Individual soldier physical ability dynamic evaluation and intelligent intensive training system and method thereof

Abstract

The invention relates to the technical field of military training, in particular to an individual soldier physical ability dynamic evaluation and intelligent intensive training system and a method thereof.The system comprises a hardware sensing layer, an intelligent analysis layer and a decision execution layer, training parameters are optimized, regulated and controlled by collecting electromyographic signals and kinematics data, and the training performance of the individual soldier physical ability dynamic evaluation and intelligent intensive training system is improved. The intelligent analysis layer comprises a multi-modal space-time alignment unit, a motion efficiency evaluation unit and a genetic algorithm optimization unit, through the units, the system can calculate a training efficiency index and generate a training parameter optimization scheme, the scheme comprises a nonlinear fluctuation type load parameter, real-time parameter optimization and personalized training path planning can be realized, and the training efficiency is improved. Through the application of the system, the new soldier standard reaching time is shortened from 12 weeks to 8.4 weeks (reduced by 30%), meanwhile, the invalid action proportion is reduced from 35% to 12%, and the occurrence rate of training related muscle strain and joint inflammation is reduced from 17.3% to 8.7% (reduced by 50%) through a real-time biomechanical risk early warning and active intervention mechanism.

Description

Technical Field

[0001] The present invention relates to the field of military training technology, and in particular to a system and method for dynamic assessment and intelligent reinforcement training of individual soldiers, in particular to a personalized training system based on biomechanical data acquisition and intelligent analysis. Background Art

[0002] The traditional individual soldier physical training system suffers from a low degree of standardization and strong subjective dependence. Specifically, the assessment dimensions are single, relying on manual observation (such as movement standardization) or single physiological indicators (such as heart rate and blood oxygen) for physical fitness assessment, which is unable to quantify muscle co-activation efficiency (such as the synchronization of the quadriceps and hamstrings) and joint dynamic characteristics (such as knee flexion angular velocity); the training plan is static, using a "one-size-fits-all" training plan (such as fixed weight and number of repetitions), ignoring individual soldiers' biological differences in muscle recruitment ability (such as differences in the proportion of type II muscle fibers) and joint mobility (such as insufficient hip joint flexibility); injury warning is delayed, and existing physical fitness monitoring equipment (such as heart rate monitors and GPS bracelets) can only capture macroscopic physiological indicators and lack the real-time identification ability of microscopic biomechanical abnormalities (such as excessive knee internal rotation angle and overcompensation of the erector spinae muscles). As a result, sports injuries often occur during the stage of accumulated movement deformation.

[0003] In recent years, wearable sensing and intelligent algorithm technologies have been gradually applied to military training. For example, one research team used an inertial measurement unit (IMU) to interpret tactical postures, but its data sampling rate (≤500Hz) made it difficult to capture the transient characteristics of explosive movements (such as the EMG peak during a squat jump). Another study used surface electromyography (sEMG) sensors to analyze muscle fatigue, but failed to integrate these with joint kinematic parameters (such as ankle dorsiflexion angle), resulting in assessment results that deviated from actual combat requirements.

[0004] Existing technical solutions have the following key bottlenecks: fragmented data acquisition. Multi-source sensor data such as electromyography, acceleration, and angular velocity are collected in an asynchronous mode (timestamp deviation ≥ 10ms), resulting in cross-modal signal alignment errors. For example, it is impossible to accurately correlate the "oblique abdominal muscle activation timing during crawling" with the "shoulder joint rotation angle"; the algorithm adaptability is poor. Traditional machine learning models (such as SVM and random forest) rely on manually annotated static data sets and are difficult to adapt to the dynamically changing physiological states of soldiers during training (such as lactate threshold drift and improved neuromuscular adaptability); the intervention is not real-time enough. The training plan adjustment cycle is as long as 48-72 hours (relying on manual review and analysis), and it is impossible to achieve closed-loop optimization of load parameters within a single training session (such as dynamically adjusting the load based on the real-time electromyographic spectrum entropy value). Summary of the Invention

[0005] The purpose of the present invention is to provide a system and method for dynamic assessment and intelligent reinforcement training of individual soldiers, aiming to solve the technical problems of single assessment dimension, static training plan and delayed injury warning in traditional training.

[0006] The present invention proposes a system for dynamic assessment and intelligent reinforcement training of individual soldiers, comprising:

[0007] Hardware perception layer, used to collect electromyographic signals and kinematic data during training;

[0008] an intelligent analysis layer, connected to the hardware perception layer, configured to receive and process the electromyographic signals and the kinematic data, and generate a training parameter optimization solution; and

[0009] A decision execution layer, connected to the intelligent analysis layer, configured to receive the training parameter optimization plan and execute training control;

[0010] The intelligent analysis layer includes:

[0011] a multimodal spatiotemporal alignment unit, configured to synchronize the myoelectric signal with the kinematic data;

[0012] a motion effectiveness evaluation unit, configured to calculate a training effectiveness index based on the time-synchronized electromyographic signals and kinematic data;

[0013] A genetic algorithm optimization unit is used to generate the training parameter optimization scheme based on the training effectiveness index using a genetic algorithm with a double-layer parallel architecture, wherein the training parameter optimization scheme includes a nonlinear fluctuating load parameter.

[0014] Preferably, the hardware perception layer includes:

[0015] A high-density myoelectric sensor array is used to collect muscle activation data and obtain the myoelectric signal;

[0016] A nine-axis IMU fusion unit, used to collect joint kinematic parameters and obtain the kinematic data; and

[0017] A data preprocessing unit is used to filter and normalize the electromyographic signal and the kinematic data.

[0018] Preferably, the high-density electromyography sensor array includes an 8-channel wireless sEMG module with a sampling rate of 2000 Hz and a common mode rejection ratio of not less than 110 dB; the nine-axis IMU fusion unit integrates a three-axis accelerometer, a gyroscope and a magnetometer, and has an output frequency of 100 Hz.

[0019] Preferably, the multimodal spatiotemporal alignment unit adopts a method combining dynamic time warping with the maximum mutual information criterion to control the time synchronization error between the electromyographic signal and the kinematic data within 2 milliseconds.

