Artificial Intelligence Technologies and Their Application for Reform and Development of Table Tennis Training in Complex Environments.

The development of information technology has been deployed in almost all sectors and is making life easier. In this development, the sports sector has seen tremendous expansion. In traditional table tennis training, the coach and the players have to meet daily to take appropriate training for the game. This process is time-consuming in a complex environment, and it will have a significant impact on the reformation and development of table tennis training. The utilization of improved technologies can overcome this challenge by performing training of the game online with an intelligent wireless system with advanced training mechanisms. Carrying the traditional and heavier intelligent wireless devices will be difficult for the players. In this research, the fine-grained evaluation (FGE) system is incorporated into the deep learning model to analyze the player's body postures during the training and event sessions and make them develop after each session through online training in any circumstance. The proposed FGE was compared with the traditional statistical model, and it was observed that the proposed FGE had obtained higher precision and recall values of 70% and 98.9% than the statistical model.

1. Introduction

It is necessary to use artificial intelligence to study and construct the concept, method, and technique and software applications for simulation, expansion, and growth of general intelligence [1]. One of the subfields of computer science known as artificial intelligence aims to understand the essence of cognition and construct an autonomous robot that responds like a human being. The evaluation of the research process for the creation of software applications is in accordance with the software development technique [2]. Computational mathematics has a lot to do with the rejections of neural nets, decision trees, and the clustering method, which were all studied for the table tennis event. A backhand loop is one of the most effective scoring tools in table tennis, and it accounts for more than half of the difference between a winning and a losing streak in a billiards match [3]. International male table tennis players' backhand loop in preparation for a forehand strike rate is a highway. Chinese men's forehand and backhand are groundstrokes, whereas Chinese table tennis players' backhand loop in preparation for a forehand strike rate is a highway. China's women's table tennis players heavily use the forehand looping method and forehand loop technology, whereas international ladies rely on backhand and road strategies to score [4]. There has been progressed in the study of table tennis technique and strategy from concepts to statistical equations, and approaches have progressed from depending on classical statistics to the use of strong computational tools and computer-aided technology. An upgraded machine learning model based on artificial intelligence has been widely used in data mining, decision support, and other areas. Little research has examined the benefits of an integrated solution for diagnosis, evaluation, and prediction in the area of game strategy diagnostics [5].

Only objective evaluation, knowledge of the strategic and tactical abilities of competitors, and preparation for and planning of contest approaches could help people win the table tennis competition [6]. From a descriptive method to a quantifiable one to a combined one, diagnosis and evaluation are frequently experienced through deciding the learning outcome by assessing the current competitive resilience of both ends, by the indicator of general to particular assessment, from manual to computer-assisted study to human contact, and from basic mathematical analytics to information mining and decision-making processes [7]. It is a kind of athletic training known as “shadow play” that entails mimicking a certain talent in a sport by doing repetitive motions. Inexperienced players and those who are just starting out might benefit from this method. Using shadow play as a tactic helps players improve their stroke performance when playing table tennis, a popular handheld game [8]. The notion of table tennis shadow play is a way to train table tennis blows even without a ball. Playing in the shadows has the potential to improve a player's overall stroke performance and their awareness of proper racket posture and the ideal stroke approach. The use of shadow play can be beneficial, but only if it is performed under the direction of an expert. In addition, this strategy needs special training materials because there are so many different ways to play shadow play [9].

