Image Recognition Method of High Intensity Sports Injury in Fish

Algorithm

Lei Huang

Hankou University, Wuhan, 430212, Hubei, China

Keyword: Fish School Algorithm, Sports Injury, Image Recognition.

Abstract: In today's society, image recognition technology is widely used and plays an extremely important role in

various fields. Good recognition technology is the key. How to improve the recognition rate and speed is of

great significance, which directly relates to the practicality and security of image recognition. Athletes will

inevitably be injured after high-intensity sports training, and their injury images need to be recognized during

treatment. However, some current recognition methods are not efficient and accurate. Therefore, this paper

will study the image recognition of sports athletes' high-intensity sports injury based on fish swarm algorithm.

First, extract the contour of the injured part, then obtain the preliminary recognition of the injury image, and

finally further recognize the image based on fish swarm algorithm.

1 INTRODUCTION

The development of modern science and technology

has greatly promoted the progress of sports science

and technology. High-tech means are widely used in

sports training, improving the scientific level of

training, and promoting the continuous improvement

of sports technology level and competition results.

With the development of sports science and the

gradual deepening of sports practice, the use of

scientific training methods and monitoring methods

in sports training to improve the effect of sports

training and improve sports performance has become

one of the important development directions of

sports. Scientific sports training is a kind of sports

training mode based on the feedback information of

functional monitoring indicators and technical

monitoring indicators to control the intensity of sports

training. Sports biomechanics is to monitor sports

training from the perspective of technical monitoring,

and the main means of monitoring is to measure the

kinematic parameters and dynamic parameters of

athletes during sports in real time, and make technical

diagnosis and evaluation in time, so as to improve

technical actions for athletes Provide theoretical basis

and technical guidance for improving sports

performance. The research of sports biomechanics in

kinematics mainly depends on the analysis system of

sports video. With the rapid development of computer

technology and the emergence of the Internet, the

motion video analysis system has made great

progress. From the film analysis system mainly used

in the 1970s and 1980s to the video analysis system

used at home and abroad in the 1990s; From the study

of single camera plane at the beginning to the study

of three-dimensional space of two or more cameras at

present; The camera's shooting frequency ranges

from dozens to hundreds or even thousands per

second, and the image is also clearer; With the

application of OPENGL, 3DMAX and other

software, the data analysis methods are more flexible

and diverse, and the display is more realistic and

detailed.

Image recognition is also called image pattern

recognition. It is a specific application of pattern

recognition technology in the image field. It is a

technology that establishes an image recognition

model for the input image information, analyzes and

extracts the characteristics of the image, and then

establishes a classifier to classify and recognize

according to the characteristics of the image. The

main purpose of image recognition is to process and

recognize images, pictures, scenes, characters and

other information to solve the direct communication

process between the computer and the external

environment (Yuan, Zhu, et al. 2017).

The development of image recognition has gone

through three stages: character recognition, digital

image processing and recognition, and object

Huang, L.

Image Recognition Method of High Intensity Sports Injury in Fish Algorithm.

DOI: 10.5220/0013544000004664

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 3rd International Conference on Futuristic Technology (INCOFT 2025) - Volume 1, pages 383-390

ISBN: 978-989-758-763-4

383

recognition. In short, it is a recognition process from

simple to complex. The improvement of computer

processing speed and the improvement of

corresponding algorithms provide foundation and

convenience for this. Image recognition mainly

focuses on the commonality of "classification".

According to certain standards, objects with the same

attribute are classified into one category and objects

with another common attribute are classified into

another category (Tao, 2019). For example, Arabic

numerals need to be divided into 10 categories,

English letters need to be divided into 26 categories,

and thousands of Chinese characters need to be

divided into thousands of categories. In addition,

different classification standards will result in

different classification results, such as classification

by color, classification by shape, classification by

other attributes, etc (Hamlin, Lizamore et al. 2017).