[0020] Preferably, the exercise performance evaluation unit is used for:

[0021] Calculating movement completion, including joint angle compliance, muscle activation sequence correctness, and movement stability index;

[0022] Calculating energy expenditure efficiency, which is the ratio of mechanical work to muscle energy expenditure to complete the task; and

[0023] The training effectiveness index is generated based on the movement completion degree and the energy consumption efficiency.

[0024] Preferably, the two-layer parallel architecture of the genetic algorithm optimization unit includes:

[0025] An outer parallel structure for simultaneously evaluating multiple candidate training solutions; and

[0026] The inner parallel structure is used to parallelly calculate the muscle-joint coupling matrix during the evaluation of each candidate training program;

[0027] The genetic algorithm optimization unit further includes a dynamic constraint mechanism for dynamically adjusting the range of training parameter changes according to the fatigue index.

[0028] Preferably, the intelligent analysis layer further includes:

[0029] An individualized training path planning unit, configured to generate an individualized training path based on individual feature vectors and training response patterns;

[0030] The individual feature vector includes muscle fiber type distribution characteristics, joint mobility characteristics, muscle strength ratio, movement learning curve parameters, fatigue recovery mode parameters and injury risk factors.

[0031] Preferably, the decision execution layer includes:

[0032] Augmented reality interface for projecting biomechanical feedback in real time;

[0033] an adaptive resistance control module, configured to adjust the training load according to the training parameter optimization scheme; and

[0034] A biofeedback intervention unit is used to trigger intervention measures when abnormal biomechanical patterns are detected.

[0035] Preferably, the biofeedback intervention unit includes an electrical stimulation module for applying electrical stimulation of a specific frequency to the relevant muscles to suppress abnormal compensation when abnormal joint angles are detected.

[0036] A method for dynamic assessment of individual soldier physical fitness and intelligent reinforcement training, applied to any of the above systems, comprises the following steps:

[0037] Collect electromyographic signals and kinematic data during training;

[0038] Time-synchronizing the electromyographic signal with the kinematic data to obtain time-synchronized data;

[0039] Calculating action completion and energy consumption efficiency based on the time synchronization data;

[0040] generating a training effectiveness index based on the movement completion degree and the energy consumption efficiency;

[0041] Using a genetic algorithm with a two-layer parallel architecture, generating a training parameter optimization scheme based on the training effectiveness index, the training parameter optimization scheme including a nonlinear fluctuating load parameter; and

[0042] Training regulation is performed according to the training parameter optimization plan, and intervention measures are triggered when abnormal biomechanical patterns are detected.

[0043] The beneficial effects of the present invention include:

[0044] 1. Significantly improved training efficiency: Through real-time parameter optimization and personalized training path planning, the time it takes for new recruits to reach training standards has been shortened from 12 weeks to 8.4 weeks (a 30% reduction), resulting from Pareto optimization of daily training effectiveness (the proportion of ineffective movements has been reduced from 35% to 12%).

[0045] 2. Training safety has been significantly improved: The incidence of training-related muscle strains and joint inflammation has dropped from 17.3% to 8.7% (a 50% reduction), attributed to real-time biomechanical risk warning (sensitivity 92%, specificity 88%) and proactive intervention mechanisms.

[0046] 3. Comprehensive enhancement of combat capabilities: explosive power increased (30-meter sprint results increased by 0.4 seconds, quadriceps peak power increased by 18%); endurance optimized (400-meter obstacle race average heart rate reduced by 11 bpm under a 25 kg load, oxygen uptake utilization increased by 15%).

[0047] 4. Significant reduction in logistics costs: The system reduces reliance on professional fitness coaches (saving 40% in labor costs) and indirectly saves approximately RMB 270,000 per year in medical expenses (calculated per trip) by preventing injuries. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is the overall architecture diagram of the individual soldier physical fitness dynamic assessment and intelligent reinforcement training system of the present invention;

[0049] Figure 2 Schematic diagram of the hardware perception layer of the present invention;

[0050] Figure 3 Schematic diagram of the composition of the intelligent analysis layer of the present invention;

[0051] Figure 4 Schematic diagram of the composition of the decision execution layer of the present invention;

[0052] Figure 5 This is a workflow diagram of the multimodal spatiotemporal alignment unit of the present invention;

[0053] Figure 6 is a calculation flow chart of the exercise efficiency evaluation unit of the present invention;

[0054] Figure 7 Schematic diagram of a double-layer parallel architecture of the genetic algorithm optimization unit of the present invention;

[0055] Figure 8 This is a workflow diagram of the individualized training path planning unit of the present invention;

[0056] Figure 9 Flowchart of the method of the present invention. DETAILED DESCRIPTION

[0057] Please refer to the attached Figure 1-9 , the specific implementation of the present invention is further described in detail below with reference to the accompanying drawings.

[0058] like Figure 1 As shown in the figure, the individual soldier dynamic physical fitness assessment and intelligent reinforcement training system provided by the present invention includes three main layers: hardware perception layer 1, intelligent analysis layer 2, and decision execution layer 3. Hardware perception layer 1 is used to collect electromyographic signals and kinematic data during training; intelligent analysis layer 2 is connected to hardware perception layer 1 to receive and process electromyographic signals and kinematic data and generate training parameter optimization solutions; decision execution layer 3 is connected to intelligent analysis layer 2 to receive training parameter optimization solutions and execute training control.

[0059] First, the structure and function of the hardware perception layer 1 are described in detail. Figure 2 As shown, the hardware perception layer 1 includes a high-density electromyography sensor array 11, a nine-axis IMU fusion unit 12 and a data preprocessing unit 13.

[0060] The high-density electromyographic sensor array 11 is used to collect muscle activation data and obtain electromyographic signals. In a preferred embodiment of the present invention, the high-density electromyographic sensor array 11 includes an 8-channel wireless sEMG module with a sampling rate of 2000Hz and a common mode rejection ratio of not less than 110dB. This high sampling rate design can effectively capture the transient muscle activation characteristics in explosive movements, such as the electromyographic peak in the squat take-off phase. Preferably, the high-density electromyographic sensor array 11 covers the main muscle groups of the lower limbs, such as the rectus femoris, vastus lateralis, semitendinosus, etc., and can comprehensively monitor the muscle synergistic activation pattern. For example, in a squat movement, the system will synchronously collect the activity of the quadriceps femoris, hamstrings, and gluteus maximus, and analyze whether their activation timing and synergistic pattern meet the standard movement requirements.

