Personalized intelligent pelvic floor muscle training system and method based on deep learning

Abstract

The invention relates to the technical field of human body rehabilitation training, and discloses a personalized intelligent pelvic floor muscle training system and method based on deep learning, and the method comprises the steps: collecting first multi-dimensional data through sensor equipment; preprocessing the first multi-dimensional data to obtain second multi-dimensional data; inputting the second multi-dimensional data into a 1D-CNN-Transformer model, and formulating a personalized pelvic floor muscle training scheme according to the pelvic floor muscle state and basic information of the user; related data of pelvic floor muscles of a user are continuously collected, and according to the training effect monitored in real time, a pelvic floor muscle training scheme is dynamically adjusted through a PPO algorithm; according to the method, individual differences are fully considered, the training pertinence is improved, the training scheme is dynamically adjusted according to the evaluation result, the training is more accurate and effective, the training efficiency and the training effect are improved, the dynamically adjusted training scheme can enable the user to feel the scientificity and the effectiveness of the training, and the training enthusiasm and the satisfaction degree of the user are improved.

Description

Technical Field

[0001] This invention relates to the field of human rehabilitation training technology, specifically to a personalized intelligent pelvic floor muscle training method and system based on deep learning. Background Technology

[0002] Pelvic floor muscles are crucial for maintaining normal physiological functions, playing a key role, especially in postpartum recovery for women and prevention of urinary incontinence in middle-aged and elderly people. Currently, pelvic floor muscle training mainly adopts the traditional Kegel exercise method. This method lacks personalized training guidance, and the training intensity and training plan are difficult to adjust precisely according to individual circumstances. At the same time, traditional training methods cannot monitor training effects in real time, making it difficult for users to understand the improvement of their own pelvic floor muscles, resulting in low training efficiency and poor training effects. Although some devices for assisting pelvic floor muscle training have appeared on the market, most of these devices only provide simple training reminders or basic training intensity control, lacking the ability to deeply analyze the user's pelvic floor muscle status and develop personalized training programs. Summary of the Invention

[0003] The purpose of this invention is to solve the above problems by designing a personalized intelligent pelvic floor muscle training method and system based on deep learning.

[0004] The first aspect of this invention provides a personalized intelligent pelvic floor muscle training method based on deep learning, the method comprising the following steps: The system uses sensor devices to collect bioelectrical signals, muscle contraction force, and muscle fatigue of the user's pelvic floor muscles, forming the first multidimensional data. The first multidimensional data is preprocessed, including noise removal and missing value imputation, to obtain the second multidimensional data; The second multidimensional data is input into the 1D-CNN-Transformer model, and a personalized pelvic floor muscle training plan is formulated based on the user's pelvic floor muscle status and basic information. We continuously collect data related to users' pelvic floor muscles and dynamically adjust the pelvic floor muscle training program using the PPO algorithm based on real-time monitoring of training effects.

[0005] Optionally, in a first implementation of the first aspect of the present invention, the preprocessing of the first multidimensional data, including noise removal and missing value imputation, to obtain the second multidimensional data, includes: Adaptive Gaussian white noise is added to the first multidimensional data, and multiple ensemble empirical mode decompositions are performed. After each decomposition, the extracted IMF components are removed, and different noise is added again for the next round of decomposition until the remaining signal is a monotonic trend term. Analyze the frequency characteristics and energy distribution of each IMF component, remove the high-frequency IMF components containing noise, and reconstruct the remaining effective IMF components to obtain the noise-removed data.

[0006] Optionally, in a second implementation of the first aspect of the present invention, the preprocessing of the first multidimensional data, including noise removal and missing value imputation to obtain the second multidimensional data, further includes: The noise-removed data is converted into a low-rank matrix. By constructing an augmented Lagrangian function, the low-rank matrix is ​​decomposed into a singular value matrix, a left singular matrix, and a right singular matrix using singular value decomposition. Based on the singular value matrix, left singular matrix, and right singular matrix, the singular values ​​are processed according to a preset threshold. Singular values ​​smaller than the threshold are set to zero. The low-rank matrix is ​​reconstructed through matrix operations to fill missing values ​​and obtain the second multidimensional data.