A table tennis automaton is a clever and considerate robot that plays table tennis with people. It is able to understand the supply chain and trajectory of competing objects, make intelligent judgments and trade strategies, and perform a variety of different hits. Research into table tennis robots spans a wide range of areas of research. Machine learning, artificial intelligence, robotics, kinematics and dynamics, and computer animation are just a few of the topics that are represented. Many other applications may be found for it [10], which makes it an important part of the scientific community. In table tennis, the game is short and the flight velocity is fast, necessitating that table tennis robots should be able to quickly respond. Table tennis routes that quickly move must be predicted in a short amount of time [11]. Real-time and precise table tennis robot control requires a lot of time and effort. Table tennis robots will only be successful if their trajectory predictions are accurate, which is why this is an important consideration [12]. In order for the table tennis robot to perform at its best, it must first thoroughly study the ball's characteristics. A shift between qualitative and empirical research on table tennis strategy and technique is required in order to fully understand the subject [13]. There was a shift in methods from relying on the previous human analytical models to the current computational models and computerization [14]. Diagnostics for virtual computerized ball games offer a brand-new avenue for the study of table tennis game strategy diagnostics. Some studies have tried to merge convolutional neural network techniques with a computer model to diagnose all matches, but strategic and tactical indicators broadly hindered its development and training application [15]. More than 350 table tennis teams were examined for their performances at the past five World Championships and the Summer Olympic Games. It uses a task force-designed “table tennis strategic and tactical gathering system” and collects the strategic and technological index of data from every match [16]. Cloud technology, data science, and other factors have helped artificial intelligence systems reach a new level of development. The performance of such hard task categorization has been significantly enhanced, such that machine learning, computer vision, language processing, robotic devices, and speech-based technologies are rising [17]. On these confined days, the trainer must devote a significant amount of time, which is not possible. To keep the student continually supported throughout these last few days, an accessible, pragmatic, and clever system with minimal expense might be an appropriate choice [18]. As a result, machine learning (ML) approaches may be used in many real-world contexts, such as the education business, the industrial sector, healthcare, and transportation infrastructure [19]. While the cutting technology is responsible for reliable and exact information collection about individuals' actions, sensors are also responsible for that information. A combination of sensing devices and machine learning algorithms might be an efficient solution to the problem of the lack of suitable coaches during these emergency days (under confined circumstances for defined reasons) [20]. The use of machine learning (ML) algorithms has made significant progress in addressing, measuring, and anticipating difficulties related to ADL, sports activity tracking, and health and illness concerns. ADL and athletic activity assessments are discussed in this section in connection with one another. Implementations, kinds of major objectives, and methods utilized are used to assess the research. The task of estimating in ADL (activity of daily life) entails the use of sensation, which has been extensively studied and commercialized [21].

Sensor-equipped methods have been used by several researchers to record biometrics or human movement. [22] This study has addressed the subject of assessing the performance of man's daily life activities. The pulse rate was estimated by using the convolutional neural network with regression (CNNR). An ActigraphGT9X was performed as a number of sensors recorded human interactions. The author used a wrist-worn inertial measurement unit (IMU) and Gaussian process analysis to develop a walking pace assessment system [23]. One of the most popular areas of study in this department is sport-related activities and accomplishment evaluation techniques. Sports players' motions have been collected and analyzed in a variety of ways, including cameras, sensors, and a combination of the two [24]. There have been a lot of studies performed on the subject of sport-related analysis and estimation. [25] A researcher has developed an efficient method for coaches to provide their students with excellent coaching by employing just one IMU. Using a twin spatial and temporal CNN model, a new table tennis eyeball categorization method was developed [26]. To aid table tennis coaches, researchers used three body-worn IMUs. An LSTM network including stochastic properties was used to categorize the keystrokes as either unidirectional or bidirectional. Table tennis players might benefit from the high-dimensional time series data generated by the sensors [27]. It was used in the proposed jogger exhaustion-predicting program that relied on a wrist-worn IMU. As a more accurate method of predicting athletes' performance, support-vector regression (SVR) was proposed. Football, in particular, is plagued by the issue of undiagnosed injuries and the time it takes to recover from them. Based on data from Tottenham Hotspur's professional football players, the model predicted “healing time” [28]. Footballers' injury forecasting issues were also addressed in this study. The SVR was used in the first phase of their system, and KNN in the second. On the basis of the GPS and data model of an Australian football club player, the regression and SVM analysis were given. Based on performance monitoring system data for athletes' datasets, the research team devised a model with LSTM to predict soccer players' peak levels of preparation in the future [29]. When faced with confusing or linguistically based information, the researchers recommended an ellipsoid-shaped inference system with the neuro approach that outperforms traditional fuzzy systems. According to their findings, the developed model may be used in real-world applications with a low-processing overhead [30]. Besides these findings, this study focused on evaluating the development of table tennis training using artificial intelligence. This study focused on artificial intelligence technology and its application for the reform and development of table tennis training in a complex environment using deep learning.