After high-intensity training, athletes will

inevitably suffer from a certain degree of physical

damage, and athletes will repeat a certain action,

which will also cause wear and tear of body joints or

skin. When there is damage, it needs to be treated.

Identifying the damaged position of the damaged

image is conducive to improving the treatment effect.

And many scholars have proposed different

recognition methods. Chen Hua et al. proposed a

recognition method based on linear discrimination

and ultrasonic image features. This method has good

recognition efficiency, but its recognition accuracy is

relatively low. Cai Shuyu et al. Proposed a

recognition method based on improved spectral

clustering, which has a certain recognition effect, but

its accuracy is not high. Yan Pei et al. Proposed a

recognition method based on wavelet coefficient Hu,

which can obtain more accurate recognition results,

but it takes a long time. So this paper will study a

recognition method based on fish swarm algorithm,

aiming to improve the accuracy and efficiency of

recognition (Chen and Yuan, 2021).

2 RELATED WORKS

2.1 Research Status of Fish Swarm

Algorithm

Fish Swarm Algorithm (FSA) is a new and efficient

swarm intelligence algorithm proposed by Li Xiaolei,

which has good robustness and strong searching

ability. Once the algorithm was proposed, it has

attracted the attention of many researchers. On the

one hand, researchers have conducted in-depth

research on the algorithm itself. For example, Ma

Xuan proposed a dual domain model fish swarm

algorithm. The algorithm first uses the coding method

directed by the precursor node to form a multicast tree

to represent individuals, and divides the search space

into feasible and infeasible regions. Then, the feasible

and infeasible regions are given different swimming

targets respectively to make full use of the swimming

behavior of feasible and infeasible fish individuals,

Improve the performance of the algorithm; Shi L has

made an empirical study on the performance of the

fish swarm algorithm, and has adaptively modified

the field of vision and the moving step size of the

individual fish during the implementation of the

algorithm, effectively balancing the local search

ability and the global search ability of the algorithm;

Xian S introduced the chemotactic behavior of

bacterial foraging optimization into the foraging

behavior of fish school, and improved the

optimization method of fish school algorithm

(Petushek, Sugimoto et al. 2019). On the other hand,

researchers have further expanded the application

scope of the algorithm. For example, Zhu Qiang used

the fish school algorithm in the network virtualization

mapping research, established a binary combination

optimization model based on the constraint

relationship between virtual network requests and the

underlying network nodes and links, and used the fish

school algorithm to achieve the approximate optimal

mapping of virtual network resources to the

underlying network resources; Liu Ding proposed a

multi-objective optimization fish swarm algorithm

and applied it to the threshold segmentation of silicon

single crystal diameter detection image to improve

the segmentation accuracy of bright halo in the

detection image; Zhu X proposed a selective

integration algorithm based on extreme learning

machine and discrete fish swarm algorithm, and

applied the fused algorithm to haze weather

prediction; The minimum mean square error (MMSE)

and fish swarm algorithm are combined and applied

to the research of multiple access interference

suppression in multi-user detection (Young, Gap-

Taik, et al. 2017). As mentioned above, researchers

at home and abroad have conducted more in-depth

research on fish swarm algorithm, but how to further

improve the performance of the algorithm and expand

the scope of application is still a research hotspot in

this field.

The structure of artificial fish includes such

factors as perception, behavior, behavior evaluation,

execution, parameters and data. When external

stimuli are added to the artificial fish, it makes

corresponding response by its fins. In order to reach

the global extreme, the artificial fish is always

INCOFT 2025 - International Conference on Futuristic Technology

384

swimming, but it does not swim aimlessly. The

artificial fish makes judgments according to its own

vision and the surrounding environment, and swims

in a better direction. As shown in Figure 1, it

represents the visual concept of artificial fish.