[0007] Optionally, in a third implementation of the first aspect of the present invention, the 1D-CNN-Transformer model includes a 1D-CNN branch and a Transformer branch, wherein the 1D-CNN branch includes three layers of convolutional kernels and pooling layers of different scales, and the Transformer branch includes position encoding, multi-head attention mechanism and skip connections.

[0008] Optionally, in a fourth implementation of the first aspect of the present invention, the step of inputting the second multidimensional data into the 1D-CNN-Transformer model and formulating a personalized pelvic floor muscle training program based on the user's pelvic floor muscle state and basic information includes: The second multidimensional data is organized into a continuous data stream according to the time series, and the user's basic information is encoded into a feature vector, which is then fused with the second multidimensional data. The fused overall data is used as the input of the 1D-CNN-Transformer model. After receiving input data, the 1D-CNN branch performs sliding operations on the time dimension of the data through three layers of convolutional kernels of different scales. Multiple feature maps are generated through convolution operations, and multiple feature maps are downsampled through pooling layers. After multiple convolution and pooling operations, the 1D-CNN branch forms a feature vector containing local and partial time series information. The Transformer branch receives the feature vectors output by the 1D-CNN branch, adds positional encoding information to the features at each time step, captures the long-distance dependencies between features in the data through a multi-head attention mechanism, and deeply fuses the outputs of the 1D-CNN branch and the Transformer branch through skip connections to form a sequence of feature vectors containing temporal context information. By combining the feature vector sequence of temporal context information with the user's basic information feature vector, a personalized pelvic floor muscle training program is developed, which includes training intensity, frequency, and movement type, based on the analysis of the user's current pelvic floor muscle status and personal basic information.

[0009] Optionally, in a fifth implementation of the first aspect of the present invention, a shallow convolutional layer with 7×1 convolutional kernels captures the μ-wave and β-wave features of bioelectrical signals, a middle convolutional layer with 15×1 convolutional kernels extracts the explosive force and endurance features of muscle contraction, and a deep convolutional layer with 31×1 convolutional kernels mines the nonlinear variation features of muscle fatigue, wherein the μ-wave is 8-13Hz and the β-wave is 13-30Hz.

[0010] Optionally, in a sixth implementation of the first aspect of the present invention, the step of continuously collecting user pelvic floor muscle-related data and dynamically adjusting the pelvic floor muscle training program using the PPO algorithm based on real-time monitored training effects includes: The system continuously collects data related to the user's pelvic floor muscles, which forms the state input of the PPO algorithm. The PPO algorithm generates a training adjustment plan based on the current state, and quantifies the training effect through a reward function to drive strategy optimization and obtain the adjusted pelvic floor muscle training plan.

[0011] A second aspect of the present invention provides a personalized intelligent pelvic floor muscle training system based on deep learning, the system comprising: The data acquisition module is used to collect bioelectrical signals, muscle contraction force, and muscle fatigue of the user's pelvic floor muscles using sensor devices, forming the first multidimensional data. The preprocessing module is used to preprocess the first multidimensional data, including noise removal and missing value imputation, to obtain the second multidimensional data; The program development module is used to input the second multidimensional data into the 1D-CNN-Transformer model and develop a personalized pelvic floor muscle training program based on the user's pelvic floor muscle status and basic information. The dynamic adjustment module is used to continuously collect data related to the user's pelvic floor muscles and dynamically adjust the pelvic floor muscle training program using the PPO algorithm based on the real-time monitoring of the training effect.

[0012] A third aspect of the present invention provides a deep learning-based personalized intelligent pelvic floor muscle training device, the deep learning-based personalized intelligent pelvic floor muscle training device comprising a memory and at least one processor, the memory storing instructions; the at least one processor calling the instructions in the memory to cause the deep learning-based personalized intelligent pelvic floor muscle training device to perform the various steps of the deep learning-based personalized intelligent pelvic floor muscle training method as described in any of the preceding claims.

[0013] A fourth aspect of the present invention provides a computer-readable storage medium storing instructions that, when executed by a processor, implement the steps of the deep learning-based personalized intelligent pelvic floor muscle training method as described in any of the preceding claims.