2. Proposed System for the Table Tennis Training

In our proposed methodology, people have employed serving robots to serve the table tennis balls, which are also known as ping-pong and whiff-whaff. The serving robots in table tennis work with the aid of a wireless cord to function from a remote location. The robots are programmed to fire the tennis ball once the power is switched on. These robots will follow the firing mechanism of one or more positions on the table, and hence, the players have to make an attempt to hit the ball back to the robot. The table tennis serving robots are machines, which automatically shoot the ball one by one at a preset strike rate. The ball storage capacity can be varied according to the need of the player and type of training. These types of robots are used to train the player where there is no need to wait for an opponent to play/train. The serving robots give the experience of playing with multiple players through their preprogrammed ball-serving module. The speed, frequency, and angle differ from module to module and can be selected by the player. Since the nature of the game is so fast, high-speed cameras are used to monitor the training session. High-speed cameras record fast-moving objects as photographic images on a storage medium. The images stored on the media can be played back in slow motion once after being recorded. A high-speed camera can capture moving images with exposures of less than 1/1000 seconds and frame rates of more than 250 frames per second. Deep learning technology is deployed in this proposed system. Deep learning is a form of machine learning algorithm used in artificial intelligence that mimics the activities of the human brain. Artificial neural networks are used to extract data. In this situation, multiple-layered artificial neural networks are deployed. Convolutional neural networks (CNNs) are a type of neural network that excels at picture categorization and recognition. CNNs are utilized in computer vision techniques like object recognition, face recognition, and traffic sign recognition. It is also employed in self-driving automobiles and robotic vision systems. Feature extraction is the process of translating raw data into numerical features that can be processed while keeping the information in the original dataset. It also produces better outcomes than directly applying machine learning to raw data. The collected data are analyzed and displayed on the LCD. With the help of these advanced technologies, the player can be trained with multiple skills in table tennis. The deep learning techniques offer various tasks, such as tracking the performance of the players for further assessment, which is represented in Figure 1. Thus, artificial intelligence is found to be an efficient model for improving and reforming the table tennis game in a complex environment.

Figure 1

Proposed model for the table tennis training in the context of deep learning.

To extract data, neural networks are deployed. Multiple-layered artificial neural networks are used in this case. Convolutional neural networks (CNNs) are a type of neural network that is particularly good at image categorization and recognition. CNNs are used in computer vision applications like as object recognition, face recognition, and traffic sign recognition. It is also used in self-driving cars and robotic vision systems. The process of converting raw data into numerical features that can be processed while retaining the information in the original dataset is known as feature extraction. It also yields better results than merely applying machine learning to raw data. For the data extraction process, the convolutional neural network (CNN) model, which is a type of deep learning concept, is implemented. The neural model is trained to perform the processes of analyzing the data of the player's movements and extract the required features for taking further training in playing the game.

On the other hand, the usage of multicategory classification is very prevalent. This classification appears to have two options. The number of softmax structural models distributed in logistic regression is represented by r, in addition to numeric classification approaches. There is currently a multicategory classifier with a high number of r categories represented as p(i)1,2,…, n. The classification possibility considered in the softmax similarity classification for such testing dataset d is shown inwhere M is the quantity of softmax structural models distributed in logistic regression and p(i)1,  2, ...,  n, with a high number of r  classifications. r is a numeric classification addition, and A specifies an attribute for a single category. M denotes the model's specifications, which are likewise denoted by a  r-line structure. As shown in equation (2), each dividing line may be considered as both a FGE algorithm employed classification attribute for a single category and a dividing line for several categories.

