Feasibility Study of Mass Sports Fitness Program Based on Neural Network Algorithm.

Mass sports has become a world trend, setting off a new health revolution in the world. Mass fitness programs not only enrich people's lives. It not only relieves the psychological pressure of modern people but also promotes people's health and improves people's quality of life. According to the time-consuming stability of neural network algorithm, this paper proposes a sports video recognition algorithm based on BP neural network. The static and dynamic features are classified by BP neural network, and the basic probability assignment is constructed according to the preliminary recognition results. At the same time, we use evidence theory to fuse the preliminary results and get the results of motion video recognition. It can be applied to the generation model of the feasible scheme of mass sports fitness. Relevant experiments show that the whole model that generates the feasible mass sports fitness scheme can accurately generate the sports fitness scheme of multiple patient users and ensure the rationality and safety of the sports fitness scheme.

1. Introduction

With the development of social economy and the improvement of people's living standard, mass fitness sports has developed rapidly and has become an important part of people's daily life, involving more and more projects. People's living standard is gradually changing to a well-off and affluent type. Physical fitness has become an important means for people to keep fit, entertain, and promote health and is an indispensable part of modern people's daily life [1]. Efforts should be made to provide the people with the conditions, rights, and opportunities for fitness activities in social development, attach importance to the improvement of people's physical quality, advocate the establishment of a scientific, civilized, and healthy modern lifestyle, and improve the quality of life of the people. Generally speaking, people who insist on physical exercise for a long time will have their own physical exercise program [2, 3]. Different objects, different aims and pursuits, and different conditions can have different physical exercise programs, so physical exercise programs have obvious personality characteristics.

With the development of social urbanization, more and more people are flocking to cities, and the urban population is expanding rapidly. At the same time, people are getting farther and farther away from nature. Socialization and electrification of housework have greatly shortened the time of housework. People have more leisure time, and on the premise of meeting basic needs, they have new pursuits and developments. Caspersen et al. [4] hold that the establishment of physical fitness program is beneficial to reduce the blood pressure of hypertensive patients and put forward five key factors: safety, exercise mode, exercise intensity, exercise duration, and exercise frequency. Liu et al. [5] affirmed the importance of physical fitness programs in the field of health, and the physical fitness programs began to develop more comprehensively, with different new application fields appearing. With the development of sports fitness programs in China, in order to popularize fitness knowledge, many sports colleges and universities have offered courses related to sports fitness programs [6]. Barker et al. [7] put forward a sports video recognition algorithm based on the combination optimization principle, and the accuracy of sports video recognition is improved. Yan [8] applied the grey system and BPNN model to sports prediction in order to find a better method for quantitative prediction of sports results. Liang [9] established a risk early warning model of sports events based on BPNN. Finally, an empirical study is made on the sports event risk early warning model. The empirical study shows that using the BPNN-based sports event risk early warning model can achieve better early warning effect [10].

Fitness has become a way of life and is accepted by people. However, there are many problems in urban community residents' physical exercise. How to keep fit scientifically and reasonably is an important issue for modern society to explore. The neural network model is a nonlinear dynamic system, which has low requirements on the nature of data and can fit any nonlinear relationship, thus providing a new idea for the processing of nonlinear data [11]. Neural networks have a high degree of nonlinear dynamic processing ability, and it is not necessary to know the distribution form and variable relationship of data. For some complex systems, the input and output relationship is too complicated to express neural networks with general formulas, but it is easy to realize their high nonlinear mapping relationship. This paper focuses on the analysis of the generation of sports fitness programs based on neural network algorithm, aiming at providing scientific and effective sports programs for the public and promoting the development of sports fitness industry in China.

2. Research Method

2.1. Physical Fitness Test Data Mining

The purpose of physical exercise has two levels. The primary purpose is obviously utilitarian, such as curing diseases and prolonging life. The purpose of higher level is to enjoy the nature on the basis of health. Because the development of mass sports in China started late and its basic point is low, physical exercise has not become an important part of people's life, and most people have not developed the habit of physical exercise. Therefore, the specific national conditions and people's feelings make the development strategy of mass sports in China significantly different from the implementation strategies of western countries. The large population has become the biggest obstacle to China's economic development, and it is also the biggest obstacle to China's mass sports development. The difference of economic base makes our country and western developed countries have differences in formulating the development strategies of mass sports.