Figure 1: Visual concept of artificial fish

The current status of artificial fish i is 𝑋



which is

represented by 𝑋



=(𝑥



,𝑥



,,𝑥



). The field of

vision of artificial fish is visual, 𝑋



,𝑋



,𝑋



. which is

the status of other artificial fish in the visual field of

vision of the current artificial fish i, X, which is the

status of the current artificial fish in the visual field of

vision, which is represented by 𝑋



=(𝑥



,𝑥



,,𝑥





Compare the food concentration of state 𝑋



and state

𝑋



if the latter is greater than the former, then the

current artificial fish is in state 𝑋



The corresponding

artificial fish moves forward one step, and the state

after moving forward is Xnext. 𝑋



and Xnext can be

expressed by the following formula:

𝑋



𝑋



+visual ⋅rand()

(1)

𝑋

𝑛𝑒𝑥𝑡 =

𝑋



𝑋



−

𝑋



∥

𝑋



−

𝑋



∥

⋅

ℎ𝑎𝑛𝑑()

(2)

Where, step represents the step length of the

artificial fish, rand() is a random number, ranging

from 0 to 1. The food concentration is expressed as

an objective function value, let Y=f (X), where X is

the state of the artificial fish mentioned above, that is,

the function value of X, that is, the food concentration

corresponding to X. According to the above process,

the three behaviors of artificial fish swarm algorithm,

namely, foraging, crowding and rear-ending, always

tend to the places with high food concentration.

2.2 Research Status of Image

Recognition Technology

The development of image recognition

technology has gone through many stages, from

traditional recognition technologies such as

classification and extraction to the introduction of

Artificial Intelligence AI, now image processing has

become an important topic in the AI field. The most

basic part of image processing includes image

segmentation and recognition technology, which is

also a difficulty in image processing. The analysis and

processing of image data is very difficult, and it is a

key point of current research to imitate human image

operation. Researchers have developed different

computer programs to simulate the recognition

process of human image information (Liu, and Ji,

2021). Pattern recognition is one of the important

means of image recognition. The method based on

this recognition technology needs to analyze a large

number of data and information. At the same time,

combining expert experience and existing

knowledge, it makes corresponding judgments on

numbers, characters, curves, shapes, etc. through

mathematical reasoning and a large number of

computer calculations to complete the recognition,

evaluation and operation of pictures similar to human

beings (Shi, Guo, et al. 2017). The information

processing flow of the image recognition system is

shown in Figure 2, specifically including graph

segmentation, image feature extraction, classifier

recognition and other processes.

Figure 2: Flow of image recognition system

(1) First, convert the image into a grayscale

image.

(2) Normalized gamma space

In order to reduce the impact of illumination on

the image and normalize the image, the gamma

compression formula is as follows:

𝐼(𝑥,𝑦) =𝐼(𝑥,𝑦)



(3)

Generally, gamma is 1/2.

Image Recognition Method of High Intensity Sports Injury in Fish Algorithm

385

(3) Calculate the gradient size and direction of

each pixel in the image

Use the horizontal gradient operator [- 1,0,1] and

the vertical gradient operator [1,0,−1]



convolution the image to obtain the horizontal

gradient and the vertical gradient. The formula is as

follows:

𝐺



(𝑥,𝑦)=𝐻(𝑥+1,𝑦)−𝐻(𝑥−1,𝑦)

(4)

𝐺



(𝑥,𝑦) = 𝐻(𝑥,𝑦+ 1) − 𝐻(𝑥,𝑦 − 1)

(5)

Where 𝐺



(𝑥,𝑦), 𝐺



(𝑥,𝑦) represent the gradient in

the horizontal and vertical directions at the point

(𝑥,𝑦) respectively, and H(𝑥,𝑦) represents the pixel

value at the point (𝑥,𝑦). The gradient size and

gradient direction from point (𝑥,𝑦) are expressed by

the following formula:

𝐺(𝑥,𝑦)=



𝐺



(𝑥,𝑦)



+𝐺



(𝑥,𝑦)



(6)

𝛼(𝑥,𝑦) = 𝑡𝑎𝑛



(

𝐺



(

𝑥,𝑦

)

𝐺



(

𝑥,

𝑦

)

(7)

The gradient direction is divided into two types:

signed and unsigned. The unsigned gradient direction

range is (0, 180 °) and the signed gradient direction

range is (0, 360 °). The unsigned gradient direction is

adopted in this paper.