[0061] The nine-axis IMU fusion unit 12 is used to collect joint kinematic parameters and obtain kinematic data. In a preferred embodiment of the present invention, the nine-axis IMU fusion unit 12 integrates a three-axis accelerometer (range ±16g), a gyroscope (range ±2000° / s) and a magnetometer, and the output frequency is 100Hz. The nine-axis IMU fusion unit 12 can output the joint Euler angle in real time, and the angle accuracy reaches ±0.5°. In order to eliminate the heading angle drift in a dynamic environment, the present invention uses a quaternion extended Kalman filter algorithm to process IMU data. The core of the algorithm is to achieve high-precision posture estimation by fusing acceleration, angular velocity and magnetic field data, and its state equation can be expressed as:

[0062] x k+1 =F k x k +w k ,

[0063] Among them, x k is the state vector at the kth moment, including the quaternion attitude q k =[q0,q1,q2,q3] T and gyroscope bias b k =[b x ,b y ,b z ] T ,Right now F k is the state transfer matrix, according to the angular velocity ω k =[ω x ,ω y ,ω z ] T The integral is calculated; w k is process noise, with mean zero and covariance matrix Q k Gaussian distribution, that is In practical applications, for example, when a soldier performs a squat, nine-axis IMU fusion units 12 are installed in the mid-thigh and mid-calf to measure knee flexion angles in real time. Using the quaternion extended Kalman filter algorithm, the system maintains high accuracy in angle measurement, within ±0.5°, even in rapidly changing training environments (such as rapid squats or explosive power training).

[0064] The data preprocessing unit 13 is used to filter and normalize the electromyographic signals and kinematic data. For the electromyographic signals, a bandpass filter (20-450Hz) is used to remove baseline drift and high-frequency noise; for the IMU data, a median filter window (window width is 5 sampling points) is used to eliminate spike noise. The quality of the preprocessed data is significantly improved, laying the foundation for subsequent analysis. In actual training, for example, when soldiers conduct armed raid training, equipment vibration and strenuous body movement will cause an increase in signal noise. At this time, the role of the data preprocessing unit 13 is particularly important, which can effectively filter out the noise caused by the environment and movement to ensure signal quality.

[0065] The structure and function of the intelligent analysis layer 2 are described in detail below. Figure 3 As shown, the intelligent analysis layer 2 includes a multimodal spatiotemporal alignment unit 21 , a motion efficiency evaluation unit 22 , a genetic algorithm optimization unit 23 and an individualized training path planning unit 24 .

[0066] The multimodal spatiotemporal alignment unit 21 is used to synchronize the time of myoelectric signals with kinematic data. In the prior art, multi-source sensor data is acquired asynchronously, and timestamp deviations are typically greater than 10ms, resulting in cross-modal signal alignment errors. The multimodal spatiotemporal alignment unit 21 of the present invention combines dynamic time warping with the maximum mutual information criterion to control the time synchronization error between myoelectric signals and kinematic data to within 2 milliseconds, achieving millisecond-level precision alignment.

[0067] The dynamic time warping (DTW) algorithm is an effective method to solve the time series alignment problem. Its core idea is to find the optimal nonlinear mapping relationship between two time series. In this paper, the DTW algorithm is used to find the optimal time correspondence between electromyographic signals and IMU data. Suppose the electromyographic signal sequence is X = {x1, x2, ..., x n}, the IMU data sequence is Y={y1,y2,...,y m}, the DTW algorithm calculates the minimum cumulative distance matrix D:

[0068] D(i,j)=d(x i ,y j )+min{D(i-1,j),D(i,j-1),D(i-1,j-1)},

[0069] Where D(i,j) represents the minimum cumulative distance between the first i points of sequence X and the first j points of sequence Y; d(x i ,y j ) is the point distance metric function, and the Euclidean distance is usually used to calculate the i-th electromyographic signal sampling point x i and the jth IMU data sampling point y j The distance between them; the value range of i is 1≤i≤n, and the value range of j is 1≤j≤m; min{D(i-1,j),D(i,j-1),D(i-1,j-1)} means taking the minimum value among the three, which corresponds to the distance cost of the three operations of insertion, deletion and matching respectively.

[0070] In practical applications, such as when soldiers perform squats, muscle activation often precedes joint movement. The DTW algorithm can capture this nonlinear temporal relationship and accurately align quadriceps EMG activity with knee joint angle changes. To further improve alignment accuracy, for rapid movements (such as the burst phase of a squat), this paper employs a local constraint strategy to limit the degree of temporal bending, avoid excessive deformation, and ensure physiologically sound alignment.

[0071] To further improve the alignment accuracy, the present invention combines the maximum mutual information criterion and performs fine adjustments based on the DTW coarse alignment. The mutual information calculation formula is:

[0072]

[0073] Where I(X;Y) represents the mutual information value between the EMG signal sequence X and the IMU data sequence Y, which characterizes the statistical correlation between the two signals; p(x,y) is the joint probability distribution of the EMG signal x and the IMU data y, estimated by constructing a two-dimensional histogram; p(x) and p(y) are marginal probability distributions, representing the probability distributions of the EMG signal and the IMU data, respectively; log represents the logarithmic function with base 2; ∑x ∈X ∑y ∈Y In practice, kernel density estimation is used to obtain the probability distribution, thus avoiding information loss caused by discretization.

[0074] In military training scenarios, such as crawling exercises, mutual information calculation can precisely align the activation pattern of the oblique abdominal muscles with the trunk rotation angle, thereby analyzing the coordination of technical movements. By searching for the time offset corresponding to the maximum mutual information value within a ±20ms range using a sliding window (50ms width, 10ms step length), and then using a cubic spline interpolation algorithm, sub-millisecond alignment can be achieved.

[0075] The alignment quality is evaluated by the Phase Synchronization Index (PSI), which is calculated as follows:

[0076]

[0077] Where PSI represents the phase synchronization index, which ranges from [0,1], and a larger value indicates a higher degree of synchronization; N is the signal length; φ X (k) and φ Y (k) represents the instantaneous phase of the Hilbert transform of the EMG signal and IMU data at time k, respectively; Represents Euler's formula e iθ =cosθ+isinθ, where I is the imaginary unit and |·| represents the modulus of the complex number. In the present invention, the PSI threshold is set to 0.85. Signal segments below this threshold are locally realigned to ensure stable data synchronization throughout the training process.

