation is detected, a voice command guides the user to relax the abdomen and refocus on the pelvic floor muscle lifting sensation.

[0041] Step T6, Solution Iteration: After the user completes a training cycle (7 days, including 5 training days and 2 rest days; the duration of a single training session on each training day is 10-20 minutes, including 3-5 gamified training levels), the evaluation method of Example 1 is re-executed to update the digital profile. Assuming that the muscle endurance score improves to 58 points after the update, the main control module 140 automatically adjusts the single session duration, number of levels, and target pressure parameters of the next training cycle based on the user's completion rate, compensation improvement rate, and improvement magnitude of the digital profile in the previous training cycle. For example, the step-by-step acceleration rate is adjusted from +3 seconds per level to +5 seconds per level, and the target pressure is increased from 35% to 40% of MVC, forming a closed-loop optimization of "evaluation-training-re-evaluation".

[0042] Example 3: Remote Closed-Loop Management System Reference Figure 4 This embodiment provides a remote closed-loop management system based on the device 100 of Embodiment 1, which is suitable for batch management of the pelvic floor muscle rehabilitation process of postpartum care centers.

[0043] In the postpartum care center setting, each new mother undergoes her first pelvic floor muscle function assessment on the third day postpartum to establish an individualized baseline. Nursing staff place 100 sets of devices on the chairs in the mothers' rooms; the assessment process is automatically triggered when the mother sits down. After the assessment, the digital profile and PFI (Pelvic Floor Function Index) are automatically uploaded to cloud server 300. Cloud server 300 is equipped with a nursing management backend, displaying the PFI distribution of all resident mothers in tabular form: red rows represent high-risk users with PFI < 40, yellow rows represent medium-risk users with PFI 40 ≤ PFI < 70, and green rows represent low-risk users with PFI ≥ 70. The nursing supervisor can prioritize assigning postpartum rehabilitation specialists to provide one-on-one rehabilitation guidance to users in the red rows. After leaving the center, users continue to use the home version of device 100 for home training, with data seamlessly integrated between the postpartum care center and home lifecycle via cloud server 300.

[0044] The cloud server 300 executes the following swarm intelligence optimization algorithms: (1) Cluster analysis: The K-means algorithm is used to group users with similar initial digital profiles. For example, users with "Type I muscle defects predominate and initial PFI 40-60" are grouped into one category; (2) Effect comparison: The improvement effect of different training scheme parameter configurations in this user group is compared, with the PFI improvement as the optimization target; (3) Parameter push: The reinforcement learning algorithm is used to automatically explore the optimal training parameter space and push the best-performing parameter configuration to the devices of this user group to realize the scheme iteration driven by swarm intelligence. For example, if the data shows that "step-by-step increasing rate +5 seconds / off" improves the average PFI by 12 points in the Type I muscle defect group, while "+3 seconds / off" only improves by 8 points, then +5 seconds / off is automatically used as the recommended configuration for this group.

[0045] In addition, the cloud server 300 also performs anomaly monitoring: when a user fails to upload training data for 7 consecutive days, it automatically pushes a reminder message to the user's mobile terminal 200; when a user's PFI shows a downward trend in 3 consecutive assessments, it pushes a reminder to the nursing terminal 400, suggesting a manual follow-up to investigate the cause (such as incorrect user training posture, equipment failure, or the emergence of new health problems).

[0046] Example 4: Training and Deployment of Multi-Muscle Group Compensation Identification Algorithm This embodiment details the training and edge deployment process of the spatiotemporal convolutional neural network (ST-CNN) described in step S4 of embodiment 1.

