Optimisation Strategies and Mathematical Modelling of the Path to Improvement of Students’ Physical Fitness Levels in Higher Education Physical Education

College students’ physical fitness not only affects the future employment development of college students but also is an important part of China’s education reform and development and has a profound impact on the cultivation of comprehensive technical talents. Based on this, based on the significance of college students’ physical fitness enhancement, we analyse the core influencing factors of college students’ physical fitness, formulate strategies for college students’ healthy physical fitness enhancement, and establish a mathematical model of college students’ physical fitness change, so as to help colleges and universities to better enhance college students’ healthy physical fitness [1–3].

Physical fitness training plays an important role in developing students’ specialised sports skills and improving sports performance [4]. It not only enables students to establish a solid sports foundation but also inspires their love and passion for sports, develops their competitive awareness and teamwork spirit, and leads students through simple to complex sports training methods, thus improving their sports ability [5–8]. Through systematic physical training, students’ physical fitness can be gradually improved to support their performance in specific sports [9–11]. In addition, there is a close relationship between the physical state of college students and their learning efficiency, and good physical fitness can help college students better cope with the pressure of study and improve their learning efficiency [12–14]. Therefore, physical fitness training is undoubtedly an important part of students’ development and growth in college physical education. Explore better physical training methods, constantly promote the content and training methods of physical training in depth to ensure that the physical training improves the physical quality of students in college physical education at the same time, but also to improve their athletic performance, so as to improve the athletic ability [15–18]. It is necessary for sports workers to seriously think and explore in order to master the scientific and systematic training enhancement methods and apply them to actual training [19–21].

Good physical fitness level determines the height of one’s future career and study, and physical education in colleges and universities has become extremely important for improving students’ physical fitness level and health. The article proposes a physical function sensing device + intelligent data collection in the physical fitness diagnosis process of college physical education with personalised physical fitness tests and longitudinal tracking assessments. The main method is to use the LSTM model to acquire data on students’ physical fitness training in different time series. Secondly, the wavelet transform algorithm is used to extract the dynamic features of students’ physical fitness training data, the dynamic feature data are input into the CNN model to achieve the accurate identification of physical fitness training features, and the optimised output model of students’ physical fitness training monitoring data is constructed by combining with the correlation dimension retrieval method. Finally, a quantitative analysis of the data was carried out to investigate the effectiveness of wavelet transform, CNN and real-time monitoring architecture of physical training, and based on the analysis results, an optimisation strategy of students’ physical training was designed from the core strength and specialised physical training, which provides auxiliary decisionmaking for enhancing students’ physical fitness level.

2

Dynamic monitoring of physical training of physical education students in higher education institutions

The optimisation of the teaching mode of physical education in colleges and universities is an important means of promoting the comprehensive development of student’s physical fitness and ability, and it has a greater significance and influence. We need to improve the content of training, cultivate students’ sense of autonomy, pay attention to fun training, etc., and deeply understand the importance of physical training in physical education in order to lay an important foundation for the cultivation of students’ good physical fitness level, physical fitness and potential stimulation.

2.1

Diagnostic process of students’ fitness level

The physical fitness level of students in college physical education is often very personalised, and only through personalised training based on students’ characteristics and problems can the physical fitness level of students in sports be effectively improved.

First of all, it is necessary to conduct personalized physical diagnoses of students to accurately assess their physical condition, athletic ability, and physical fitness level. The systematic testing of students to analyse their physical strengths and weaknesses and their level of athletic ability, to help physical education teachers to clarify where to start for students to carry out sports training, and to improve the relevance of physical training programmes.

Secondly, on the basis of personalised diagnosis of students’ physical fitness levels, a longitudinal tracking and assessment mechanism is established to make an objective assessment of students’ training effects and accordingly determine whether adjustments need to be made to physical training.

Physical education teachers provide students with scientific and rigorous personalised physical fitness diagnoses, which can promote trust and respect between students and physical education teachers, and teachers and students can directly experience the changes brought about by physical fitness training through testing. From the perspective of modern technology, the combination of technology and physical education training can achieve real-time monitoring of students’ physical training through the data to assist physical education teachers in further optimize the physical training programme in physical education and guide to enhance the physical fitness level of students in college physical education.