[0014] The technical solution provided by this invention utilizes sensor devices to collect bioelectrical signals, muscle contraction force, and muscle fatigue of the user's pelvic floor muscles, forming first multidimensional data. The first multidimensional data is preprocessed, including noise removal and missing value imputation, to obtain second multidimensional data. This second multidimensional data is then input into a 1D-CNN-Transformer model, and a personalized pelvic floor muscle training plan is developed based on the user's pelvic floor muscle state and basic information. The system continuously collects relevant data on the user's pelvic floor muscles and dynamically adjusts the training plan using the PPO algorithm based on real-time monitoring of the training effect. This invention fully considers individual differences, improves the targeting of training, and dynamically adjusts the training plan based on evaluation results, making training more precise and effective, improving training efficiency and results. The dynamically adjusted training plan allows users to experience the scientific nature and effectiveness of the training, increasing user motivation and satisfaction. Attached Figure Description

[0015] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention.

[0016] Figure 1 A flowchart illustrating a personalized intelligent pelvic floor muscle training method based on deep learning, provided in an embodiment of the present invention. Figure 2 A schematic diagram of the structure of a personalized intelligent pelvic floor muscle training system based on deep learning provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of a personalized intelligent pelvic floor muscle training device based on deep learning, provided in an embodiment of the present invention. Detailed Implementation

[0017] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” or “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, apparatus, product, or device that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0018] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 The flowchart of the personalized intelligent pelvic floor muscle training method based on deep learning provided in this embodiment of the invention includes the following steps: Step 101: Use sensor devices to collect bioelectrical signals, muscle contraction force, and muscle fatigue of the user's pelvic floor muscles to form the first multidimensional data. In this embodiment, before starting training, the user needs to wear professional sensor equipment. The surface electromyography (EMG) sensor is closely attached to the surface area of ​​the corresponding pelvic floor muscles through specially designed electrode pads. Utilizing the principle of bioelectrical conduction, it captures the weak electrical signals generated when muscle cells are excited in real time and converts them into bioelectrical signal data that can be recorded and analyzed. Its high sampling frequency can accurately capture every subtle change in muscle electrical activity. The miniature pressure sensor is placed according to different usage scenarios. For example, the vaginal dumbbell-type pressure sensor changes the pressure on the sensor when the pelvic floor muscles contract and relax. Through the built-in pressure sensing element, the pressure signal is transmitted. The signal is converted into an electrical signal, and after analog-to-digital conversion, the muscle contraction force data is accurately recorded. In addition, a wearable temperature sensor works in conjunction with the bioelectric signal acquisition to monitor changes in heat generated by local muscle metabolism. By combining the frequency and amplitude analysis of the bioelectric signal, the muscle fatigue is comprehensively assessed. These different types of sensors work simultaneously and integrate the collected pelvic floor muscle bioelectric signals, muscle contraction force, and muscle fatigue data according to a unified time series standard. Finally, a first multidimensional data set containing multidimensional information is formed, providing a rich and accurate data foundation for subsequent pelvic floor muscle status analysis and personalized training program development.

[0019] Step 102: Preprocess the first multidimensional data, including noise removal and missing value imputation, to obtain the second multidimensional data; In this embodiment, adaptive Gaussian white noise is added to the first multidimensional data, and multiple ensemble empirical mode decompositions are performed. After each decomposition, the extracted IMF components are removed, and different noises are added again for the next round of decomposition until the remaining signal is a monotonic trend term. The frequency characteristics and energy distribution of each IMF component are analyzed, the high-frequency IMF components containing noise are removed, and the remaining effective IMF components are reconstructed to obtain the noise-removed data.