[0047] Dataset Construction: Two hundred postpartum female volunteers, aged 25-40, with a postpartum period of 6 weeks to 6 months, were recruited. Each volunteer, after signing an informed consent form, simultaneously underwent data collection using the device 100 of this invention, an invasive vaginal pressure probe, and transperineal ultrasound imaging. The invasive probe data served as the gold standard for pressure, while the pelvic floor muscle contraction and displacement data acquired through ultrasound imaging served as the anatomical gold standard. Three pelvic floor rehabilitation experts with over 5 years of experience independently reviewed the ultrasound images and pressure curves, assigning a five-category label to each contraction (correct contraction / Valsalva / abdominal muscle compensation / gluteal muscle compensation / adductor muscle compensation), with the majority vote used as the label. The final dataset contained 12,000 valid contraction samples, including 4,200 correct contractions, 2,100 Valsalva contractions, 2,800 abdominal muscle compensations, 1,600 gluteal muscle compensations, and 1,300 adductor muscle compensations. The dataset was divided into training, validation, and test sets in an 8:1:1 ratio.

[0048] Model Training: End-to-end training was performed on a GPU server using the cross-entropy loss function. The optimizer was AdamW, with an initial learning rate of 1e-4, a batch size of 32, and 200 epochs of training. A cosine annealing learning rate scheduling strategy was employed. To prevent overfitting, Dropout (rate=0.3) and L2 regularization (weight decay=1e-5) were introduced. The F1-macro metric was monitored on the validation set, and early stopping was triggered when there was no improvement for 10 consecutive epochs. After training, the five-class classification accuracy on the test set was 92.3%, and the F1-macro score was 0.91. Specifically, the correct contraction recall rate was 94.5%, the Valsalva maneuver recognition rate was 89.2%, the abdominal muscle compensation recognition rate was 91.7%, the gluteal muscle compensation recognition rate was 90.4%, and the adductor muscle compensation recognition rate was 88.6%.

[0049] Edge Deployment: Due to the limited computing power and memory of the STM32H743, the trained ST-CNN model needs to be compressed into a lightweight version. A knowledge distillation strategy is employed: using the trained large model as the teacher model, a student model is designed—the spatial branch is compressed into a 2-layer inverted residual module, the temporal branch is compressed into a 1-layer BiLSTM (64 hidden units), and the fusion layer uses simple feature concatenation + fully connected layers. The number of parameters in the student model is compressed from 4.2M in the teacher model to 280K, and the test set accuracy decreases by only 2.1% (to 90.2%), which is still within an acceptable range. The student model is quantized to INT8 using a model quantization tool, resulting in a final model size of 320KB, which can be fully loaded into the 1MB Flash memory of the STM32H743. During inference, a single forward propagation takes approximately 85ms, meeting the real-time requirements (<100ms).

[0050] Variation Example Based on the above embodiments, those skilled in the art can make the following modifications, all of which should fall within the protection scope of this invention: (1) The number of sensing units of the high-density flexible pressure sensing array 120 can be adjusted to 16×16, 24×24 or 64×64, and the spatial resolution can be adjusted to 5mm, 6mm or 3mm respectively, as long as the basic requirements of covering the complete sensing area and the spatial resolution is sufficient to distinguish the four functional zones are met. (2) The microcapacitive pressure sensing unit can be replaced with a piezoresistive, piezoelectric or triboelectric pressure sensing unit, and the corresponding signal conditioning circuit (charge amplifier, instrumentation amplifier, etc.) can be adapted accordingly. (3) The ST-CNN in the multi-muscle group compensation discrimination algorithm can be replaced by a pure Transformer architecture (such as TimeSeries Transformer) or a graph neural network (GNN, which treats the sensing unit as a graph node), as long as it can achieve multimodal feature fusion and five-class classification output; (4) The virtual objects in the gamified training can be replaced by any virtual image such as a hot air balloon, a bird, or a submarine. The control mapping relationship can be adjusted to pressure control horizontal displacement, symmetry control rotation angle, etc., as long as the core mechanism of pressure signal → game control variable → real-time visual feedback is maintained. (5) The device 100 can be expanded into a full-body cushion to increase the pressure sensing area of ​​the waist and back, and is used to simultaneously assess the synergistic function of the pelvic floor muscles and core muscles. (6) The swarm intelligence optimization algorithm of the cloud server 300 can use reinforcement learning algorithms such as PPO or SAC to replace K-means+ comparative analysis, so as to maximize the improvement of the average PFI of the group as the reward function and automatically explore the optimal training parameter space.