2.2

Framework for dynamic monitoring of physical training

Through the effective monitoring of students’ physical training, obtaining students’ heart rate, respiratory rate and other information after training can help teachers to adjust and correct students’ physical training, give students targeted training methods, and thus help students to improve the efficiency of physical training and physical fitness level. This paper develops a framework to monitor students’ physical training in real-time for physical education purposes in colleges and universities, as demonstrated in Figure 1 [22]. Its hardware part mainly includes six parts: the MCU minimum system, six-axis sensor MPU, heart rate monitoring sensor, digital-to-analogue conversion module, display circuit, and Bluetooth circuit.

The MCU minimum system selects the STM35 model microcontroller as the main control chip, which is the core control board in the physical fitness monitoring system. The six-axis sensor is primarily comprised of an acceleration sensor and a gyroscope sensor. The role of the sensor is to collect motion information for pre-processing operations, i.e., to reduce the load on the processor and other effective processing. The heart rate sensor selects the photoelectric reflective sensor, connects it to the STM chip using the conversion function pin, and then the heart rate signal can be converted. The CC2545 chip is chosen for the Bluetooth circuit design due to its ability to communicate at multiple rates in the 2.4GHz band. The pin connection is 24 general purpose and the storage size is 8kB, which can meet the system’s low-power wireless transmission requirements.

2.3

Immediate data collection for physical training

In students’ physical training, timely information related to training plays an important role in improving the effectiveness and quality of training. Aiming at the immediate collection of students’ physical training data in college physical education, this paper constructs a physical training data collection model using a long short-term memory neural network (LSTM). The specific steps are as follows: Step 1

For the observation points that need to collect data instantly, first collect historical statistical data from the observation points for multiple time periods, and then normalise the statistical data of each time period according to the following formula, so as to standardise the time series. Then: 1 $z_{n} = \frac{x_{n} - x_{\min}}{x_{\max} - x_{\min}}$ Where n denotes the number of dimensions, and x_max, x_min denotes the maximum and minimum values of a certain set of dimensions, respectively.

Step 2

In a one-dimensional standardised one-dimensional array, a sliding window is introduced to acquire multiple training sample sizes, each of which is consistent with the statistical data of the previous time point, and is treated as a labelled training sample, each of which contains a large amount of data at different time points.

Step 3

In order to improve the accuracy of the data acquisition model, a weighted beta distribution method with a continuous probability distribution is applied to calculate the mean of the beta distribution of the weighting parameters of the sample features. Then: 2 $d_{k} = λ_{k} \sqrt{1 - b^{2}} + {\bar{λ}}_{k} (1 - \sqrt{1 - b^{2}})$ Where λ_k, ${\bar{λ}}_{k}$ denotes the weighting parameter and mean value after the k nd calculation respectively, and b denotes the beta distribution constant.

Step 4

Construct the data acquisition model input gate of the long and short-term memory network to filter invalid data and add useful data to the long and short-term memory unit. The input gate sample data can be expressed as: 3 $ε (ω_{x k} x_{k} + ω_{h k} h_{k - 1} + ω_{g k} g_{k - 1} + a_{k})$ Where ω_xk, ω_hk, ω_gk represents the weight coefficients between the input gate and feature data, the input gate and hidden state, and the input gate and training state, respectively, ε represents the activation function coefficients, h_k–1, g_k–1 represents the hidden state and training state, respectively, a_k represents the input gate bias term, and x_k represents the input data.

Step 5

Construct the forgetting gate and delete the invalid data in the memory unit. Then: 4 $E_{k} = ε (ω_{x k}^{'} x_{k} + ω_{h k}^{'} h_{k - 1} + ω_{g k}^{'} g_{k - 1} + b_{k})$ Where $ω_{x k}^{'}$ , $ω_{h k}^{'}$ , $ω_{g k}^{'}$ denotes the weight coefficients between the forgetting gate and the feature data, the forgetting gate and the hidden state, and the forgetting gate and the training state, respectively, and b_k denotes the forgetting gate bias term.