In the face of big data processing, special data mining technology has emerged to meet the increasing data volume in sports data processing. This new data mining technology will be of great help and vital importance to sports statistics. Data mining technology is different from traditional statistical computing. Data mining is usually related to computer science and achieves the above goals through many methods, such as statistics, online analysis and processing, information retrieval, machine learning, expert system (relying on past experience rules), and pattern recognition. The biggest difference is that the statistical final inference is only a hypothetical final inference [12, 13]. Data mining and processing technology in sports data analysis and management is just a kind of data management and analysis technology to analyze and study the hidden information in these big data. This article mainly makes some technical analysis and research on sports data mining and processing technology for these aspects and gives a simple explanation.

Physical health intervention means that physical education teachers make sports training plans according to the public's achievements and help improve the public's physical health test scores [14]. The distance between two masses is the Euclid Distance between the vectors representing the achievements of these two masses.

For two people to measure the public x=(x, x,…, x), y=(y, y,…, y), the distance between them is

The closer the distance between two objects is, the more similar they are, that is, the closer their physical measurement data are. Of course, when calculating the distance, it is necessary to standardize the vector first because the calculation of the distance is affected by the dimension.

For a physical test crowd x, his k-nearest neighbor N(x, k) is the set of k physical test crowds closest to other physical test data, and the density of an object is defined as the reciprocal of the average distance to k nearest neighbors. The size of k is specified by the user:where N(x, k) is the nearest neighbor of physical mass x, including x itself, because density(x, k)=0 must be the nearest. |N(x, k)| represents the size of the collection.

In fact, the independent existence of two transaction sequences is the fundamental reason for this principle. For the case where both sequences have frequent itemsets, we can get the following formula:where n is the length of two sequences and Lenmax is the maximum length of an independent sequence. According to the theorem, in theory, we find that Apriori also scans the itemsets in the independent sequence, but for these two transactions, this independent sequence is meaningless.

At present, in the technical platform, there are several common technical implementation methods to realize the persistence of entity class object instances. (1) Standard JDBC technology: the JDBC API is directly adopted and standard SQL statements are directly embedded in the data access objects, and the data access operations in the target database tables—add, delete, modify, and query function operations, are realized. (2) CMP (container managed persistence) is implemented by the J2EE EJB container to manage the persistence function of EJB components of each entity. (3) ORM (object/relational mapping) uses object/relational mapping technology to implement the persistence function of entity class object instances. (4) JDO (Java data object) is a standard API developed by Sun company, which is used to describe the persistence semantics of object instances. It supports the persistence of object instances to any storage system. It includes relational database, object-oriented database, XML-based database, and other special storage systems. If a physical mass has a low relative density, the higher the probability that the mass is an outlier, the more attention should be paid to the physical data of the mass [15, 16].

2.2. Public Physical Health Knowledge Discovery and Decision Support

In the modern society with highly developed information, it is difficult for traditional statistical methods to meet the demand of how to discover useful information from massive data. Data mining has become an important means of knowledge discovery based on big data. The larger the amount of data obtained, the more complicated things are and the more fuzzy the final result is. Usually, the theory of fuzzy set is used to make fuzzy speculation and estimation of things. It is difficult to put forward lifelong exercise programs for exercisers, but it is entirely possible to put forward programs or physical exercises that have great exercise value, are easy to stick to for a long time, and have been selected by more exercisers.

Generally speaking, the event detection needs to define an event pattern first and then match the content of the video with this event pattern. If it matches the pattern, it means that the event happened; otherwise, it means that the event did not happen [17]. If there are too many feature choices, on the one hand, it will affect the efficiency of motion mining; on the other hand, it may also have a negative impact, resulting in the mined knowledge deviating from the real situation. If the feature selection is too little, the required knowledge cannot be mined, or the mined knowledge is too little to meet the user's requirements.

The object Ω that defines an entity class E can be described by a structural model, which is represented by a 3 yuan formula:where RE represents a plane area in the video, which is called the space area of Ω, δmin represents the minimum time threshold of Ω, and δmax represents the maximum time threshold of Ω.

With the increasing number of sports video data, it is more and more important to manage sports video data effectively, so that users can find the required sports video efficiently and quickly. BPNN (BP neural network) has strong nonlinear classification ability and fast training speed, which can better meet the real-time requirements of video recognition. The sports video recognition process of BPNN is shown in Figure 1.

Figure 1

Sports video recognition.

The input layer node of BPNN only transmits the input signal to the hidden layer, and the Gaussian function constitutes the function of the hidden layer node, namely:where σ is the variance of radial basis function and c represents the center of the hidden layer.

The output of each node in the hidden layer is linearly weighted and summed:where w is the weight of the ith hidden layer node and the jth node of the output layer and m is the number of nodes in the output layer.