The maximum activity is identified in a sequence of A, where p is the number of color data points. The color scatterplots are represented by p and G; R is used to denote a separate greatest activity. Where p denotes the number of color data points, G and G denote the color scatterplots of clips r and r − 1, comprising both height and width. Height denotes the number of pixels in each frame, and R denotes the limitations for distinguishing a distinct greatest activity in a succession of A discontinuities, as illustrated in equation (3). In this, du x and du y are also linearization elements for a particular structuring element. Users refer to the sequential parameters as linearization. The coordinates of a centerline are denoted by x and y, respectively, and sequential parameters are denoted by k′s.

Its row vector r=(x, y) is described. The coordinate matrix d is constructed anew with all frames that are not marked as exceptions by vertically appending the row arrays r. U is a k  × 3 matrix, where k is the number of macroframes that are not designated as exceptions. The vectors Q and Q are created by adding all of du x and du y that are not identified as outliers. Finally, the curve shape parameters k=(k1, k2, k3) are employed, where k=(k4, k5, k6) is blended together as indicated in equation (4). Based on such equations, one can write Q=Uk and Q=Uk, then b and b were computed utilizing R's nucleation and growth matrix shown in

Its total probability seems like one when the modeling approach is normalized by k and k. The extract of the system is shown in

For these suggestive functions, the FGE algorithm applied profitability rules. 1{.} consists of one 1{expression for correct value}=1 and 1{expression for incorrect value}=0. The situations of r categories are then aggregated using softmax correlation. Equation (6) calculates the likelihood of d being classified into j groups.

The regression analysis extraction generalization is shown in equation (3). The similarity goal function is shown in

The extract generalization of regression analysis is represented by G(ϑ). An iterative optimization approach, such as in the statistical model, can minimize the performance parameters in equation (8). Equation (8) shows how and where to calculate any modified version of an iterative procedure as a result.

G(ϑ) is a vector, and its lth(γJ(ϑ)/γδ) is the lth term whether it is in the jth category of such a cost function. With the simplification of the problem, the above equation is fed into regression analysis and iteratively amended. Since the same amount is removed from each procedural control parameter, the importance of such a failing function does not actually start to change, implying that the parameter might not be the only answer. As seen in the following equation (9), every scientific proof technique is represented.

Frequency (b) appears being the mathematical concept of a typical frequency scale in the given equation. As indicated in equation (10), its frequency scale has a frequency spectrum of 0–45050 Hz and a fixed frequency distance.

This difference in voice signals is measured by energy. Such as recorded chats A, where n = 1, 2,…, u, its energy is estimated to use the equation (11) as those of the recording of just such a signal energy.

Y denotes the recording of such a signal energy, as in recorded talk A.

A(p(=jd(; ϑ) denotes that the energy loss is computed, and the results are enforced. It also denotes that such a functional is close to meeting the optimal requirements. Its energy loss is estimated, and the findings are used to confirm that the model's parameters are increased and that the functional form seems to be the restricted function indicated by equation (12). Equation (13) depicts the differential equation as it approaches the ideal criteria.where τ > 0 represents the fractional generated function in exponential functional.

Subsequently, by reducing the goal function using equation (15), a viable soft limit similarity classification model is created.

3. Results and Discussion

The data in this study were evaluated utilizing a fine-grained evaluation technique. This study was carried out in order to assess the effectiveness of table tennis training.

Figure 2 shows the fine-grained evaluation (FGE) process, which employs intelligence to perform record classification of distinct dataset types, diffuse them, and remove discrepancies. It also needs to take into account the characteristics of various tennis-playing strategies. Multiple standards are represented in arrays, and their dataset integration is further distinguished. The number of deep learning technology employed for the analysis of various table tennis training of arrays varies by category. The observation of flying and the view of the playing surface methods have occupied higher-frequency classifications of 14%, respectively, for a total of 30.5%.

Figure 2

Frequency classification performance analysis.