Step 6

Construct the output gate, i.e., the output of the data acquisition model based on the long and short-term memory network. Then: 5 $O_{k} = ε (ω_{x k}^{''} x_{k} + ω_{h k}^{''} h_{k - 1} + ω_{g^{k}}^{''} g_{k - 1} + c_{k}) + λ_{k} x_{k}$ Where $ω_{x k}^{''}$ , $ω_{h k}^{''}$ , $ω_{g k}^{''}$ denotes the weight coefficients between the output gate and the feature data, the output gate and the hidden state, and the output gate and the training state, respectively, and c_k denotes the forgetting gate bias term. The data acquisition model constructed by the long and shortterm memory network is able to update the model parameters in real time and has the function of model fine-tuning.

To enable selective training of the model, a focused control mechanism based on long short-term memory is designed using the data acquisition model. Using the control vector and the gating control unit of the gating mechanism, the current input is selectively controlled according to the control vector. In the feature level focused attention calculation, the input feature values are composed of entity features, and the data use a bilinear model to select the focused control scoring function to obtain the final training results.

3

Optimisation of dynamic feature extraction and recognition of student fitness data

Physical training is an important part of physical education in colleges and universities. Taking physical fitness improvement as an important training direction to optimise and innovate physical education classroom teaching practices can highlight the value and role of physical education classroom teaching, provide good support for students’ vocational physical fitness strengthening, and effectively improve the physical fitness level of students in colleges and universities, as well as achieve the overall development of students. Relying on the dynamic real-time monitoring of students’ physical training, the analysis of students’ physical training data from the perspective of physical fitness level improvement can provide scientific and effective guidance for the development of physical education and teaching activities in colleges and universities.

3.1

Dynamic feature extraction for physical fitness data

3.1.1

Wavelet transform algorithm

Wavelet transform is a time-frequency domain analysis method with a fixed window area. Wavelet transform is used to obtain different wavelet coefficients by scaling and translating the mother wavelets. Generally, the translation is used to obtain the time information of the data, while the scaling is used to obtain the frequency information of the data [23].

Wavelet transform mainly discusses the existence of a signal f(t) in the space L²(R) consisting of square productable functions satisfying the following conditions: 6 $f (t) \in L^{2} (R) \Leftrightarrow \int_{R} | f (t) |^{2} d t < \infty$

Then f(t) is said to be an energy limited signal.

If ψ(t) ∈ L² (R), its Fourier transform $\hat{φ} (ω)$ satisfies the tolerance condition: 7 $C_{ψ} = \int_{- \infty}^{+ \infty} | ω |^{- 1} | \hat{ψ} (ω) |^{2} d ω < \infty$

Then ψ(t) is called permissible wavelet, mother wavelet function or basis wavelet, and the fulfilment of the above permissible conditions indicates that this wavelet has zero mean and volatility. The wavelet function ψ_a,b(t) is obtained by stretching and translating the mother wavelet ψ(t): 8 $ψ_{a, b} (t) = \frac{1}{\sqrt{| a |}} ψ (\frac{t - b}{a}) (a, b \in R, and a \neq 0)$ Where ψ_a,b(t) is said to be the wavelet basis function, a denotes the scale factor, which reflects the scale of the wavelet function, i.e., the width of the wavelet function, and b is the translation factor, which is usually used to detect the translation distance of the wavelet function on the time axis.

The continuous wavelet transform of the data f(t) to be processed is the unfolding of the data f(t) in any L² space under the selected wavelet basis function, corresponding to the mathematical expression: 9 $W T_{f} (a, b) = 〈 f, ψ_{a, b} 〉 = \frac{1}{| a |} \int_{- \infty}^{+ \infty} f (t) \bar{ψ (\frac{t - b}{a})} d t a (a \neq 0)$

The left side of the above equation represents the wavelet transform coefficients obtained after the wavelet transform. The wavelet basis function has both scale factor and translation factor, so the data in the time domain is equivalent to the projection onto the time one-scale two-dimensional plane after the wavelet transform.