In this paper, the activation function ϕ(x) adopts the commonly used sigmoid function:

Because the input is a quantity that changes with time, it should be developed into a spatial variable. In order to reduce the randomness of the input vector, the data should be normalized to a value in the range of [0,1] before input. It is necessary to use logarithms or error functions to improve the regularity of the data [18, 19]. We need to standardize the input vector with the following function:where ymin, ymax represent the minimum value and the maximum value in the input vector of the ith unit; μ, v are the parameters, where 2μ+v=1, and represents the normalized input vector.

Using the above-mentioned influence analysis method, one or several variables with small influence values are eliminated to obtain a new variable set. Then, the new variables are used as input variables to reconstruct the neural network and retrain and establish a new neural network model, and the above-mentioned influence analysis method is used to eliminate the variables with small influence values again. By repeating this process, we can find out the most important influencing factors on dependent variables.

2.3. The Feasibility Plan of Mass Sports Fitness Is Generated

In the research, firstly, each influencing factor is quantified by a certain method, and then the structural equation model is used for analysis, and the theoretical basis of the analysis is: Fitness literacy concept and people-oriented thinking. After improving our understanding of fitness knowledge, we should change our physical exercise behavior, so as to improve our physical quality and health. Based on the theory of health literacy, this paper constructs a mass fitness model [20, 21]. The model is checked and revised by questionnaire and physical fitness test data, which accord with the confirmatory factor analysis in structural equation model. It is to verify whether the causal relationship between independent variables and corresponding dependent variables is consistent with the hypothesis through the test results of each factor.

There will be a lot of data in sports, and if these data are not used, then this pile of data will be useless. These data are just like rich material mines, and only after corresponding processing and data analysis can we extract useful information from them and turn it into useful data, rather than a pile of useless data. Under such special advantages, many parents have suggested that schools offer more physical fitness programs to promote the combination of physical education and other disciplines. The society and state organs pay more and more attention to physical education class, and physical training is regarded as an important course for lifelong benefit. In many colleges and universities, physical training is a compulsory course for the public. .Therefore, for men who actively participate in physical exercise, we should guide them to carry out targeted scientific exercise. Strengthen the fitness program, mainly to develop their cardiopulmonary function and strength exercise, and psychologically cultivate their spirit of hard work and hard work.

One of the purposes of the generation model of mass sports fitness feasibility plan is to provide centralized services for multiregional health management centers, including user management, physical data management, sports fitness plan management, and sports project management. In order to realize the generation model of mass sports fitness feasibility scheme based on association rules, the overall architecture is designed with three layers, and the overall frame structure of the system is shown in Figure 2.

Figure 2

Integral frame structure.

The overall business process is that users use browsers, mobile phones, mobile devices, and other clients to register and log in and conduct physical fitness detection with identity information to obtain a comprehensive evaluation. The physical fitness detection process is started by scanning the two-dimensional code. The process of scanning the two-dimensional code can ensure the user's identity information confirmation, accurate operation of physical fitness detection data, and data security.

Data evaluation requires scientific evaluation of users' test data. For example, the national physique item refers to the national physique standard issued by the state, and the scoring system is designed. Corresponding evaluation is made according to the physique measurement data such as BMI, age range, and so on, and the overall evaluation of users is carried out for each total score. In the original analysis of physical fitness programs, only a few available options can be obtained when a higher value is set, and this will result in the absence of related multidisease sports items, resulting in the failure of the expert reasoning system to generate physical fitness programs. Therefore, according to the data characteristics of the current physical fitness programs, more candidate results can be obtained when the support degree is set to 0.01 and the credibility is set to 0.1 after many attempts.

3. Result Analysis

Because there are great differences in physical quality among individuals, if we directly check the excellent rate and pass rate according to the whole grade, we cannot know the details of the changes in grades. In order to make the vertical comparison more objective and scientific, we should further standardize the basic conditions of vertical comparability, such as the scientific construction of the index system. Therefore, when analyzing the overall physical change trend of the sample masses, we choose to check what changes each person has made on the basis of their original grades and calculate its proportion in the whole.

Physical measurement data also include areas with different densities. For example, the density of people at normal level will be higher (more people), while the density of people at excellent level will be lower (less people). According to the analysis of the excellent retention rate of mass sports fitness results, the trend chart of excellent retention rate of male and female sports fitness items is shown in Figure 3.

Figure 3

Excellent retention rate of male and female sports fitness programs.

It can be found that compared with other physical fitness events, boys' 1,000-meter running keeps a low proportion of excellence, while girls' corresponding physical fitness events are 50-meter running and 800-meter running, which indicates that the establishment of physical fitness events plays an important role in maintaining the public's sports ability, but it has no great influence on the examination of skills and flexibility of sitting body flexion, indicating that the current physical fitness focuses on the training of physical fitness events for the public, but the training of the public's skills is insufficient. This is a problem that should be paid attention to in physical fitness.