[0078] The exercise performance evaluation unit 22 is used to calculate the training performance index based on the time-synchronized electromyographic signals and kinematic data. Traditional physical fitness assessments rely primarily on subjective observations or single physiological indicators, which cannot quantify training quality and efficiency. The exercise performance evaluation unit 22 of the present invention establishes a quantitative evaluation system based on multidimensional biomechanical indicators, including the two core dimensions of movement completion and energy consumption efficiency.

[0079] The calculation of movement completion involves three aspects: joint angle compliance, muscle activation sequence correctness, and movement stability index. Joint angle compliance is calculated by comparing the deviation of the actual joint angle sequence with the standard template; muscle activation sequence correctness analyzes the timing characteristics of the activation of major muscle groups; and the movement stability index evaluates the fluctuation range of key joint angles. The formula for calculating movement completion is:

[0080] Movement completion = w1·joint angle conformity + w2·muscle activation sequence correctness + w3·movement stability index,

[0081] Among them, the action completion degree indicates the degree of standardization of the training action, and its value range is [0,1]. The larger the value, the more standardized the action. w1, w2, and w3 are weight coefficients. In the preferred embodiment of the present invention, w1 = 0.5, w2 = 0.3, and w3 = 0.2, and the sum of the weights satisfies w1 + w2 + w3 = 1. The joint angle conformity is calculated as the similarity between the actual joint angle and the standard template, which is defined as where θ i is the actual angle, is the standard template angle, N is the number of sampling points, θ max and θ minThe maximum and minimum values ​​of the joint angles are respectively; the correctness of the muscle activation sequence evaluates the consistency of the muscle activation timing with the standard pattern, which is calculated based on the order of the electromyographic peaks; the movement stability index measures the smoothness of the movement,

[0082]

[0083] Where σ is the standard deviation of the key joint angle, σ threshold is the set threshold. By introducing the max function, it is ensured that when σ exceeds σ threshold When , the lower limit of the motion stability index is 0, rather than a negative value. This ensures that the index value is always within the range of [0,1], meeting the mathematical requirements of the weight coefficient.

[0084] In practical applications, when the joint angle fluctuation σ is small (far less than the threshold σ threshold ), the calculated stability index is close to 1, indicating that the action is very stable; when σ is close to the threshold, the stability index is close to 0, indicating that the action is less stable; when σ exceeds the threshold, the stability index is 0, indicating that the action is unstable.

[0085] For example, in military weighted squat training, joint angle compliance focuses on the maximum knee flexion angle (ideal range 105°±3°) and hip extension angular velocity (around 85° / s); muscle activation sequence focuses on the activation sequence of gluteus maximus-quadriceps femoris-hamstrings. The correct activation sequence can reduce knee joint pressure; movement stability detects fluctuations in the trunk forward tilt angle. The smaller the fluctuation, the stronger the control ability.

[0086] Energy consumption efficiency measures the energy efficiency consumed to complete a specific action. The calculation formula is:

[0087]

[0088] Among them, energy consumption efficiency refers to the mechanical work generated by unit muscle energy consumption, and the unit is J / (μV·s); the mechanical work to complete the task is calculated by multiplying the joint torque and angular displacement, and the unit is J. The calculation formula is: where τ j (t) is the moment of the j-th joint at time t, Δθ j (t) is the corresponding angular displacement increment, J is the number of joints, and T is the number of time sampling points; iEMG i is the integrated electromyographic value of the ith muscle, in μV·s, and is calculated as follows: k represents the integral of the absolute value of the electromyographic signal in the time interval [t1, t2], reflecting the total amount of muscle activity; ik is the energy consumption coefficient per unit of myoelectric value of the i-th muscle, in J / (μV·s), which is determined by muscle type and location. For example, muscles with a high proportion of type I (slow muscle) fibers have higher energy efficiency and correspond to lower k i value; M is the number of muscles monitored; represents the summation of all monitored muscles.

[0089] In military training scenarios, such as weighted marching, energy efficiency is a key metric for evaluating a soldier's technical economy. By calculating the energy efficiency of the major lower limb muscle groups (quadriceps, hamstrings, triceps surae, etc.), the system can identify energy "leakage" points, such as excessive muscle co-contraction or unnecessary compensatory movements, and thus guide technical optimization.

[0090] The training efficiency index is a comprehensive evaluation of movement completion and energy consumption efficiency. The calculation formula is:

[0091] Training efficiency index = movement completion × energy consumption efficiency norm ,

[0092] Among them, the training effectiveness index is a comprehensive evaluation value, reflecting the comprehensive level of training quality and efficiency; the action completion degree is as mentioned above; the energy consumption efficiency norm The normalized energy consumption efficiency is calculated by standardizing the maximum and minimum values:

[0093]

[0094] Among them, energy consumption efficiency min and energy consumption efficiency max They are the minimum and maximum values ​​in the historical data respectively.

[0095] In practical applications, to ensure comparability across different training exercises, this invention uses a maximum-minimum normalization method to map each indicator to the [0, 1] range. For example, for soldiers performing armed cross-country training, the system evaluates the training effectiveness index for key technical movements (such as carrying a load uphill and overcoming obstacles), identifies areas of low performance, and provides targeted technical guidance and training optimization.

[0096] The genetic algorithm optimization unit 23 is used to generate a training parameter optimization plan based on the training effectiveness index using a genetic algorithm with a two-layer parallel architecture. This plan includes nonlinear fluctuating load parameters. Traditional training plan adjustment cycles can take as long as 48-72 hours, making parameter optimization impossible within a single training session. The genetic algorithm optimization unit 23 of the present invention achieves real-time optimization of training parameters, with each iteration time controlled to less than 5 seconds.

[0097] A core innovation of the present invention is the dual-layer parallel architecture, comprising an outer parallel structure and an inner parallel structure. The outer parallel structure is used to simultaneously evaluate multiple candidate training plans. In a preferred embodiment, 50 candidate plans are evaluated simultaneously. The inner parallel structure is used to parallelize the calculation of the muscle-joint coupling matrix during the evaluation of each candidate training plan, significantly improving computational efficiency.