[0020] In this embodiment, when processing the first multidimensional data, in order to overcome the mode aliasing problem in traditional empirical mode decomposition, adaptive Gaussian white noise needs to be added to the data. The so-called adaptive means that the intensity and distribution of the added Gaussian white noise are dynamically adjusted according to the characteristics of the data itself and the noise level. The purpose of adding Gaussian white noise is to use the uniform distribution characteristics of noise to separate the features of different scales in the original signal in the frequency space, so that the subsequent ensemble empirical mode decomposition can more accurately decompose the signal into different intrinsic mode function (IMF) components. The signal after adding noise retains the original pelvic floor muscle data features while having characteristics that are more conducive to decomposition. Multiple ensemble empirical mode decompositions are performed. After each decomposition, the extracted IMF components are removed, and different noises are added again for the next round of decomposition, until the remaining signal is a monotonic trend term. After noise addition is completed, ensemble empirical mode decomposition is performed on the data. In each decomposition process, the algorithm decomposes the signal into an IMF component and a residual component based on the local extremum characteristics of the signal. The IMF component represents the fluctuation characteristics of the signal within a specific frequency range, while the residual component contains the remaining signal information. After extracting an IMF component, it is removed from the current signal, and then new and different Gaussian white noise is added to the residual component. This is because adding different noise each time can make the decomposition process random. Taking the average result after multiple decompositions can effectively suppress mode aliasing. The above process of decomposition, removal of IMF components, and addition of new noise is repeated continuously. As the decomposition proceeds, the signal characteristics in the residual component gradually decrease until the remaining signal becomes a monotonic trend term. At this point, it is considered that all meaningful fluctuation characteristics in the original signal have been decomposed into the corresponding IMF components. After obtaining a series of IMF components, each IMF component needs to be analyzed in depth. By calculating and studying the frequency distribution of each IMF component, its main frequency range is determined. At the same time, the energy carried by each IMF component is analyzed. Since the added Gaussian white noise is mainly concentrated in the high-frequency part, most of the high-frequency IMF components in the decomposed IMF components contain noise information and have little effect on reflecting the true state of the pelvic floor muscles. By setting appropriate frequency thresholds and energy thresholds, IMF components with frequencies higher than a certain value and low energy proportions are identified as components that mainly contain noise. These high-frequency IMF components are removed from all IMF components, and only the effective IMF components that can reflect the true signal characteristics of the pelvic floor muscles are retained. After removing the high-frequency IMF components containing noise, the remaining effective IMF components represent useful information of different frequencies and features in the original pelvic floor muscle data. These effective IMF components are superimposed and reconstructed in the order of decomposition. Through signal synthesis, the information contained in each component is integrated to reconstruct a new signal. This new signal removes the noise interference in the original data and more clearly and accurately reflects the real data characteristics of pelvic floor muscles, such as bioelectrical signals, muscle contraction force, and muscle fatigue. This provides a high-quality data foundation for subsequent pelvic floor muscle status analysis and personalized training program development based on these data.

[0021] In this embodiment, the noise-removed data is converted into a low-rank matrix. By constructing an augmented Lagrangian function, the low-rank matrix is ​​decomposed into a singular value matrix, a left singular matrix, and a right singular matrix using singular value decomposition. Based on the singular value matrix, the left singular matrix, and the right singular matrix, the singular values ​​are processed according to a preset threshold. Singular values ​​smaller than the threshold are set to zero. The low-rank matrix is ​​reconstructed through matrix operations to fill missing values ​​and obtain the second multidimensional data.