The inverse transform exists for wavelet functions that satisfy the tolerance conditions, and the inverse transform formula for its continuous wavelet is: 10 $f (t) = \frac{1}{c_{ψ}} \int_{- \infty}^{+ \infty} \int_{- \infty}^{+ \infty} [(W_{ψ} f) (a, b)] \cdot ψ_{a, b} (t) \frac{d a d b}{a^{2}}$

One of them is $C_{ψ} = 2 π \int_{- \infty}^{+ \infty} | \hat{ψ} (ω) |^{2} | ω |^{2} d ω$ .

In practical application analysis, in order to facilitate computer analysis and processing, the data signals f(t) need to be discretised. Continuous time data signals into a discrete sequence before the computer can be used to discrete wavelet transform. The main difference with the continuous wavelet transform is that the scale factor a and translation factor b are discrete. The actual detection of ground-penetrating radar data is time-discrete data. Therefore, the discrete wavelet transform is used to process it, and its mathematical formula is as follows: 11 $W T_{f^{j}} (2^{j}, 2^{j} k) = 〈 f (t), ψ_{2^{j}, 2^{j} k,} (t) 〉 = 2^{\frac{j}{2}} \int_{- \infty}^{+ \infty} f (t) \bar{ψ (\frac{t - 2 J k}{2 J})} d t$

The inverse discrete wavelet transform is: 12 $f (t) = \sum_{a, b} (W_{ψ} f) (a, b) ψ_{2^{j}, 2^{j} k} (t)$

After the wavelet transform data f(t) from the time domain to the scale - displacement domain, so the data can be processed and analysed in the time-frequency domain to distinguish the characteristics of the effective signal and noise, and then effectively remove the noise.

The wavelet transform threshold denoising principle in the time-frequency domain is based on the wavelet coefficients obtained from the wavelet transform of the data, a threshold is set to exclude the wavelet transform coefficients corresponding to the noise signal, and retain the wavelet transform coefficients corresponding to the effective reflected waveform signals, so as to carry out the inverse wavelet transform and reconstruct the data.

3.1.2

Dynamic feature extraction

In the monitoring of physical training data in college physical education, this paper uses wavelet transforms to dynamically extract features from students’ physical training data. Since each student has different physical functions, to achieve the constraint control and autocorrelation constraint detection of all students’ physical training data, it is necessary to develop an analysis method for different groups of physical functions, and to explore the deep-seated features of each student’s training data [24].

Given a physical fitness training data m(t), m(t) the continuous wavelet variation is: 13 $W T m (p, q) = \frac{1}{\sqrt{p}} \int_{- \infty}^{+ \infty} m (t) ψ \cdot (\frac{t - q}{p}) d t$ Where, p(p > 0), q denotes the scale parameter and displacement parameter of the training data, respectively, $ψ \cdot (\frac{t - q}{p})$ denotes the sub-wavelet, which is obtained from the mother wavelet ψ(t) after scaling and displacement, and dt denotes the wavelet transform coefficients.

In the wavelet transform process, scaling can obtain certain spectral information, and within the signal width, the wavelet shift can also obtain part of the time domain information. Usually, the continuous wavelet transform can show the transform results in 3D model, but it cannot reach the standard of quantitative analysis in some special environments. Therefore, it is necessary to obtain WTm(p, q) after continuous wavelet transform and calculate it by adding it with the absolute value on the time axis, so that the 3D WTm(p, q) is converted to the 2D WTm(p), then: 14 $W T m (p) = \int_{t_{1}}^{t_{2}} W T m (p, t) ∣ d t$

The scale p is converted to the corresponding frequency f_a according to $\frac{f_{a} f_{b}}{p}$ (wherein f_a denotes the centre frequency of the wavelet transform and f_b denotes the sampling rate of the wavelet transform), and the wavelet spectrogram of the physical training data is obtained, and the physical training data fusion model is further constructed to obtain the multidimensional parameter detection model of the physical training data as: 15 $B_{j} = ((k (t) + 1), (k (t) + 2), \dots, (k (t) + n))$

Where k(t) denotes a discrete sample of physical fitness training data.