Gabor features have the following advantages:

1) It is not sensitive to changes in local

illumination.

2) The filter with small frequency can reflect the

global information and reduce the influence of noise;

The filter with high frequency reflects local

information for image filtering, but is sensitive to

noise.

3) The image rotation, stretching and other

transformations have little impact on it.

4) Because of these advantages, Gabor features

have been widely used in face recognition. But at the

same time, Gabor features also have certain

limitations, mainly including:

1) Computing Gabor features requires

convolution of the image, which will take time.

2) Because Gabor filters with different

frequencies and directions convolve the image, each

convolution operation will eventually get a column

vector, and the final feature dimension is large,

resulting in a large amount of computation.

3 FISH SCHOOL ALGORITHM

In a piece of water, the place where fish gather most

is generally the place where food is most abundant.

When no food is found, fish will search for food by

randomly swimming. When they find food, they will

swim in the direction of food, and other fish will also

swim with the fish that find food. According to the

characteristics of fish looking for food, fish are

constructed, and the activity behaviors of fish in the

process of searching for food are defined: foraging

behavior, crowding behavior, tail chasing behavior,

and random behavior. Each fish corresponds to an

optimization solution, the virtual water area where the

fish lives corresponds to the solution space of the

optimization solution, and the food concentration

corresponds to the objective function value. The

optimization is achieved by executing the defined

behavior of the fish in the virtual water area. This is

also the basic idea of fish school algorithm

(Kahlenberg, Nair et al. 2016).

A fish is an entity that encapsulates its own data

and has the ability to perform behavior. The flow

chart of the fish swarm algorithm is shown in Figure

3. It can receive the stimulus information of the

environment through its senses, evaluate and select

the appropriate behavior, execute and make

corresponding stress behavior by controlling its tail

fin. The environment in which the fish lives is the

solution space of the problem. The behavior it

chooses to perform at the next moment mainly

depends on the current individual's own state and the

state of the surrounding environment. Any behavior

activities it performs will also serve as a feedback to

the surrounding environment. This feedback will be

used to affect the surrounding environment, thereby

affecting the behavior choices of other nearby peers,

so as to ultimately achieve the goal of optimization.

Aggregation mechanism: in the process of

searching for food, fish will spontaneously gather

together to avoid injury and ensure their survival. For

any artificial fish, when the fitness of the central

position of its field of vision is high and not too

crowded, the artificial fish moves forward to the

central position, otherwise, the foraging mechanism

is implemented. Wherein, the expression of the

individual advancing to the central position is as

follows:

INCOFT 2025 - International Conference on Futuristic Technology

386

Figure 3: Flow chart of fish school algorithm

𝑢



(

𝑡

)

=𝑢



(

𝑡

)

+ 𝐿(𝑡)(𝑒



(

𝑡

)

+𝑒



(𝑡))

(8)

Tail chasing mechanism: when one or several fish

in the school find food, the fish in the vicinity will

follow them to the location of the food quickly. For

any artificial fish, when the fitness of an artificial fish

within its field of vision is high and not too crowded,

the artificial fish will move forward to the position of

the better individual, otherwise, the foraging

mechanism will be implemented. Wherein the

expression for the individual to advance toward the

better fish individual is as follows:

𝑠𝑢𝑝









‖

𝐼−𝐿

(

𝑡

)

𝐶

(

𝑡

)

𝑃



(𝑡)𝐵

(

𝑡

)‖

(9)

3.1 Pixel Calculation of Damaged

Image Damage Location Based on

fish Swarm Algorithm

Based on the above analysis, the damaged position

can be preliminarily identified, but the accurate

position cannot be obtained. Therefore, the paper will

further identify the damaged part by using the fish

school algorithm, so as to obtain more accurate

damaged part and calculate the damaged area.