[0022] In this embodiment, the noise-removed data contains multi-dimensional information such as bioelectrical signals of pelvic floor muscles, muscle contraction force, and muscle fatigue. In order to facilitate the subsequent use of matrix properties for missing value filling, it needs to be converted into a low-rank matrix form. The low-rank matrix assumes that the pelvic floor muscle data has certain intrinsic correlations and structural features in the high-dimensional space, so that most of the information of the matrix can be represented by a few main feature vectors and singular values. By arranging and organizing the data in a reasonable way, the pelvic floor muscle data of different time points and different types are mapped to the rows and columns of the matrix, and a low-rank matrix that can reflect the intrinsic structure of the data is constructed. After this conversion, the feature structure of the data is clearer. The augmented Lagrangian function (ALS) is constructed to introduce constraints during matrix factorization, thereby better solving for the optimal decomposition form of low-rank matrices. The ALS transforms the low-rank matrix factorization problem into an optimization problem. By minimizing this function, the matrix factorization result that best represents the characteristics of the original data can be found. In this process, the powerful matrix factorization tool Singular Value Decomposition (SVD) is used to decompose the low-rank matrix into three matrices: the singular value matrix, the left singular matrix, and the right singular matrix. The singular value matrix contains the singular values ​​of the original matrix, reflecting the importance of different features within the matrix. The left and right singular matrices are then correlated with the singular value matrix, together forming a complete representation of the original matrix. Through this decomposition, the information of the original low-rank matrix is ​​reorganized and refined. After obtaining the singular value matrix, left singular matrix, and right singular matrix, the singular values ​​need to be processed. Since the magnitude of the singular values ​​reflects the importance of the corresponding features in the original matrix, smaller singular values ​​usually correspond to noise or unimportant information in the data. To remove this noise and redundant information while retaining the main data features, a threshold is preset based on the characteristics of the data and actual needs. Then, all singular values ​​in the singular value matrix smaller than this threshold are set to zero. After this processing, the non-zero singular values ​​and their corresponding features retained in the singular value matrix are the parts that significantly contribute to the original pelvic floor muscle data, while the information corresponding to the singular values ​​set to zero is considered negligible noise or secondary information. In this way, further filtering and optimization of the original data features are achieved. After processing the singular values, the processed singular value matrix, left singular matrix, and right singular matrix are recombined using matrix multiplication and other operations to reconstruct a new low-rank matrix. This reconstructed low-rank matrix retains the main features of the original data while effectively filling in missing values. Because the algorithm utilizes the overall structure and correlation information of the data during matrix decomposition and singular value processing, it can reasonably infer the possible values ​​of missing values ​​based on the characteristics and distribution patterns of the known data and fill them in the corresponding positions. The resulting low-rank matrix containing the missing value filling results, after appropriate transformation and organization, is restored to a multidimensional data form, i.e., the second multidimensional data. At this point, the second multidimensional data not only removes noise interference but also completes the filling of missing values, significantly improving the integrity and quality of the data.

[0023] Step 103: Input the second multidimensional data into the 1D-CNN-Transformer model, and formulate a personalized pelvic floor muscle training plan based on the user's pelvic floor muscle status and basic information. In this embodiment, the 1D-CNN-Transformer model includes a 1D-CNN branch and a Transformer branch. The 1D-CNN branch includes three layers of convolutional kernels and pooling layers of different scales, while the Transformer branch includes position encoding, multi-head attention mechanism, and skip connections.

[0024] In this embodiment, the second multidimensional data is organized into a continuous data stream according to the time series order, and the user's basic information is encoded into a feature vector, which is then fused with the second multidimensional data. The fused overall data is used as the input of the 1D-CNN-Transformer model. After receiving input data, the 1D-CNN branch performs sliding operations on the time dimension of the data through three layers of convolutional kernels of different scales. Multiple feature maps are generated through convolution operations, and multiple feature maps are downsampled through pooling layers. After multiple convolution and pooling operations, the 1D-CNN branch forms a feature vector containing local and partial time series information. The Transformer branch receives the feature vectors output by the 1D-CNN branch, adds positional encoding information to the features at each time step, captures the long-distance dependencies between features in the data through a multi-head attention mechanism, and deeply fuses the outputs of the 1D-CNN branch and the Transformer branch through skip connections to form a sequence of feature vectors containing temporal context information. By combining the feature vector sequence of temporal context information with the user's basic information feature vector, a personalized pelvic floor muscle training program is developed, which includes training intensity, frequency, and movement type, based on the analysis of the user's current pelvic floor muscle status and personal basic information.

[0025] In this embodiment, the shallow convolutional layer with 7×1 convolutional kernels captures the μ-wave and β-wave characteristics of bioelectrical signals, the middle convolutional layer with 15×1 convolutional kernels extracts the explosive force and endurance characteristics of muscle contraction, and the deep convolutional layer with 31×1 convolutional kernels mines the nonlinear variation characteristics of muscle fatigue, wherein the μ-wave is 8-13Hz and the β-wave is 13-30Hz.

[0026] Step 104: Continuously collect data related to the user's pelvic floor muscles, and dynamically adjust the pelvic floor muscle training program using the PPO algorithm based on the real-time monitoring of the training effect.