On the basis of the above equation, the results of regression analysis of physical training data were obtained as: 16 $G = \sum_{k = 1}^{i = 0} {(k (t) - ω_{i} (t))}^{2} + B_{j}$

Among them, 0 ≤ ω_i(t) ≤ 1. The collection and fusion of training data is completed using the 3rd order autocorrelation feature matching method. In the process of data fusion, the threshold control model of physical training data was obtained by quantitatively analysing students’ cardiorespiratory function. 17 $g_{i c} = G (β_{1}, β_{2}, \dots, β_{n})$

According to the results of the fusion analysis of physical training data, combined with the association detection algorithm, the fusion rule model for the AT critical value point anaerobic valve is constructed as: 18 $E_{2} = {(g_{i c} + R M)}^{2}$

According to the dynamic detection results of the physical training data, the feature extraction distribution results of the students’ physical training data can be obtained. Referring to the feature extraction distribution results of the physical training data, the feature extraction of the physical training data of each student can be realized.

3.2

Feature recognition output of physical fitness data

3.2.1

Identification of physical characteristics

Based on the results of the feature extraction distribution of students’ physical fitness training data, this paper introduces a convolutional neural network (CNN) model for physical fitness feature recognition. The convolutional neural network model consists of a convolutional layer, a pooling layer, a fully connected layer, and a classification layer [25]. The convolutional layer extracts features using convolutional kernels and contains multiple convolutional kernels inside. Each neuron is connected to only a local region of the previous layer, i.e., the receptive field, and the size of the region depends on the size of the convolutional kernel. Then: 19 $o u t p u t = f (\sum_{i}^{p \times q} w_{i} x_{i} + b)$ Where p×q is the convolution kernel size, w is the weight of the convolution kernel, x is the image grey value, after convolution plus bias b, f is the activation function, and output is the output after convolution.

The pooling layer is a downsampling operation, the main goal is to reduce the size of the feature map, the most common use of the maximum pooling function, the entire image is split into several small blocks of the same size, length and width of M and N respectively without overlapping, and only the largest number is taken within each block. After discarding other nodes again, the original planar structure is maintained, resulting in a new feature graph x_new, computed as: 20 $x_{n e w} = M a x (x_{m, n}) 0 \leq m \leq M, 0 \leq n \leq N$

The fully connected layer is located at the end of the convolutional neural, and its main function is to reintegrate the local features extracted in the previous steps and input the results to the activation function, which ultimately accomplishes the desired classification goal.

The classification layer is a general form of logistic regression, which can be used to implement a multiclassification problem. For input data with k category, Softmax estimates the probability that the input data x belongs to each of the categories in k. Softmax is a commonly used activation function for the classification layer, which normalises a vector of values into a vector of probability distributions with individual probabilities summing to 1. Softmax can be used as the neural network’s final layer for the output of a multi-classification problem. Then: 21 $Softmax (x_{i}) = \frac{e^{x_{i}}}{\sum_{i = 1}^{n} e^{x_{i}}}$

For the training of convolutional neural network is mainly distributed forward transmission and backward transmission two steps, through the training of sampling and transported through the input layer to the output layer. After reverse training each layer to obtain the deviation of each layer, followed by full-time neuron modification, under the current batch of sports target samples, to complete the end of training. Finally, the classification results for physical training features are output through the output layer, and the recognition of the features is completed according to the results of the classification.

3.2.2

Monitoring of optimised outputs

Analyse the convergence of physical training data, adopt the method of cardiovascular function as well as pulmonary ventilation function exchange to achieve the dynamic feature detection of physical training data, and obtain the feature extraction results of physical training data according to the results of training experiments and metabolic analysis. Using the method of correlation dimension retrieval, the amount of detection statistical features for real-time monitoring of physical training data was established as: 22 $U_{o g} = (E_{2} (g_{i c} + k (t)) - A g e)$

The moments of the distribution of linearly mapped combinatorial features of the physical training load data were established, denoted as: 23 $J = {\begin{matrix} A g e \\ U_{o g} + \frac{E_{2}}{R Q} \end{matrix}}$

According to the results of distributed detection of dynamic features of physical training load data, coherent fusion processing of physical training load data is carried out to obtain the test model for monitoring and evaluation of physical training load data.

The real-time monitoring output of physical training data is realised by fuzzy degree detection and dynamic recognition technology.