Its probability density result (see Table 1) can also be interpreted as a frequency categorization for further explanation. A survey may be divided into two groups for training and testing, with the frequency classification result based on the training category identification with the highest likelihood. The correlation between the normal human table tennis motion and the conventional movement with a certain parameter is evaluated using such testing categories' greatest probability significance.

Table 1

Classification frequencies for training and testing categories' results.

Parameter	Classification of frequency (%)
	Training	Testing
View of the playing field	97.8	98.34
Flight observation	99.7	96.78
Prospects for a fast-playing surface	95.6	97.56
Aerial view of the slow-playing surface	86.7	89.78
Prospects for a fast-playing medium surface	84.3	86.58
Close-up surface (or view outside of playing)	87.4	85.89
Others	92.6	94.63

Figure 3 shows the characteristics that have a greater impact on training for sporting activities. To begin establishing a healthy relationship between detection and tracking and varied circumstance performances, this model integrates multiple artificial neural analytical approaches and selected high-resolution characteristics. Sporting events have a variety of answers to problems that arise during training, as has been discovered. To perfect the versatile implementation of extra information and facts, they conducted a series of control trials.

Figure 3

Deep learning-based performance analysis for table tennis height.

Furthermore, the speed of development is unaffected by the strong magnetic transition and indicates the difference in the sequential direction and speed of a rigid body through time. Analyzing table tennis activity with a three-axis synthetic acceleration screen can assist the FGE algorithm in clarifying the process (refer to Table 2). If two convolutions create a chain, the larger filter size may increase the quantity of testing required, resulting in a shortage of computational resources and a failure to complete the training process on time. The following table shows that the standing range parameter has larger values than the others when compared to table height, table width, wingspan, and standing range analysis. Though the table height and width in table tennis are constant, the player has to get accordingly trained, as the player's height may get varied and also the players have to focus on how much speed, steps, and other criteria to achieve in the game.

Table 2

Results of a deep learning performance analysis for table tennis training.

	Table height	Table width	Wingspan	Standing range
Total	71.00	71.00	71.00	71.00
Mean	87.84	89.11	92.89	102.49
Standard deviation	4.42	3.40	4.71	5.98
Max	95.25	96.25	95.75	122.50
Min	78.25	79.50	74.00	98.50

Figure 4 shows how the FGE algorithms are used in sports. It was discovered that in the continuous learning computational connection of tennis sports, a distinct linguistic methodology exclusively based on the deep learning algorithm is being used for data processing and that this method appears to have far more advantages.

Figure 4

FGE algorithm in vertical table tennis utilizing a deep learning method using performance analysis.

This probability density result (see Table 3) is also known as a neural network deduction. A test can be split down into different categories, and also the graph's performance of the classifier is determined by the classification label that corresponds to the highest significance level. The resemblance between the more important human table tennis movements and the normal action is evaluated using this highest valuation. This has a specific point of reference.

Table 3

FGE algorithm with performance analysis in table tennis vertical using the deep learning method.

Parameters	Minimum range of tennis vertical	Maximum range of tennis vertical	Tennis vertical (number of steps)
Total	59.00	59.00	59.00
Mean	42.75	148.05	54.52
Standard deviation	4.62	3.29	5.54
Min	34.00	131.54	32.60
50%	46.56	146.00	34.60
100%	41.00	153.00	37.60
Max	48.00	139.00	32.60

Deep learning pioneered the technique to persuade table tennis players to embrace various training styles, as seen in Figure 5. An FGE algorithm is solely based on a randomized matrix and a swarm-based technique for optimal control evaluation. Its methodology can aid in the future detection of human activity in athletes even when they are fluidly moving but at the expense of a partial accuracy.

Figure 5

FGE algorithm in deep learning method employing performance evaluation of table tennis.