[0027] In this embodiment, data related to the user's pelvic floor muscles are continuously collected and used as the state input for the PPO algorithm. The PPO algorithm generates a training adjustment plan based on the current state and quantifies the training effect through a reward function to drive strategy optimization and obtain the adjusted pelvic floor muscle training plan.

[0028] In this embodiment, the PPO algorithm treats the user's pelvic floor muscle training process as a dynamic virtual environment. In this environment, the real-time collected bioelectrical signals of the pelvic floor muscles, muscle contraction force, muscle fatigue data, and basic information such as the user's age, gender, and weight are combined to form a state that the algorithm can perceive. For example, when the user's muscle contraction force continuously decreases during training, combined with information such as age, it forms a specific manifestation of the current training state. At the same time, a set of operable actions is defined, including adjusting the training intensity, such as increasing or decreasing the contraction force, changing the training frequency, increasing or decreasing the number of training sessions, and changing the type of training action from slow muscle contraction training to fast muscle contraction training. The reward is the key indicator for the algorithm to evaluate the adjustment effect. It is set as a feedback signal that comprehensively considers multiple factors such as the degree of muscle strength improvement, fatigue recovery speed, and user training comfort. For example, if the user's muscle strength is significantly improved and fatigue recovery is good after adjusting the training plan, the algorithm will give a higher reward. During user training, sensors continuously collect data related to pelvic floor muscles. This data is transmitted to the PPO algorithm in real time. The algorithm integrates the new data with the previous state information and quickly analyzes the latest state of the user's pelvic floor muscle training. For example, when the user's bioelectrical signal shows that the muscles are overly fatigued and the training intensity is at a high level, the algorithm determines that the current training state is risky and needs to be adjusted. The PPO algorithm evaluates the current training state based on existing strategies and selects appropriate adjustment actions from the action set. This strategy is formed by the algorithm through continuous learning and experience accumulation. It takes into account the historical effects of different actions in similar states. For example, in the past, when encountering similar muscle fatigue and high-intensity training states, actions that reduced training intensity and increased rest time achieved good results. Therefore, the algorithm may prioritize similar adjustment actions in this case. When selecting actions, the algorithm does not follow the previous strategy in an unchanging manner, but will try new actions with a certain probability in order to explore better training schemes and adjustments, and avoid getting stuck in local optima. For example, it will occasionally try a new combination of training actions and observe its impact on training results. After the adjustment is performed, the algorithm obtains corresponding reward feedback based on the changes in the user's training effect. If the adjusted training plan significantly improves muscle strength and speeds up fatigue recovery, the algorithm will receive a positive reward, indicating that the action selection is effective. Conversely, if the training effect is poor, such as increased muscle fatigue or no improvement in strength, the algorithm will receive a negative reward. Based on the reward feedback obtained, the PPO algorithm optimizes the original strategy. For actions that receive positive rewards, the algorithm increases the probability of that action being selected in similar situations; for actions that receive negative rewards, it reduces the probability of selection. By continuously optimizing the strategy based on reward feedback, the PPO algorithm gradually forms a scheme adjustment method that is more suitable for the current user's training situation, realizing dynamic optimization of the pelvic floor muscle training scheme and ensuring that the training scheme always fits the user's physical condition and training goals.

[0029] In this embodiment, a certain sample size of users is selected and divided into a system-assisted training group and a traditional training group. Rehabilitation experts regularly evaluate users using standardized assessment scales, such as quantitative indicators like the rate of pelvic floor electromyography signal attainment, the extent of improvement in contractile strength, and the speed of fatigue recovery. For example, if users in the system group have a 30% higher rate of muscle contractile strength attainment and a 25% faster rate of fatigue reduction after 4 weeks of training compared to the traditional group, and the experts score the rationality of the program adjustment at over 90%, the quantitative data from the expert evaluation avoids the subjectivity of the effect description and provides a clear reference and verification basis for the improvement of the technical effect.