4

Monitoring and analysing the physical fitness level of physical education students in higher education institutions

Physical education activities in higher education have a profound impact on the physical fitness level of university students, which is beneficial to the formation of good exercise habits and directly affects the improvement of students’ physical fitness levels. Physical activity helps college students to become more physically fit and improve their metabolic capacity. It can improve the function of the respiratory and immune systems, enhance students’ physical fitness, and physical training of appropriate intensity can reduce the number, chance, and severity of illnesses. Dynamic monitoring of physical training based on physical education in higher education provides new strategic guidance for enhancing students’ physical fitness levels.

4.1

Effectiveness of feature extraction for physical fitness data

In this paper, the physical fitness professional athletes in all kinds of sports in the College of Physical Education of University Y are taken as the research objects, and the movement information and electromyographic information of the athletes before, during and after training collected by the image acquisition equipment and the sensor equipment, respectively, are taken as the experimental samples of a total of 400. In this paper, image feature extraction and EMG signal feature extraction at different frames in the sports video are used as the test indexes.

4.1.1

Characteristics of motor EMG signals

While image feature extraction tests are carried out, EMG signal feature extraction tests are carried out. The EMG signal feature extraction test was carried out for the biceps brachii muscle and the rectus femoris muscle of the leg of the students in athletics, and the wavelet transform algorithm was used to carry out before and after EMG signal feature extraction. The EMG signal feature extraction results for students’ physical training are depicted in Fig. 2, in which Fig. 2(a)~(d) shows the feature extraction results of the biceps brachii muscle and rectus femoris muscle before and after, respectively.

As can be seen from the figure, there is no obvious regularity when the wavelet transform algorithm is not used to extract the EMG signal feature information, and the biceps brachii muscle has obvious fluctuations between 13.5s and 14.7, with its amplitude fluctuating between [-0.45mV,0.45mV], while the rectus femoris muscle has significant fluctuations before 12.8s and after 15.7s, which leads to the bad effect of the EMG feature extraction for physical training. When the wavelet transform algorithm was used for EMG signal extraction, the fluctuation intervals of the biceps brachii muscle and rectus femoris muscle of the leg had a strong regularity, and the EMG signal feature extraction effect was better. According to the results of EMG, feature extraction of students’ physical training can help physical education teachers to clarify the changes of students’ EMG level during physical training and provide auxiliary decision support for assessing students’ physical fitness level.

4.1.2

Results of physical fitness feature extraction

In order to further verify the effectiveness of the wavelet transform algorithm in performing students’ physical fitness feature extraction, Improved Background Subtraction (IBS)-based and Improved Hybrid Gaussian Model (IHGM)- are chosen as the comparison algorithms to adequately extract the dynamic feature changes of students’ physical fitness training data. Figure 3 shows the comparison results of eigenvalue extraction of physical fitness training data under different algorithms, in which the physical fitness eigenvalues are related to the student’s physical fitness training status, and the higher eigenvalues indicate the more obvious students’ physical fitness training status.

As can be seen from the figure, the wavelet transform algorithm can better extract the physical training features of students, and its extracted physical training eigenvalue can reach up to 0.98, which can present the physical training state of students more significantly. On the other hand, the eigenvalues of students’ physical training extracted based on the improved background subtraction method and the improved hybrid Gaussian model are lower, and their eigenvalues are below 0.5 and 0.6, respectively. The reason is that the threshold set by the method based on improved background subtraction is too simple, resulting in false positives, and the method based on an improved hybrid Gaussian model has a poor correlation with the students’ physical training state, which is only applicable to the situation where a student’s physical training state is maintained for a long period. In conclusion, it is shown that the results of the wavelet transform algorithm-based feature extraction of students’ physical training are more significant, which can improve the feature extraction effect of students’ physical training data and provide guidance for the optimisation of students’ physical training strategies in college physical education.

4.2

Identification monitoring of physical training features

4.2.1

Physical training feature recognition

Based on the physical training monitoring framework, 400 sets of physical training actions for students in physical education were obtained, of which 80 sets of sample data were available for each of the five categories of standing, walking, running, jumping and squatting. Then, 300 sets of samples were selected from the action samples for constructing a new training set, and the remaining 100 sets of samples were used for constructing a test set. In this paper, a convolutional neural network is used to identify students’ physical training actions, and the loss function curve changes during the model training process, as shown in Figure 4.