Claims

1. A postpartum pelvic floor muscle function assessment and intelligent rehabilitation training device based on a flexible pressure sensor array, characterized in that, include: The flexible seat cushion body adopts a three-layer composite structure, which consists of a medical-grade antibacterial silicone surface layer, a flexible pressure sensor array middle layer, and a high-resilience memory foam bottom layer from top to bottom. A high-density flexible pressure sensing array is integrated into the middle layer of the flexible pressure sensing array, which contains M×N flexible pressure sensing units. The high-density flexible pressure sensing array is divided into four functional zones based on the anatomical structure of the female pelvic floor: Zone A is the perineal body zone, corresponding to the surface projection area of ​​the puborectalis and pubococcygeus muscles; Zone B is the bilateral ischial tuberosity zone, used to detect sitting stability and compensatory tension of the gluteus maximus; Zone C is the coccyx / sacrum zone, used to detect the involvement of the posterior pelvic floor muscles. Zone D is the inner thigh area, used to identify compensatory behavior of the adductor muscles. The multimodal sensing module includes: flexible surface electromyography electrodes arranged on both sides of the A area for acquiring superficial perineal electromyography signals; a six-axis inertial measurement unit embedded in the flexible cushion body for detecting pelvic tilt posture; and a temperature compensation unit arranged at the edge of the high-density flexible pressure sensing array for eliminating the influence of ambient temperature drift on pressure measurement. The main control module includes a signal preprocessing unit, an edge computing unit, and a storage unit. The edge computing unit is configured to execute: (a) a pressure field reconstruction algorithm, used to map the raw signals collected by the high-density flexible pressure sensing array into a two-dimensional pressure distribution map and extract spatiotemporal features; (b) a multi-muscle group compensation discrimination algorithm, used to perform multimodal fusion analysis on the temporal sequence of the two-dimensional pressure distribution map, the posture data collected by the six-axis inertial measurement unit, and the electromyographic signals collected by the flexible surface electromyographic electrodes, and output a contraction mode classification result, the classification result including at least correct contraction and compensation behavior; (c) a pelvic floor muscle function digital profile construction algorithm, used to construct a digital profile containing five dimensions including muscle strength level, muscle endurance, explosive power, symmetry, and coordination based on the spatiotemporal features; and (d) an adaptive gamified rehabilitation training program generation algorithm, used to match training programs according to the defect-dominant type of the digital profile and map the pressure signal into game control variables. A communication module is used to transmit the digital profile and training data externally; and A power module is used to supply power to the device.

2. The apparatus according to claim 1, characterized in that, The high-density flexible pressure sensing array comprises M×N microcapacitive pressure sensing units, where M≥16 and N≥16, with a spatial resolution ≤5mm, covering the complete ischial tuberosity-perineum-coccyx sensing area. The microcapacitive pressure sensing units employ fabric-integrated printed electronics technology, using PET or PI film as a flexible substrate, printed silver paste as the electrode layer, and a microstructured elastomer as the dielectric layer, ensuring that the capacitance change of the high-density flexible pressure sensing array is linearly related to the pressure within the 0-200kPa pressure range.

3. The apparatus according to claim 1, characterized in that, The flexible surface electromyography electrodes are arranged on the left and right sides of the A area, corresponding to the 3 o'clock and 9 o'clock positions of the human body; the six-axis inertial measurement unit is embedded in the front center of the flexible cushion body; the temperature compensation unit is an NTC thermistor, arranged diagonally on the high-density flexible pressure sensing array, used to collect the ambient temperature reference value and perform temperature drift compensation on the pressure signal during the signal preprocessing stage.