Fish swarm algorithm is used in image damage

recognition, that is, each solution is regarded as a fish,

and then all solutions form a solution set of 7%. There

are two ways to find the final solution in the solution

set, namely taking the cluster center as the solution

and the clustering result as the solution. In order to

improve the recognition accuracy, this paper uses the

method of cluster center as the solution (Tan, Song,

et al. 2020). Then, the vectors of all the cluster center

points are regarded as each fish, and any pixel point

in the image can be represented by the position state

of the fish. Then the objective function of fish can be

expressed by the following formula:

(, )

gik

jVxdxy

=−•



(10)

Where, g represents the number of cluster centers,

x represents the cluster object, and V represents the

pixel cluster center. When j is the minimum value in

the formula, it is set as the best clustering point, which

is helpful to achieve the goal of damage graph

segmentation.

The gray pixel value of the clustered image will

correspond to the original pixel. After clustering

results, color description of pixels is realized, so

different colors in the image will represent different

representations. The pixel RGB representation value

can then be calculated by summing the GRB flux of

each type of pixel value and dividing it by the total

number of pixels.

In the process of sports, athletes may suffer

injuries at various positions of the human body, such

as shoulder joint injury, back injury, eye injury, etc.

The above fish school based algorithm can accurately

identify the injured parts of athletes, and can improve

the efficiency of identification.

3.2 Segmentation Method of Digital

Motion Injury Image

The motion injury image segmentation divides the

motion injury image according to the similarity

criteria of some features or feature sets of the digital

motion injury image, and divides the image into

several specific non-overlapping regions with unique

properties (Wang, 2019), (Han, Tang et al. 2020). The

motion injury image segmentation algorithms are

mainly divided into the following categories:

threshold type motion injury image segmentation,

regional type motion injury image segmentation and

Image Recognition Method of High Intensity Sports Injury in Fish Algorithm

387

edge detection type motion injury image

segmentation.

The threshold-type segmentation method of sports

injury image determines an appropriate threshold, and

classifies the pixel value of sports injury image by

comparing the pixel gray value with the threshold

value. The contrast threshold can accurately

distinguish the target area from the background area,

but if the threshold is too low, the information will be

redundant, and if the threshold is too high, the

information in the target area will be lost. Therefore,

selecting the appropriate threshold is the key point to

achieve (Gharaat, 2016). Threshold-based

segmentation is the most common segmentation

method because of its advantages of high efficiency

and speed. However, it is difficult to use a single

threshold for simple segmentation of motion damage

images with similar contrast between the target area

and the background area. Therefore, this method is

not applicable when the background of the motion

damage image is complex or there are many noises in

the motion damage image.

The basic idea of the region-based segmentation

method is to first find a pixel as the pixel point in the

original target area, which is called the seed pixel. If

the pixel around the seed pixel is similar to its pixel

value, then merge this pixel point into the target area,

and then judge the pixels around the pixel point. In

this way, gradually expand the target area until no

eligible pixels can be merged, and the target area is

selected. This method of growing from small to large

has ideal segmentation effect for relatively uniform

connected objects (Xu, 2021). Another method of

growing from large to small is called region splitting

and merging method. In contrast to the order of the

former method, the range is gradually reduced from

the whole motion damage image, and the sub-region

is segmented and then merged with the previously

segmented region. When it is judged that the pixel is

no longer similar to the previous pixel, it will stop

splitting, The previously divided area can be used as

the target area. The split-merge method is more

complex and has the risk of destroying the region

boundary, but the segmentation effect is good.