As can be seen from the figure, the loss value of the model decreases with the increase in the number of iterations of the model, and the accuracy increases with the increase in the number of iterations of the model when the convolutional neural network carries out the recognition of physical fitness training features. The loss value of the model has a significant change in the first 100 iterations, the loss value of the model changes abruptly in 120 iterations, and the loss value stabilises at around 0.01 since 230 iterations. The accuracy of the model in the first 100 iterations has obvious improvement changes, and since 230 iterations, the accuracy of the model for physical training feature recognition tends to stabilise at about 0.99. It shows that the physical training features of students obtained by the wavelet transform algorithm as input of the convolutional neural network in this paper can effectively achieve the recognition of physical training features of students.

After the model has been trained in the training set, the recognition performance of the model is now verified using the test set. The CNN model designed in this paper is compared with SVM, RF, and GBDT algorithms, and the recognition results of student physical training features of different models are obtained as shown in Fig. 5. Where Figure 5(a)~(b) shows the recognition accuracy and training time of different models, respectively.

From the figure, it can be seen that the recognition accuracy of student physical training features of the RF algorithm ranges from 85.2% to 87.3%, the recognition accuracy of GBDT ranges from [89.2%,90.4%], and the recognition accuracy of the SVM model is within the interval of 92.5% to 94.8%. The recognition accuracy of the deep learning-based CNN model for student physical training features ranges from 98.5% to 99.5%, and this paper’s model is better at recognising student physical training features than the comparison model. In addition, the average training time of this paper’s model for student physical training feature recognition is about 148ms, which is 19.13%, 42.41%, and 53.61% lower than SVM, GBDT, and RF algorithms, respectively. The experimental results show that the CNN model has the shortest training time and the highest recognition accuracy for students’ physical training features. Therefore, the effective identification of student physical training features can help physical education teachers understand the physical training of students and provide support for optimizing student physical training strategies.

4.2.2

Errors in monitoring physical fitness data

In order to verify the accuracy and timeliness of the physical training monitoring architecture for real-time monitoring of students’ physical training, two sensor monitoring nodes were assembled on each of the students’ chest, back, left arm and left calf, and then the physical training signals outputted from the sensor nodes in each part of the student’s body were fused. Let the monitored objects do walking and running, etc., to monitor the physical training data of their bodies in real time, and select the root mean square error (RMSE) as the evaluation index, choose infrared sensor-based (A) and pyroelectric infrared sensor-based physical training monitoring system (B) as a comparison, and get the RMSE results of different monitoring systems as shown in Fig. 6.

From the figure, it can be seen that the overall trend of the RMSE curve under the monitoring method based on this paper is relatively smooth, and its fluctuation range is between [0.127,0.165], and the mean value of the RMSE is 0.139dB, which is the smallest change in amplitude as well as the value of the error among the three curves. The other two methods have relatively high error amplitudes, large differences in values, and a wide range of variations, and the overall accuracy of monitoring is lower compared to the methods presented in this paper. The other two methods used to monitor students’ physical training failed to effectively reject interference data and were too dependent on external signals. This leads to a large amount of noise data in the sensor, interfering with the normal work of the transmission mechanism, increasing the signal transmission delay, and affecting the accuracy and timeliness of the monitoring of students’ physical training. Through the technology to achieve effective monitoring of student physical training data in college physical education, so as to provide data support for teachers to analyse the physical fitness level of students, but also to optimize the physical training strategy of students to provide auxiliary decision-making and guidance.

4.3

Strategies for optimising students’ physical training

Starting from the data related to the dynamic monitoring of students’ physical training, the optimisation and analysis of physical training strategies in students’ physical education is aimed at further improving students’ physical fitness level, providing guidance for enhancing students’ physical fitness and promoting students’ all-around development.