[0030] Please see Figure 2 A schematic diagram of the structure of a personalized intelligent pelvic floor muscle training system based on deep learning provided in this embodiment of the invention. The system includes: The data acquisition module is used to collect bioelectrical signals, muscle contraction force, and muscle fatigue of the user's pelvic floor muscles using sensor devices, forming the first multidimensional data. The preprocessing module is used to preprocess the first multidimensional data, including noise removal and missing value imputation, to obtain the second multidimensional data; The program development module is used to input the second multidimensional data into the 1D-CNN-Transformer model and develop a personalized pelvic floor muscle training program based on the user's pelvic floor muscle status and basic information. The dynamic adjustment module is used to continuously collect data related to the user's pelvic floor muscles and dynamically adjust the pelvic floor muscle training program using the PPO algorithm based on the real-time monitoring of the training effect.

[0031] Figure 3This is a schematic diagram of a personalized intelligent pelvic floor muscle training device based on deep learning, provided in an embodiment of the present invention. The personalized intelligent pelvic floor muscle training device 300 based on deep learning can vary significantly due to different configurations or performance. It may include one or more central processing units (CPUs) 310 (e.g., one or more processors) and a memory 320, and one or more storage media 330 (e.g., one or more mass storage devices) for storing application programs 333 or data 332. The memory 320 and storage media 330 can be temporary or persistent storage. The program stored in the storage media 330 may include one or more modules (not shown in the diagram), each module including a series of instruction operations on the personalized intelligent pelvic floor muscle training device 300 based on deep learning. Furthermore, the processor 310 may be configured to communicate with the storage media 330 and execute the series of instruction operations in the storage media 330 on the personalized intelligent pelvic floor muscle training device 300 based on deep learning to implement the method provided in the above embodiment.

[0032] The deep learning-based personalized intelligent pelvic floor muscle training device 300 may also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input / output interfaces 360, and / or one or more operating devices 331, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that... Figure 3 The structure of the personalized intelligent pelvic floor muscle training device based on deep learning shown does not constitute a limitation on the computer device provided by the present invention. It may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0033] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when the instructions are executed on a computer, cause the computer to perform the various steps of the deep learning-based personalized intelligent pelvic floor muscle training method provided in the above embodiments.

[0034] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the above-described equipment or apparatus / unit can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0035] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0036] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A personalized intelligent pelvic floor muscle training method based on deep learning, characterized in that, The method includes the following steps: The system uses sensor devices to collect bioelectrical signals, muscle contraction force, and muscle fatigue of the user's pelvic floor muscles, forming the first multidimensional data. The first multidimensional data is preprocessed, including noise removal and missing value imputation, to obtain the second multidimensional data; The second multidimensional data is input into the 1D-CNN-Transformer model, and a personalized pelvic floor muscle training plan is formulated based on the user's pelvic floor muscle status and basic information. We continuously collect data related to users' pelvic floor muscles and dynamically adjust the pelvic floor muscle training program using the PPO algorithm based on real-time monitoring of training effects.

2. The personalized intelligent pelvic floor muscle training method based on deep learning as described in claim 1, characterized in that, The preprocessing of the first multidimensional data, including noise removal and missing value imputation, to obtain the second multidimensional data includes: Adaptive Gaussian white noise is added to the first multidimensional data, and multiple ensemble empirical mode decompositions are performed. After each decomposition, the extracted IMF components are removed, and different noise is added again for the next round of decomposition until the remaining signal is a monotonic trend term. Analyze the frequency characteristics and energy distribution of each IMF component, remove the high-frequency IMF components containing noise, and reconstruct the remaining effective IMF components to obtain the noise-removed data.

3. The personalized intelligent pelvic floor muscle training method based on deep learning as described in claim 2, characterized in that, The preprocessing of the first multidimensional data, including noise removal and missing value imputation, to obtain the second multidimensional data, further includes: The noise-removed data is converted into a low-rank matrix. By constructing an augmented Lagrangian function, the low-rank matrix is ​​decomposed into a singular value matrix, a left singular matrix, and a right singular matrix using singular value decomposition. Based on the singular value matrix, left singular matrix, and right singular matrix, the singular values ​​are processed according to a preset threshold. Singular values ​​smaller than the threshold are set to zero. The low-rank matrix is ​​reconstructed through matrix operations to fill missing values ​​and obtain the second multidimensional data.