[0098] The muscle-joint coupling matrix C is calculated as:

[0099] C ij =corr(RMS i ,θ j ),

[0100] Among them, C ij is the element in the i-th row and j-th column of the matrix, which represents the correlation coefficient between the RMS value of the i-th muscle and the j-th joint angle; corr(·,·) represents the Pearson correlation coefficient calculation function, and the formula is Where cov(X,Y) is the covariance, σ X and σ Y are standard deviation and RMS respectively. i is the root mean square value of the i-th muscle, and the calculation method is EMG i (k) is the electromyographic signal value of the i-th muscle at time k, N is the number of samples; θ j is the angle value of the jth joint. The dimension of matrix C is M×J, where M is the number of monitored muscles and J is the number of joints.

[0101] In military training scenarios, such as when soldiers perform tactical squat shooting training, the muscle-joint coupling matrix can reveal the key biomechanical characteristics of a stable shooting posture. By analyzing the correlation coefficient between quadriceps RMS and knee flexion angle, the system can determine the optimal squat angle to maintain stability while minimizing muscle fatigue.

[0102] The training parameters are optimized using a real-number coded genetic algorithm. Each chromosome contains the following genes: load (kg), rest time between sets (s), movement speed (times / min), number of sets, and number of repetitions per set. The chromosome coding format is:

[0103] chromosome=[W,T rest ,V,S,R],

[0104] Among them, chromosome represents a chromosome, representing a complete set of training parameter schemes; W is the training weight, in kg; T rest is the rest time between sets, in seconds; V is the movement speed, in times / minute; S is the number of training sets; R is the number of repetitions in each set.

[0105] The fitness function uses the training efficiency index. The algorithm process includes initial population generation, fitness evaluation, selection, crossover, mutation, and population update. The selection operation uses the roulette wheel method, with the probability of an individual being selected proportional to its fitness. The crossover operation uses simulated binary crossover (SBX) to generate two new individuals. The mutation operation uses polynomial mutation to randomly change the values ​​of certain genes in the chromosome.

[0106] The present invention also introduces a dynamic constraint mechanism to dynamically adjust the range of training parameters according to the fatigue index. The fatigue index is calculated based on the downward shift of electromyographic frequency and the change of amplitude:

[0107]

[0108] Among them, FI represents fatigue index, and its value range is [0,1]. The larger the value, the higher the fatigue level. initial and MPF current The median frequencies of the electromyographic signals in the initial and current states are in Hz. A decrease in the median frequency reflects muscle fatigue. RMS initial and RMS current are the root mean square values ​​of the electromyographic signals in the initial state and the current state, respectively. An increase in RMS indicates that more motor units are required to maintain the same force output. α and β are weight coefficients. In the preferred embodiment of the present invention, α = 0.7 and β = 0.3, reflecting the dominant role of median frequency changes in fatigue assessment.

[0109] In military training scenarios, such as continuous tactical physical training, the fatigue index accurately reflects a soldier's fatigue state. When the system detects that the fatigue index of certain muscle groups (such as the forearm muscles in a gun grip) exceeds a threshold (typically set at 0.4), the load change range is adjusted from [-5kg, +5kg] to [-5kg, +2kg], reducing the maximum increase and preventing overtraining that can lead to technical breakdown and increased injury risk.

[0110] In addition, the present invention also uses an adaptive mutation rate to dynamically adjust according to population diversity:

[0111] P m =P m0 +0.1·(1-D),

[0112] Among them, P m Indicates the mutation rate of actual application; P m0 is the basic mutation rate, set to 0.05; D is the current population diversity, calculated as where σ population is the average standard deviation of the population parameter, σmax is the theoretical maximum standard deviation; (1-D) represents the degree of insufficiency of population diversity; the larger the value, the lower the diversity.

[0113] In practical applications, such as when designing strength training programs for special forces, the adaptive mutation rate mechanism can automatically increase the mutation rate when training programs converge (for example, most programs are concentrated in a narrow parameter range), increase the parameter search range, avoid falling into local optimal solutions, and discover more innovative training models.

[0114] The nonlinear fluctuation load strategy is another innovation of this invention, which breaks through the traditional linear incremental training load model. For example, in squat training, the system may generate the following training parameter combination:

[0115] Group 1: 5kg weight, 60s rest, speed 1 time / s, repeat 10 times;

[0116] Group 2: 3kg weight, 45s rest, 1.5 reps / s, 15 repetitions;

[0117] Group 3: 7 kg weight, 75 seconds rest, 0.7 times / second speed, 8 repetitions;

[0118] This fluctuating loading pattern aligns with the physiological mechanism of "overload-unload recovery-supercompensation," effectively activating type II muscle fibers and enhancing training effectiveness. In military training, this strategy is particularly well-suited for rapid strength training, such as accelerating during tactical sprints. By interspersing high-speed, light-load combinations within the training sequence, the system effectively enhances neuromuscular recruitment efficiency and explosive power without overly fatigued soldiers.

[0119] The individualized training path planning unit 24 is used to generate a personalized training path based on the individual feature vector and training response pattern. Traditional training ignores the biological differences in individual soldiers' muscle recruitment ability and joint mobility. The present invention achieves deep personalization of training programs through detailed individual modeling.

[0120] The individual feature vector includes six dimensions: muscle fiber type distribution characteristics, joint mobility characteristics, muscle strength ratio, movement learning curve parameters, fatigue recovery mode parameters and injury risk factors, a total of 24 parameters, which can be expressed as:

[0121] F=[F fiber ,F ROM ,F strength ,F learning ,F recovery ,F risk ],

[0122] Where F represents the individual feature vector; Ffiber is the muscle fiber type distribution feature vector, which contains 4 parameters and estimates the ratio of fast and slow muscles through electromyographic spectrum analysis, such as F fiber =[r quad ,r ham ,r calf ,r glute ], where r represents the fast-twitch ratio of each muscle group; F ROM is the joint range of motion feature vector, which contains 6 parameters and measures the range of motion and flexibility of each major joint; F strength is the muscle strength ratio subvector, which contains 4 parameters and evaluates the strength balance between the main muscle groups, such as the strength ratio of the quadriceps and hamstrings; learning is the sub-vector of the action learning curve, which contains 3 parameters and analyzes the speed and accuracy of mastering new actions; F recovery F is the fatigue recovery mode subvector, which contains 4 parameters and evaluates the recovery time curve after training of different intensities; risk It is an injury risk factor vector containing three parameters to identify historical injuries and biomechanical risk points.