4 DAMAGE SAMPLE

COLLECTION SIMULATION

ANALYSIS

Although motion tracking based on human model

matching has advantages in identifying the stability

of joint points when the image quality is poor, it also

has the complexity of application and the inaccuracy

of joint point recognition due to individual

differences (Naoko, Terasawa, et al. 2019). The

accuracy of the human body model, the model

parameters and the amount of computation are in

direct proportion, that is, the higher the accuracy of

the model describing the human body, the more the

model parameters, and the greater the amount of

computation in the tracking process, reflecting the

complexity of the application. Although the

improvement of automatic scaling is adopted, the

scale of each link has been determined after the model

is established. Due to individual differences, the

model cannot be completely consistent with the

moving human body, which will cause certain errors.

However, the traditional method of determining joint

points directly from feature points on the contour line

is easy to use, but the contour features caused by poor

image quality are not obvious, especially for the

method of determining feature points based on the

curvature of the contour line, and even cannot be

recognized (Kojima, Kasai, et al. 2017), (Kahlenberg,

Nair, et al. 2016), (Singh, Gupta, et al. 2022). This

sample collection method provides a clearer and

intuitive real-time image of the damage in the

underground movement. Through the collection, a

total of 180 sample images, including three types of

damage images, are obtained in this paper. As shown

in Figure 4.

Figure 4: Example of damage type image

Because the experiment in this paper needs to

build a deep learning network model for recognizing

sports injury images, the size of the data sample set

will affect the level of the model learning rate. In

order to prevent the over-fitting situation caused by

the small amount of sample data in the training

process, the obtained training sample library needs to

be enhanced by data to expand the image sample

library (Wang, Bin, et al. 2019). First, the original

image is grayed out, and then the expansion of the

sample database is considered to be divided into two

steps: first, the training sample image is translated

into four directions of pixels, and then moved 30

pixels in the direction of 45 ° clockwise shift to the

horizontal and vertical axes, and the original position

after the movement is filled with the pixel value 255,

INCOFT 2025 - International Conference on Futuristic Technology

388

which can expand the sample data by four times. On

this basis, the sample image is horizontally mirrored,

Vertical image and horizontal vertical image can also

expand the number of sample databases by three

times (Zhu, Sun et al. 2020). The number of motion

injury image samples collected in this paper is

expanded to 3600, from which 3000 (1000

perforation, 1000 crack, 1000 deformation) are

selected as the experimental training samples, 600

(200 perforation, 200 crack, 200 deformation) are

used as the experimental test samples, and the size is

normalized to 224 × two hundred and twenty-four ×

3 pixel size. The simulation results are shown in

Figure 5 below.

Figure 5: Simulation result

In order to meet the requirements of fast feedback

and simple operation for the technical analysis of

sports injury, and provide convenient digital images

for the automatic recognition of the joints of the

subsequent moving human body of the analysis

system, this study adopts the method of first

segmenting the ghosted moving human body, then

removing the image ghosting and frames by filling,

merging the processed moving human body with the

background, and finally obtaining the image sequence

without ghosting and shaking.

5 CONCLUSIONS

Athletes will inevitably be injured when they are

playing sports. Identifying their injury pictures will

help improve the therapeutic effect of athletes. In

this paper, the damaged part of the image is

recognized based on the fish swarm algorithm.

Compared with the other two recognition methods,

the method in this paper helps to improve the

recognition efficiency and accuracy. Athletes are

very vulnerable to injury during sports. In order to

reduce the degree of injury, they should strictly

follow the action standards during training to avoid

serious injury.

REFERENCES

Yuan Y , Zhu H . An optimization approach for pattern

recognition image simulation based on improved

artificial fish swarm algorithm[J]. Revista de la

Facultad de Ingenieria, 2017, 32(16):1070-1077.

Tao W . Multi-joint Cooperative Control of Athletes under

High Intensity Training[J]. Journal of Physics:

Conference Series, 2019, 1237(3):032077 (8pp).