4. The personalized intelligent pelvic floor muscle training method based on deep learning as described in claim 1, characterized in that, The 1D-CNN-Transformer model includes a 1D-CNN branch and a Transformer branch. The 1D-CNN branch includes three layers of convolutional kernels and pooling layers of different scales, while the Transformer branch includes position encoding, multi-head attention mechanism, and skip connections.

5. The personalized intelligent pelvic floor muscle training method based on deep learning as described in claim 4, characterized in that, The process involves inputting the second multidimensional data into the 1D-CNN-Transformer model to develop a personalized pelvic floor muscle training program based on the user's pelvic floor muscle status and basic information, including: The second multidimensional data is organized into a continuous data stream according to the time series, and the user's basic information is encoded into a feature vector, which is then fused with the second multidimensional data. The fused overall data is used as the input of the 1D-CNN-Transformer model. After receiving input data, the 1D-CNN branch performs sliding operations on the time dimension of the data through three layers of convolutional kernels of different scales. Multiple feature maps are generated through convolution operations, and multiple feature maps are downsampled through pooling layers. After multiple convolution and pooling operations, the 1D-CNN branch forms a feature vector containing local and partial time series information. The Transformer branch receives the feature vectors output by the 1D-CNN branch, adds positional encoding information to the features at each time step, captures the long-distance dependencies between features in the data through a multi-head attention mechanism, and deeply fuses the outputs of the 1D-CNN branch and the Transformer branch through skip connections to form a sequence of feature vectors containing temporal context information. By combining the feature vector sequence of temporal context information with the user's basic information feature vector, a personalized pelvic floor muscle training program is developed, which includes training intensity, frequency, and movement type, based on the analysis of the user's current pelvic floor muscle status and personal basic information.

6. The personalized intelligent pelvic floor muscle training method based on deep learning as described in claim 5, characterized in that, The shallow convolutional layer with 7×1 kernels captures the μ-wave and β-wave characteristics of bioelectrical signals, the middle convolutional layer with 15×1 kernels extracts the explosive force and endurance characteristics of muscle contraction, and the deep convolutional layer with 31×1 kernels mines the nonlinear variation characteristics of muscle fatigue, where the μ-wave is 8-13Hz and the β-wave is 13-30Hz.

7. The personalized intelligent pelvic floor muscle training method based on deep learning as described in claim 1, characterized in that, The continuous collection of user pelvic floor muscle-related data, and the dynamic adjustment of the pelvic floor muscle training program using the PPO algorithm based on real-time monitoring of training effects, includes: The system continuously collects data related to the user's pelvic floor muscles, which forms the state input of the PPO algorithm. The PPO algorithm generates a training adjustment plan based on the current state, and quantifies the training effect through a reward function to drive strategy optimization and obtain the adjusted pelvic floor muscle training plan.

8. A personalized intelligent pelvic floor muscle training system based on deep learning, characterized in that, The system includes: The data acquisition module is used to collect bioelectrical signals, muscle contraction force, and muscle fatigue of the user's pelvic floor muscles using sensor devices, forming the first multidimensional data. The preprocessing module is used to preprocess the first multidimensional data, including noise removal and missing value imputation, to obtain the second multidimensional data; The program development module is used to input the second multidimensional data into the 1D-CNN-Transformer model and develop a personalized pelvic floor muscle training program based on the user's pelvic floor muscle status and basic information. The dynamic adjustment module is used to continuously collect data related to the user's pelvic floor muscles and dynamically adjust the pelvic floor muscle training program using the PPO algorithm based on the real-time monitoring of the training effect.

9. A personalized intelligent pelvic floor muscle training device based on deep learning, characterized in that, The deep learning-based personalized intelligent pelvic floor muscle training device includes a memory and at least one processor, wherein the memory stores instructions; the at least one processor invokes the instructions in the memory to cause the deep learning-based personalized intelligent pelvic floor muscle training device to perform each step of the deep learning-based personalized intelligent pelvic floor muscle training method as described in any one of claims 1-7.

10. A computer-readable storage medium storing instructions thereon, characterized in that, When the instructions are executed by the processor, they implement the steps of the deep learning-based personalized intelligent pelvic floor muscle training method as described in any one of claims 1-7.

Images