[0123] In the context of military training, individual eigenvectors can capture key biological differences between soldiers. For example, some soldiers may have a higher proportion of fast-twitch muscle fibers (advantageous for explosive power training), while others may have superior recovery ability (suited for high-frequency training). By comprehensively measuring these 24 parameters, the system can create a detailed "athletic gene profile" for each soldier, providing the basis for personalized training.

[0124] Based on these characteristics, the present invention uses the K-means++ clustering algorithm to classify soldiers into five types of training response patterns: rapid adaptation, strength advantage, endurance advantage, technology-oriented, and balanced development. The objective function of the K-means++ algorithm is:

[0125]

[0126] Where J is the objective function, which represents the sum of the squares of the distances from all sample points to the cluster centers to which they belong; K is the number of clusters, and in this invention, K=5; N j is the number of samples in the jth cluster; represents the feature vector of the i-th sample belonging to the j-th cluster; μ j is the center of the j-th cluster; ||·|| represents the Euclidean distance.

[0127] The system will generate a personalized training path based on the classification results and training goals. The algorithm for generating a personalized training path can be expressed as:

[0128] TP=TP base +K·(FFref ),

[0129] Among them, TP represents the personalized training path, which includes a series of training parameters and advanced rules; TP base is the basic training template, selected according to the training objectives and soldier categories; K is the individual adjustment coefficient matrix, which determines the degree of influence of different feature deviations on the training parameters; F is the individual feature vector of the current soldier; F ref is the reference feature vector, usually taken from a standard template or the average value of a similar group; (FF ref ) represents individual characteristic deviation, reflecting the difference between the current soldier and the reference standard.

[0130] In military training practice, such as comprehensive physical training for special forces, individualized training path planning can customize the most suitable training program according to the characteristics of soldiers. Taking squat training as an example, if a soldier has strength advantage but lacks hip flexibility (F ROM If the range of motion of the hip joint is lower than the standard), the system will add preparatory movements for hip joint mobility (such as dynamic hip extension before squatting) based on the standard squat template. At the same time, it will adjust the training load curve and strengthen the eccentric contraction phase (reduce the speed and increase the load) to give full play to its strength advantage and make up for the flexibility shortcomings in a targeted manner.

[0131] Finally, the structure and function of the decision execution layer 3 are described in detail. Figure 4 As shown, the decision execution layer 3 includes an augmented reality interaction interface 31 , an adaptive resistance control module 32 and a biofeedback intervention unit 33 .

[0132] The augmented reality interactive interface 31 is used to project biomechanical feedback in real time. Traditional training feedback mainly relies on verbal guidance from the coach, which has large delays and limited effects. The present invention projects biomechanical feedback in real time through a head-mounted display device, such as using a color gradient to indicate the risk of knee valgus, and provides voice prompts (such as "lower the center of gravity 2cm to reduce lumbar shear force"), which greatly improves the accuracy of training guidance. In military training scenarios, such as when conducting tactical weighted sprint training, the augmented reality interface can display a biomechanical load distribution map in real time, marking the joints that are subjected to excessive force (for example, a red warning is displayed when the force on the inner side of the knee joint exceeds the safety threshold of 600N). At the same time, the system provides immediate corrective suggestions, such as "increase the cadence to reduce impact force" or "adjust the gait to reduce the load on the knee joint", to help soldiers optimize their movement techniques while maintaining speed and reduce the risk of injury.

[0133] The adaptive resistance control module 32 is used to adjust the training load according to the training parameter optimization scheme. In a preferred embodiment of the present invention, the adaptive resistance control module 32 is connected to an intelligent weight-bearing vest, and the resistance adjustment resolution reaches 0.5kg. It can automatically adjust the external load during the rest period between groups according to the output results of the genetic algorithm, and supports a nonlinear incremental strategy. In practical applications, for example, when conducting tactical stair climbing and descending training, the system will adjust the weight of the weight-bearing vest in real time according to the performance of the soldiers. When it is detected that the standard of technical movements has decreased (such as decreased pelvic stability), the system will automatically reduce the weight by 0.5-1.5kg before the next set of training; when the soldier performs well and the fatigue level is low, the weight is increased to provide sufficient training stimulation, thereby achieving precise and personalized control of the training load.

[0134] The biofeedback intervention unit 33 is used to trigger intervention measures when abnormal biomechanical patterns are detected. The biofeedback intervention unit 33 includes an electrical stimulation module that, when abnormal joint angles are detected, applies electrical stimulation at a specific frequency to the relevant muscles to suppress abnormal compensation. For example, if the system detects a sudden increase in the internal rotation angle of the knee joint (+5°), the electrical stimulation module will immediately trigger the application of 80Hz pulses to the gracilis muscle to suppress abnormal compensation and prevent potential injury.

[0135] The electrical stimulation parameters were set as follows:

[0136] Stimulus parameters = {f, A, D, T},

[0137] Among them, f is the stimulation frequency, the unit is Hz, and the range is 20-120Hz. Low frequency (20-40Hz) is used for the recovery of fatigued muscles, medium frequency (40-80Hz) is used for muscle activation, and high frequency (80-120Hz) is used to inhibit abnormal compensation; A is the stimulation amplitude, the unit is mA, and the range is 5-5OmA, which is adjusted according to muscle location and individual sensitivity; D is the pulse width, the unit is μs, usually set to 200-400μs; T is the stimulation duration, the unit is ms, and acute intervention is usually 200-500ms.

[0138] In the context of military training, such as when performing weighted push-ups, if the system detects outward rotation of the inferior angle of the scapula ("winged scapula", a precursor to shoulder injury), it will immediately trigger electrical stimulation of the serratus anterior muscle (frequency 80Hz, amplitude 20mA, pulse width 300μs, duration 300ms), activating the muscle, correcting the scapula-thoracic relationship, and preventing rotator cuff injuries.

[0139] In practical applications, the biofeedback intervention unit 33 has a multi-level intervention mechanism:

[0140] Level 1 intervention: Visual warnings and voice guidance are provided through the AR interface. This is applicable to minor technical deviations, with an intervention intensity coefficient of 0.3.

[0141] Level 2 intervention: Automatically adjust the training load (such as reducing weight or speed), suitable for moderate technical breakdown, with an intervention intensity coefficient of 0.6;

[0142] Level 3 intervention: triggering electrical stimulation for muscle regulation, suitable for severe biomechanical abnormalities, with an intervention intensity coefficient of 1.0.