4. The apparatus according to claim 1, characterized in that, The multi-muscle group compensation discrimination algorithm is implemented based on a spatiotemporal convolutional neural network, which includes a spatial branch, a temporal branch, and a fusion layer. The spatial branch uses a two-dimensional convolutional neural network to extract spatial pattern features of the single-frame pressure distribution map. The temporal branch uses a long short-term memory network or a Transformer encoder to perform temporal modeling of the dynamic pressure change process. The fusion layer uses a cross-attention mechanism to fuse the spatial branch, the temporal branch, and the surface electromyography time-frequency features in a multimodal manner. The contraction pattern classification results include correct contraction, Valsalva maneuver, abdominal muscle compensation, gluteal muscle compensation, and thigh adductor muscle compensation.

5. The apparatus according to claim 1, characterized in that, In the algorithm for constructing a digital profile of pelvic floor muscle function, all five dimensions are scored using a percentage system, and a comprehensive functional index is generated. The muscle strength level dimension uses the maximum voluntary contraction pressure value as the evaluation index, the muscle endurance dimension uses the pressure-time integral decay curve of the sustained contraction phase as the evaluation index, the explosive power dimension uses the pressure rise rate of the rapid contraction phase as the evaluation index, the symmetry dimension uses the pressure ratio of the left and right sides of the A area as the evaluation index, and the coordination dimension uses the contraction-relaxation transition time and the frequency of compensation as the evaluation index.

6. The apparatus according to claim 1, characterized in that, In the adaptive gamified rehabilitation training program generation algorithm: when type I muscle endurance deficiency is the main feature, a step-by-step continuous contraction program is generated. When type II muscle explosive power deficiency is the main issue, a rapid response training program is generated; when the frequency of compensatory behavior exceeds a preset threshold, an isolation training mode is first entered; the game control variable mapping adopts a multi-dimensional parallel mapping strategy: contraction force is mapped to the vertical displacement of the virtual object, duration is mapped to the horizontal propulsion distance of the virtual object, and the left and right pressure difference of area A is mapped to the horizontal offset of the virtual object; the game difficulty is dynamically adjusted according to the user's real-time performance using the scoring update formula R_new=R_old+K×(SE), where R_old is the difficulty score before adjustment, S is the user's actual performance, E is the expected performance probability, and K is the learning rate coefficient.

7. The apparatus according to claim 1, characterized in that, The main control module is also configured to: execute a self-calibration procedure before each use, collect the ambient temperature reference value through the temperature compensation unit, and detect the horizontal placement status of the flexible seat cushion body through the six-axis inertial measurement unit, allowing entry into the evaluation or training mode only when the horizontal placement deviation is less than a preset threshold; the communication module supports dual-mode communication of Bluetooth Low Energy protocol and Wi-Fi protocol.