The association rule data mining of physical fitness test scores can also play a certain role in prediction. For example, when a person has excellent sitting body flexion test scores, excellent standing long jump scores, and good vital capacity, we can reasonably predict that the person's 50-meter running scores are excellent and BMI is normal. Table 1 shows the prediction function of association rule data mining.

Table 1

Example of predictive function of association rule data mining.

Serial number	Result	Support degree	Confidence level
1	Failed	0.001	0.968
2	Got through	0.001	0.937

Its prediction function is the same as that shown above, and there are many similar strong rules. Therefore, this algorithm can be used to predict the results of other public sports fitness events when the results of some public sports fitness events are obtained. By testing the sports fitness items with simple process and flexible venue, you can get all the relatively reliable results, which is of great help to save the testing cost and flexibly select the venue.

For the events obtained in the previous section, we can further mine knowledge. Taking the video of sports fitness project as an example, through six video experiments, the video accuracy rate is shown in Figure 4.

Figure 4

Event detection accuracy rate.

From the analysis, it can be seen that the correct rate of basketball and football is relatively low. The main reason is that in basketball and football matches, there are too many participants, which can easily lead to tracking loss or tracking errors, thus affecting the entities corresponding to regular expressions and further reducing the correct rate of detecting round events. However, there is a big difference in the correct rate, which may be due to the fact that the tracking is carried out through the clothing color of the athletes, and the visual quality of video 1 is relatively high, so that its corrected trajectory is relatively consistent with the real trajectory of the athletes, and the correct rate of the corresponding round event detection is also relatively high.

The established sports video recognition model is used to recognize the video test set, and the average recognition accuracy is shown in Figure 5.

Figure 5

Performance comparison of different sports video recognition models.

At the same time, the trajectory of basketball and football is more complex. The sphere swings differently with the athlete's body, so the accuracy of the motion trajectory measurement of these two items is poor. Compared with the sports video recognition model with single feature, the recognition speed of BPNN is greatly improved, and useless and redundant information is effectively eliminated. It is a sports video recognition model with high recognition rate and high speed, so the step size is the best index to evaluate the network training speed. When the learning rate is small and the number of hidden nodes is small, the smaller maximum step length often fails to reach the expected error. If the research purpose does not require a high speed of network training, a larger maximum stride length can be appropriately selected.

Online recognition of sports video action is often needed, so the recognition speed of sports action is analyzed by testing experiments. The average recognition time of sports action between BPNN and other models is shown in Figure 6.

Figure 6

Identification time of sports action.

It can be seen that the average recognition time of BPNN is shorter than that of the comparison method because BPNN reduces the input dimension of the sports classifier, speeds up the modeling speed of sports recognition, accelerates the convergence speed of BPNN, improves the recognition efficiency of sports, and better meets the practical application requirements.

The development of sports consumption has not yet formed a scale. It is believed that with the continuous improvement of people's ideological level and lifestyle, mass sports consumption will surely become an important choice for people to spend their leisure time. Therefore, we must formulate special policies to strengthen the training of sports instructors. In the process of training sports instructors, we should pay special attention to two points. First, we should work hard to improve the ability of talents and strive to ensure that instructors' ability can meet the development needs of healthy sports. Second, efforts should be made in rational allocation, that is, a specialized talent pool of sports instructors at the city level and district (county) level should be established, and sports instructors in the region should be rationally used according to the needs of residents.

4. Conclusion

This paper presents a sports behavior recognition model based on BP neural network. This model has higher recognition rate than other sports behavior recognition models, shortens the execution time, accelerates the speed of sports behavior recognition, and can provide valuable reference information for sports fitness and training. Because BPNN reduces the input dimension of motion classifier, accelerates the modeling speed of motion recognition, accelerates the convergence speed of BPNN, improves the efficiency of motion recognition, and better meets the requirements of practical application, the average recognition time of BPNN is shorter than that of the comparison method. In order to enhance the public awareness of fitness and improve the national physical quality, we must absorb foreign advanced experience, mobilize the whole society to revitalize national sports, establish and improve the development system of social sports fitness, and strengthen the skill training and guidance of national aerobics.

However, this study has some limitations. The BPNN model can find out the most important factor among many influencing factors. But for the multilayer neural network model, the calculation process is complex. Therefore, it is still impossible to reasonably explain how the independent variable affects the dependent variable. Although research can identify the most important factor of many influencing factors, for the multilayer neural network model, the calculation process is complex. Therefore, it is still impossible to reasonably explain how the independent variable affects the dependent variable.