[0143] The system will automatically select the appropriate intervention level based on the severity of the abnormality to ensure training safety. The abnormality severity calculation formula is:

[0144]

[0145] Among them, the abnormality severity is the normalized abnormality degree, and the value range is [0,1]; v is the current measurement value, such as joint angle or muscle activation ratio; v threshold is the warning threshold, which represents the upper limit of the biomechanical parameter. If this value is exceeded, intervention will be triggered. critical It is the danger threshold, indicating the limit value that may cause acute injury.

[0146] The working principle of this formula is: when the measured value v does not exceed the warning threshold v threshold When the measured value v is between the warning threshold and the danger threshold, the severity of the abnormality is proportional to the degree of exceeding the warning threshold and increases linearly; when the measured value v reaches or exceeds the danger threshold v critical When , the abnormal severity is 1, indicating an extremely dangerous state.

[0147] Based on the calculated severity value of the anomaly, the system selects the corresponding intervention level:

[0148] When the severity of the abnormality is between 0.1 and 0.3, a level 1 intervention (visual warning and voice guidance) is triggered;

[0149] When the severity of the abnormality is between 0.3 and 0.7, the second level intervention (automatic adjustment of training load) is triggered;

[0150] When the severity of the abnormality exceeds 0.7, the third level of intervention (electrical stimulation of muscle regulation) is triggered;

[0151] For example, for knee valgus angle, v threshold Usually set to 10°,v critical The setting is 20°. When the valgus angle reaches 15°, the severity level is 0.5, and the system selects the secondary intervention measure, automatically reducing the training load and correcting the posture through voice prompts. This graded intervention mechanism ensures that the system takes appropriate measures based on the degree of risk, avoiding excessive intervention and ensuring training safety.

[0152] Next, the method embodiment of the present invention is described. The present invention also provides a method for dynamic assessment and intelligent reinforcement training of individual soldiers' physical fitness, such as Figure 9 As shown, the following steps are included:

[0153] Step S1: Collect EMG signals and kinematic data during training. Specifically, a high-density EMG sensor array is used to collect muscle activation data with a sampling rate of 2000 Hz; a nine-axis IMU fusion unit is used to collect joint kinematic parameters with an output frequency of 100 Hz. The collected raw data is preprocessed, including filtering, denoising, and normalization.

[0154] Step S2: Time-synchronize the electromyographic signal with the kinematic data to obtain time-synchronized data. Specifically, a method combining dynamic time warping and the maximum mutual information criterion is used to control the time synchronization error within 2 milliseconds.

[0155] Step S3: Based on the time-synchronized data, the movement completion and energy consumption efficiency are calculated. Movement completion takes into account joint angle compliance, muscle activation sequence correctness, and movement stability index; energy consumption efficiency calculates the ratio of mechanical work to complete the task to muscle energy consumption.

[0156] Step S4: Generate a training effectiveness index based on the degree of movement completion and energy consumption efficiency. The training effectiveness index is the product of the degree of movement completion and the energy consumption efficiency, and is standardized to ensure comparability between different training movements.

[0157] Step S5: A genetic algorithm with a two-layer parallel architecture generates a training parameter optimization plan based on the training effectiveness index. This plan includes nonlinear fluctuating load parameters. The two-layer parallel architecture includes outer-layer parallelism (simultaneously evaluating multiple candidate training plans) and inner-layer parallelism (parallel calculation of the muscle-joint coupling matrix). The system also applies a dynamic constraint mechanism to adjust the parameter variation range based on the fatigue index.

[0158] Step S6: Execute training control according to the training parameter optimization plan and trigger intervention measures when abnormal biomechanical patterns are detected. Specifically, an augmented reality interface provides visual feedback, an adaptive resistance control module adjusts the training load, and an electrical stimulation module regulates specific muscles when necessary.

[0159] The following describes the implementation process of the present invention in detail using squat training as a specific example:

[0160] During the baseline assessment phase, soldiers wore the sensor device and completed a standard deep squat (initial load: 20 kg, speed: 1 rep / second). The system collected the following data: EMG characteristics—the integrated EMG value of the vastus lateralis was 120 μV·s, and the co-contraction ratio of the biceps femoris to rectus femoris was 1:0.8; joint angles—the maximum knee flexion angle was 112° (ideal range: 105°±3°), and the hip extension angular velocity was 85° / s. Through analysis, the system identified excessive knee flexion angles, indicating a risk of excessive squatting. During this phase, the system also measured key parameters, including the fast twitch muscle ratio of the quadriceps femoris (approximately 55%, estimated from the median EMG spectrum frequency of 68 Hz) and hip range of motion (maximum flexion angle of 125°, slightly below the standard 130°), providing foundational data for subsequent personalized training.

[0161] During the genetic algorithm optimization phase, the system randomly generated 50 sets of training parameters (load range 18-22 kg, rest time 45-75 seconds between sets) as the initial population. Each parameter set's chromosome is represented as a 5-dimensional vector, for example, [20 kg, 60 seconds, 1 rep / s, 4, 10], representing a 20 kg load, 60 seconds rest time, 1 rep / s frequency, 4 sets of 10 repetitions. The system evaluated these 50 sets using a two-layer parallel architecture: the outer layer simultaneously evaluated the training efficacy index of all 50 training options; the inner layer calculated an 8×3 muscle-joint coupling matrix (coupling relationships between eight major muscle groups and three key joints) during each evaluation. After 20 generations of evolution (approximately 100 seconds in total), the optimal solution was identified: a load of 19 kg and a rest time of 60 seconds, expected to improve quadriceps eccentric contraction efficiency by 23%. During the optimization process, the system applies a dynamic constraint mechanism. When it detects that the fatigue index reaches 0.36 (close to the 0.4 threshold), it automatically adjusts the weight change range from [-5kg, +5kg] to [-5kg, +3kg] to avoid the risk of overtraining.