8. A method for assessing postpartum pelvic floor muscle function based on a flexible pressure sensor array, employing the device as described in any one of claims 1-7, characterized in that, Includes the following steps: S1, Multimodal data acquisition: Pressure distribution time-series data is acquired at a sampling frequency f using the high-density flexible pressure sensor array, where f ≥ 50Hz. Simultaneously, electromyographic signals are acquired using the flexible surface electromyography electrodes, posture data is acquired using the six-axis inertial measurement unit, and ambient temperature data is acquired using the temperature compensation unit. The data acquisition is performed when the user executes a standard contraction sequence, which includes holding a maximum voluntary contraction for 5 seconds, performing 5 rapid contraction-relaxation cycles, and a 60-second sustained moderate-intensity contraction. S2, Signal preprocessing: Denoising and temperature drift compensation are performed on the pressure distribution time series data, bandpass filtering and rectification smoothing are performed on the electromyography signal, and Kalman filtering is performed on the posture data; S3, Pressure Field Reconstruction: The preprocessed signal is mapped into a two-dimensional pressure distribution map, and the peak pressure coordinates, pressure asymmetry index, pressure gradient field, pressure-time integral, and pressure rise rate are extracted. S4, Multi-muscle group compensation identification: Input spatiotemporal features, electromyographic time-frequency features and posture features into the neural network classification model, and output the contraction pattern classification result; when the classification result is correct contraction, execute step S5; when the classification result is compensation behavior, record the compensation type and mark the corresponding data segment, and reduce the weight of the data segment or remove it in the subsequent digital profile construction. S5, Construction of a digital profile of pelvic floor muscle function: Based on the spatiotemporal features extracted in step S3, quantitative scores for five dimensions—muscle strength, muscle endurance, explosive power, symmetry, and coordination—are calculated to generate a five-dimensional digital profile and a comprehensive functional index; the comprehensive functional index is compared with a preset threshold to output corresponding risk classification prompts and training suggestions. as well as S6, Result Output: Transmit the digital profile, compensation identification results and risk classification prompts to the mobile terminal for visual display; Establish a user-individualized baseline database to record the historical digital profiles of users at different postpartum recovery stages. Calculate the recovery progress of each dimension by comparing the current digital profile with the individualized baseline. When the recovery progress of a certain dimension stagnates or regresses in multiple consecutive assessments, generate training plan adjustment instructions.

9. A training control method for the device according to any one of claims 1-7, characterized in that, Includes the following steps: T1, Digital Profile Reading: Reads the user's latest digital profile of pelvic floor muscle function; T2, Defect Dominance Type Determination: Analyze the five-dimensional scores of the digital profile and identify the dimension with the lowest score as the defect dominance type; when the lowest score dimension is muscle endurance, it is determined to be a type I muscle defect dominance; when the lowest score dimension is explosive power, it is determined to be a type II muscle defect dominance; when the lowest score dimension is symmetry or coordination, it is determined to be a control defect dominance. T3, Training scheme generation: Based on the defect-dominant type, an initial training scheme is matched from a preset training scheme library, which includes a step-by-step continuous contraction scheme, a rapid response training scheme, a symmetrical balance training scheme, and a coordinated rhythm training scheme. T4, Gamified Execution: The real-time pressure data of the high-density flexible pressure sensor array is mapped to game control variables, and the game difficulty parameters are dynamically adjusted according to the user's real-time performance. T5, Training Quality Assessment: During training, the multi-muscle group compensation identification algorithm is executed in real time to count the frequency of compensation behavior; when the frequency of compensation behavior exceeds a preset threshold, the current training difficulty is reduced and an isolation guidance prompt is output. When a compensatory behavior such as breath-holding is detected, a sensory cue is output to prompt the user to stop the behavior. When abdominal muscle compensation is detected, a prompt is output guiding you to relax your abdomen and refocus on the lifting sensation of the pelvic floor muscles. as well as T6, Solution Iteration: After completing a training cycle, the evaluation process, including multimodal data acquisition, signal preprocessing, pressure field reconstruction, multi-muscle group compensation identification, pelvic floor muscle function digital profile construction and result output, is re-executed for the user to update the digital profile. The solution parameters for the next training cycle are adjusted according to the updated digital profile, forming a closed loop of evaluation-training-re-evaluation.

10. A postpartum pelvic floor muscle full-cycle remote rehabilitation management system, characterized in that, include: The device, mobile terminal, cloud server, and nursing terminal as described in any one of claims 1-7; the device is used to collect user pelvic floor muscle function data and perform edge computing; the mobile terminal is used to display the digital profile, execute a gamified training interface, and receive training program adjustment instructions; the cloud server is used to store historical data, execute swarm intelligence optimization algorithms, and push early warning information to the nursing terminal; The nursing terminal is used to receive early warning information and perform manual intervention scheduling; The swarm intelligence optimization algorithm includes: using a clustering algorithm to group users with similar initial digital profiles, comparing the improvement effects of different training scheme parameter configurations in each user group, and pushing the best-performing parameter configuration to the corresponding user group.

Images