4.3.1

Core strength optimisation training

The keys to optimizing your core strength training strategy are clear training goals, appropriate challenges, control of movement speed, alternating training styles, incorporation of functional training, and regular assessment and adjustment. Before developing a core strength training program, training goals need to be defined. Determine if it’s about improving strength or stability, whether it’s about training for a specific movement or improving core muscle capacity in general. Clarifying the goal will help teachers design a training program that is more appropriate for the student’s fitness level.

To progress in core strength training, it is necessary to have a certain level of challenge; moderate challenge is necessary, but ensure that the training is moderately difficult to avoid injury. Alternating different types of core strength training can prevent muscle adaptation and training plateaus. For example, alternating core stability training, strength training, and explosive strength training will improve the overall ability of the body in all qualities. Incorporating the latest core strength training methods can help to improve students’ overall physical fitness level. For example, functional training can better simulate actual competitive movements, strengthen core muscle stability and coordination, and improve the quality of completed movements. Regularly assess the training effect and progress of students and make corresponding adjustments according to the assessment results, increase the training intensity or introduce new training methods at the right time in order to maintain the training effect and motivation.

4.3.2

Optimisation of specific physical training

In the basic quality of physical training for students, we should take strength training as the basis, after obtaining a certain level of strength quality support, and then develop other qualities of the students in order to get the maximum training benefits. Especially for sensitivity training, students should have a certain level of strength and speed quality before training so as to maximise the harvest training effect.

For the relatively poor physical quality of the students, military physical training should be in the training process by observing the students in the training of the content and load intensity of the reaction to the actual adjustment of the training content and load. Establish a sound assessment and evaluation mechanism for physical training to ensure the implementation of training for all students and the overall improvement of quality level. Strengthen the exchange and cooperation with physical training professionals, master more advanced training concepts and methods, and constantly improve the traditional physical training mode of physical education in colleges and universities to make it more comprehensive, scientific and effective.

5

Conclusion

In the dynamic monitoring architecture of students’ physical training, the wavelet transform algorithm is used to extract the dynamic features of students’ physical training data, combined with CNN for the identification of physical features, and associated dimensional retrieval is used to achieve the real-time monitoring output of physical training data. When the wavelet transform algorithm is used to extract the features of students’ physical training, the variation amplitude of the biceps and rectus femoris muscles is between [-0.19mV,0.11mV] and [-0.18mV,0.09mV], which has obvious periodical changes compared with that before the extraction, and the highest value of the features extracted for physical training is up to 0.98. The recognition accuracy of the physical feature recognition model tends to stabilize at around 0.99 after 230 iterations. Stabilised around 0.99, and the RMSE fluctuations of students’ physical fitness training monitoring ranged between [0.127dB,0.165dB]. The analysis of students’ physical training data can provide support for optimizing students’ physical training strategies and enhancing their physical fitness levels in university physical education.

Due to the limitations of the research conditions, this study only combines the technology with the monitoring of students’ physical training to conduct a partial study of students’ basic physical fitness and partial physical fitness, which cannot be further tested on detailed physiological and biochemical indexes such as maximal oxygen consumption. Therefore, in the subsequent study, further in-depth research on physiological and biochemical indicators can be carried out when the experimental conditions permit so as to clarify the relevant influencing factors of the physical fitness level of students in college physical education and to provide guidance for the improvement of strategies for improving the physical fitness level of students in physical education.

Idioma:: Inglés

Calendario de la edición:: 1 veces al año
Temas de la revista:: Ciencias de la vida, Ciencias de la vida, otros, Matemáticas, Matemáticas aplicadas, Matemáticas generales, Física, Física, otros

RSS Feed de revista

Optimisation Strategies and Mathematical Modelling of the Path to Improvement of Students’ Physical Fitness Levels in Higher Education Physical Education

Yage Yang

Publicado en línea: 03 feb 2025

Recibido: 25 sept 2024

Aceptado: 06 ene 2025

DOI: https://doi.org/10.2478/amns-2025-0015

Palabras claveLSTM, Convolutional neural network, Wavelet transform, Physical fitness level monitoring

© 2025 Yage Yang, published by Sciendo

This work is licensed under the Creative Commons Attribution 4.0 International License.

Palabras clave
LSTM, Convolutional neural network, Wavelet transform, Physical fitness level monitoring