[0162] Real-time training monitoring phase: During the third training session, the system detected a sudden increase in knee internal rotation (from a baseline of 3° to 8°, a change exceeding 5°). It immediately calculated the severity of the abnormality as 0.4 (with a warning threshold of 5° and a danger threshold of 15°), triggering secondary intervention measures. First, a knee internal rotation risk warning (marked in red) was displayed via the AR interface, accompanied by a voice prompt: "Push your knee outward and maintain hip external rotation." Simultaneously, the electrical stimulation module applied 80Hz, 25mA, 300μs pulse width, and a duration of 300ms to the gracilis muscle, effectively suppressing abnormal compensation. In subsequent sessions, the system automatically reduced the load by 1kg (from 19kg to 18kg) and slowed the movement speed from 1 to 0.8 times per second, emphasizing movement stability and technical quality.

[0163] Training effectiveness evaluation phase: After one month of training, soldiers' squat technique improved significantly: knee flexion angle adjusted to 105° (within the standard range), quadriceps eccentric contraction efficiency increased by 19% (close to the expected 23%), quadriceps-hamstring co-contraction ratio optimized to 1:0.9 (closer to the ideal 1:1), and squat 1RM (maximum repetitions) increased by 8%, all without any injuries. The system also recorded changes in individual soldier characteristics: fast-twitch muscle recruitment increased by 12% (reflected in increased EMG amplitude and shortened latency), and joint stability index increased by 18% (reflected in improved movement trajectory consistency). These improvements directly translated into improved battlefield-related capabilities: 30-meter loaded sprint time was reduced by 0.4 seconds, and the vertical height of a 40kg squat jump increased by 3.5cm.

[0164] In summary, the individual soldier dynamic fitness assessment and intelligent reinforcement training system and method provided by this invention establish an advanced intelligent training system through the precise acquisition and synchronization of multimodal sensory data, quantitative evaluation of multidimensional biomechanical indicators, real-time genetic algorithm optimization, nonlinear fluctuating load strategy, and individualized training path planning. This system addresses the problems of traditional training, such as low standardization, strong subjective dependence, single evaluation dimension, static training plans, and delayed injury warning. It significantly improves training efficiency, safety, and soldier combat capabilities, and has broad application prospects.

[0165] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be included within the scope of protection of the present invention.

Claims

1. The system for dynamic assessment and intelligent reinforcement training of individual soldiers is characterized by: include: Hardware perception layer, used to collect electromyographic signals and kinematic data during training; An intelligent analysis layer, connected to the hardware perception layer, for receiving and processing the electromyographic signals and the kinematic data, and generating a training parameter optimization solution; as well as A decision execution layer, connected to the intelligent analysis layer, configured to receive the training parameter optimization plan and execute training control; The intelligent analysis layer includes: a multimodal spatiotemporal alignment unit, configured to synchronize the myoelectric signal with the kinematic data; a motion effectiveness evaluation unit, configured to calculate a training effectiveness index based on the time-synchronized electromyographic signals and kinematic data; A genetic algorithm optimization unit is used to generate the training parameter optimization scheme based on the training effectiveness index using a genetic algorithm with a double-layer parallel architecture, wherein the training parameter optimization scheme includes a nonlinear fluctuating load parameter.

2. The system according to claim 1, wherein: The hardware perception layer includes: A high-density myoelectric sensor array is used to collect muscle activation data and obtain the myoelectric signal; A nine-axis IMU fusion unit, used to collect joint kinematic parameters and obtain the kinematic data; and A data preprocessing unit is used to filter and normalize the electromyographic signal and the kinematic data.

3. The system according to claim 2, characterized in that The high-density electromyography sensor array includes an 8-channel wireless sEMG module with a sampling rate of 2000Hz and a common mode rejection ratio of not less than 110dB; the nine-axis IMU fusion unit integrates a three-axis accelerometer, a gyroscope and a magnetometer, with an output frequency of 100Hz.

4. The system according to claim 1, wherein: The multimodal spatiotemporal alignment unit adopts a method combining dynamic time warping and maximum mutual information criterion to control the time synchronization error between the electromyographic signal and the kinematic data within 2 milliseconds.

5. The system according to claim 1, wherein: The exercise performance evaluation unit is used for: Calculating movement completion, including joint angle compliance, muscle activation sequence correctness, and movement stability index; Calculating energy consumption efficiency, where the energy consumption efficiency is the ratio of mechanical work to muscle energy consumption to complete the task; as well as The training effectiveness index is generated based on the movement completion degree and the energy consumption efficiency.

6. The system according to claim 1, wherein: The two-layer parallel architecture of the genetic algorithm optimization unit includes: An outer parallel structure for simultaneously evaluating multiple candidate training solutions; and The inner parallel structure is used to parallelly calculate the muscle-joint coupling matrix during the evaluation of each candidate training program; The genetic algorithm optimization unit further includes a dynamic constraint mechanism for dynamically adjusting the range of training parameter changes according to the fatigue index.

7. The system according to claim 1, wherein: The intelligent analysis layer also includes: An individualized training path planning unit, configured to generate an individualized training path based on individual feature vectors and training response patterns; The individual feature vector includes muscle fiber type distribution characteristics, joint mobility characteristics, muscle strength ratio, movement learning curve parameters, fatigue recovery mode parameters and injury risk factors.

8. The system according to claim 1, wherein: The decision execution layer includes: Augmented reality interface for projecting biomechanical feedback in real time; an adaptive resistance control module, configured to adjust the training load according to the training parameter optimization scheme; and A biofeedback intervention unit is used to trigger intervention measures when abnormal biomechanical patterns are detected.

9. The system according to claim 8, characterized in that The biofeedback intervention unit includes an electrical stimulation module, which is used to apply electrical stimulation of a specific frequency to the relevant muscles to inhibit abnormal compensation when abnormal joint angles are detected.

10. A method for dynamic assessment of individual soldier physical fitness and intelligent reinforcement training, applied to the system according to any one of claims 1 to 9, characterized in that: The following steps are involved: Collect electromyographic signals and kinematic data during training; Time-synchronizing the electromyographic signal with the kinematic data to obtain time-synchronized data; Calculating action completion and energy consumption efficiency based on the time synchronization data; generating a training effectiveness index based on the movement completion degree and the energy consumption efficiency; A genetic algorithm with a two-layer parallel architecture is used to generate a training parameter optimization scheme based on the training effectiveness index, wherein the training parameter optimization scheme includes a nonlinear fluctuating load parameter; as well as Training regulation is performed according to the training parameter optimization plan, and intervention measures are triggered when abnormal biomechanical patterns are detected.

Images