Hamlin M J , Lizamore C A , Hopkins W G . The Effect

of Natural or Simulated Altitude Training on High-

Intensity Intermittent Running Performance in Team-

Sport Athletes: A Meta-Analysis[J]. Sports Medicine,

2017.

Chen X , Yuan G . Sports Injury Rehabilitation

Intervention Algorithm Based on Visual Analysis

Technology[J]. Hindawi Limited, 2021.

Petushek, Erich J.Sugimoto, DaiStoolmiller,

MichaelSmith, GraceMyer, Gregory D. Evidence-

Based Best-Practice Guidelines for Preventing Anterior

Cruciate Ligament Injuries in Young Female Athletes:

A Systematic Review and Meta-analysis[J]. American

Journal of Sports Medicine, 2019, 47(7).

Jin-Young, Choi, Gap-Taik, et al. Study on the Sports

Injury of Tennis Athletes to Respective Affiliation[J].

The Korean Journal of Growth and Development, 2017,

25(1):119-128.

Liu Y , Ji Y . Target recognition of sport athletes based on

deep learning and convolutional neural network[J].

Journal of Intelligent & Fuzzy Systems: Applications in

Engineering and Technology, 2021(2):40.

Shi L , Guo R , Ma Y . A novel artificial fish swarm

algorithm for pattern recognition with convex

optimization[C]// 2016 International Conference on

Communication and Electronics Systems (ICCES).

IEEE, 2017.

Kahlenberg C A , Nair R , Monroe E , et al. Incidence of

injury based on sports participation in high school

athletes[J]. The Physician and Sportsmedicine, 2016.

Tan L , Song Y , Ma Z , et al. Deep Learning Video Action

Recognition Method Based on Key Frame

Algorithm[J]. 2020.

Wang G M . Simulation of Damaged Part Identification

Method for High Intensity Sports Damage Image[J].

Computer Simulation, 2019.

Han L , Tang L , Tang Y . Sports image detection based

on particle swarm optimization algorithm[J].

Microprocessors and Microsystems, 2020,

80(2):103345.

Gharaat, M, Agha-Atinejad, et al. The effect of high-

intensity interval training on ventilatory threshold and

aerobic power in well-trained canoe polo athletes[J].

Science & sports, 2016.

Image Recognition Method of High Intensity Sports Injury in Fish Algorithm

389

Xu Y . Repairing waist injury of sports dance based on

multifunctional nano-material particles[J].

Ferroelectrics, 2021, 581(1):172-185.

Naoko, Terasawa, Kenta, et al. Effect of L-citrulline intake

on intermittent short-time high-intensity exercise

performance in male collegiate track athletes[J]. The

Journal of Physical Fitness and Sports Medicine, 2019,

8(4):147-157.

Kojima C , Kasai N , Ishibashi A , et al. The Effect Of

High-intensity Interval Exercise In Hypoxia On

Appetite Regulations In Female Athletes: 3443 Board

#348 June 2 3: 30 PM - 5: 00 PM[J]. Medicine &

Science in Sports & Exercise, 2017, 49.

Kahlenberg C , Nair R , Monroe E , et al. Incidence of

injury based on sports participation in high school

athletes.[J]. The Physician and sportsmedicine, 2016,

44(3):269-73.

Singh S , Gupta A K , Arora T . A Review of Machine

Learning-Based Recognition of Sign Language[J].

International Journal of Image and Graphics, 2022.

Wang Z , Bin G E , Mingyu T U , et al. Image Segmentation

Algorithm Based on Improved Otsu Algorithm and

Artificial Fish Swarm Optimization[J]. Packaging

Journal, 2019.

Zhu P , Sun F . Sports Athletes' Performance Prediction

Model Based on Machine Learning Algorithm[J].

2020.

INCOFT 2025 - International Conference on Futuristic Technology

390