According to the previous research, classic table tennis analysis methods used a variety of device facilities including processing methodologies. Algorithm research, on the other hand, is limited. As a result, device renovation is being used to investigate concerns with analyzing and recognizing table tennis competitions. The purpose is to use the mean and standard deviations, and the lowest and maximum values, to analyze and simulate bar counter tennis ball flight patterns. On this basis, physiological models of predicting the ball's trajectory have been quickly equipped with such a high-speed racing that indicates the webcam. Deep learning is utilized to forecast the flight patterns of a table tennis ball although most representations are quite complicated. A sphere tracking and extraction characteristics system and a ball path prediction are incorporated to improve predictions (refer to Table 4).

Table 4

FGE algorithm with a performance by table tennis sprint utilizing deep learning method result.

	Indicator (width)	Quickness	Race
Total	71.00	56.00	57.00
Mean	9.67	12.28	4.21
Standard deviation	1.98	0.74	0.17
Minimum	−1.00	13.21	4.16
Maximum	9.50	12.12	4.32

The results in Figure 6 show that the algorithm may definitely solve the challenge of recognizing mannerisms in table tennis games. The recognition is carried out not only with the assistance of continuous motion but with insufficient recognition precision. To increase the identification rate, the authors suggested a deep learning technique for deformation motion detection based on the FGE algorithms. It is of excellent quality and can also recognize tennis training. It is capable of determining the rotary movement of the heel joint and the training outcome.

Figure 6

Table tennis FGE algorithm using overall analysis through deep learning methodology.

The novel presented model is used to execute a new technique of recognizing human ping-pong activity while attaining the goal of additional training based on multi-FGE. This model can be used to assess human table tennis action. Between all of them, the basic definition of individual body collective activity is difficult to express. This enhanced original conception level is originally employed in the FGE prototype to provide a good representation of insert for activity data, which really is a novel framework presented by humans. Furthermore, a spiraling feature output was improved by identifying the necessity for arithmetic matrix operations that extract its conceptual characteristic of a table tennis activity. The generalizability of cluster centers and also the correlation technique from the feature space will be improved by substituting this same actual information using feature space (refer to Table 5). The precision of a fine-grained evaluation of human table tennis activities' method interventions' evaluation will be improved by these three criteria. It is the FGE algorithm's best performance when compared to the standard statistical method. Our proposed fine-grained evaluation of human table tennis action method delivers a higher ground-truth value of 30 and then a recall percentage of 2.6 % than the standard statistical method, as shown in the table above. However, the proposed system has a precision percent of 0.7%, which is lower than the statistical method.

Table 5

Results of a comparison performance analysis for table tennis using the deep learning methodology.

Objectives	Ground of the truth	Value of recall (%)	Value of precision (%)
Methodology used in standard statistics	40	96.3	96.4
Method for a fine-grained evaluation of human table tennis actions	70	98.9	95.7

4. Conclusions

Many sectors have been able to take advantage of artificial intelligence because of the advancements in technology. The goal of this study is to examine how artificial intelligence can be used to improve table tennis training in a complex setting. Two to four players are typically involved in table tennis. Small rackets are used to hit the ball back and forth across a table in this game. A net divides the playing surface. This game necessitates lightning-fast reflexes on the part of the participants, and if one of them loses the ball without passing it, the other gains a point. Table tennis training in a complicated environment can greatly benefit from the deployment of artificial intelligence applications, according to our design. When several people's needs must be taken into consideration, the environment is referred to as complicated. Table tennis training, for example, often necessitates training more than one player at a time. Individual care plans are required for each athlete due to their different needs. Two to four players are required for table tennis instruction. There is no one-size-fits-all training method for players. A training robot and artificial intelligence are, therefore, used to accomplish this goal. Table tennis may now be effectively trained in a complicated setting thanks to the proposed technology. This research focuses on the training of the table tennis player in a complex environment. For the training process, an intelligent visualization is considered in this research. Some of the parameters include the view of the play field, objects in the play field, surface of the play area, position of standing, and movements are considered. Hence, this research reflected the level of concept understanding and remembering the points that are analyzed based on the survey. The point specified in the comments will be considered as the future work.