Professional Documents
Culture Documents
Research Article Application of Face Recognition Technology in Intelligent Education Management in Colleges and Universities
Research Article Application of Face Recognition Technology in Intelligent Education Management in Colleges and Universities
Research Article
Application of Face Recognition Technology in Intelligent
Education Management in Colleges and Universities
Donghong Xiang
Wuxi Vocational Institute of Commerce, Wuxi, Jiangsu, China
Received 14 June 2022; Revised 8 July 2022; Accepted 27 July 2022; Published 13 August 2022
Copyright © 2022 Donghong Xiang. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
The application of artificial intelligence technology can help colleges and universities effectively improve the efficiency of campus
management and teaching quality. The traditional educational management concepts and models in colleges and universities can
no longer meet the growing needs of students. In this paper, the application of artificial intelligence technology in intelligent
education management in colleges and universities is proposed. The face recognition technology is constructed through the
volume neural network model and the hog algorithm. The experimental results show that the face recognition system can meet
the real-time requirements of face recognition and improve the accuracy of face recognition. In addition, in the dormitory
management application experiment, the face recognition system proposed in this paper has more advantages than the
traditional algorithm in recognition accuracy. The face recognition system in this paper can identify the problems existing in
class attendance and provides a certain reference value for the classroom on campus.
irregular spot check [4]. This method not only needs to occupy ability; and enables students to achieve all-round develop-
the classroom time and teachers’ energy, but also has a high ment. Artificial intelligence technology is a technology that
error rate. It is unable to detect the situation of being late enables computers to have the ability of analyzing and sensing
and leaving early, substituting for classes and absenteeism in and simulating corresponding responses to human thinking.
real time. With the continuous development of the Internet It has been widely applied in many fields, such as Alipay’s face
technology and information technology, the information con- registration and train station’s face screening [8]. Artificial
struction of university education management has developed intelligence technology is also one of the important technolo-
rapidly. Building intelligent education management system gies for colleges and universities to realize the Internet of
and platform has become the development trend of university Things and intelligent education management. Some colleges
education management, and it is also an important research and universities have applied artificial intelligence technology
on the development of university education management [5]. to the library management system. In the past, the library
Based on the payment request of college information identity registration was realized through the campus all-in-
construction, this paper puts forward the application research one card or password, and this method also obviously has
of artificial intelligence technology in college intelligent educa- some disadvantages [9]. The way of biometric verification
tion audience. At present, the typical applications of artificial has high requirements for biometric acquisition, and there
intelligence in the field of education mainly include intelligent are certain potential safety hazards, so it is not conducive to
tutors, intelligent partners, intelligent evaluation systems, fea- the promotion of colleges and universities [10]. The face rec-
ture recognition, and learning analysis, involving three scenes: ognition technology in artificial intelligence technology can
teaching scene, learning scene, and management assessment, complete identity recognition without contact and in the most
which basically realizes the full penetration of education. The natural state of the identified person, which is safe and reliable
face recognition and detection technology are constructed [11]. In addition, learning is still an important task for college
through convolution neural network and image gradient direc- students in college life, and at present, most colleges and uni-
tion histogram feature extraction algorithm and tested in the versities need students to reach the standard in English before
dormitory management and classroom attendance manage- graduation. However, for many students, oral English and lis-
ment module of college education management system. This tening are two major difficulties in English learning [12].
paper is mainly divided into three parts. The first part is the Therefore, scholars have built a learning system for English lis-
elaboration of the application and research status of artificial tening and speaking practice based on artificial intelligence
intelligence technology in intelligent education management technology and judge the problems of learners’ pronunciation
in colleges and universities. The second part is the construction through speech recognition by artificial intelligence technol-
of face recognition model in university intelligent management ogy, so as to assist learners to make corresponding improve-
system. The third part is the experimental results and corre- ments [13]. Other scholars apply artificial intelligence
sponding analysis of face recognition model in college intelli- technology to the detection of students’ classroom behavior
gent education management system. and recognize students’ expression and body state through
recognition technology, so as to judge students’ classroom
behavior [14]. In addition, some scholars detect students’ run-
2. Application of Artificial Intelligence ning through the Internet of Things technology and artificial
Technology in Intelligent Education intelligence recognition technology, which can improve the
Management in Colleges and Universities accuracy of students’ performance and help students correct
wrong running posture through action recognition [15].
Intelligent education management in colleges and universities In addition to college learning and education, artificial
refers to the management mode to improve the management intelligence technology is also used in college safety manage-
efficiency and teaching level of colleges and universities ment system and student life management system. With the
through the integration of college education information net- construction and development of university monitoring sys-
work application technology and college teaching information tem, some scholars have introduced video recognition tech-
technology [6]. Intelligent education management in colleges nology into the monitoring system, which can identify and
and universities needs to use modern information technology detect the vehicles and personnel entering and leaving the uni-
and science and technology to realize software requirements versity, improve the efficiency and safety of university campus
on the basis of intelligent education management hardware management, and reduce the possibility of potential safety
equipment, such as the Internet of Things technology, big data hazards [16]. In addition, with the continuous improvement
analysis technology, and artificial intelligence technology [7]. of university information construction, the role of dormitory
Intelligent education is a high-end form of educational infor- management system has gradually become prominent and
matization, which emphasizes the openness and innovation has become the focus of university education management
of education, so that teaching and learning can break through research at home and abroad [17]. Foreign scholars first inves-
the limitations of time and space. Intelligent education adheres tigated and sorted out the needs of all aspects related to dormi-
to the people-oriented concept; makes full use of high-tech tory management and developed a management mechanism
means such as big data, mobile Internet, and artificial intelli- with dormitory as the core on this basis. The management
gence; builds a networked, digital, and intelligent smart learn- mechanism essentially defines the main responsibilities of
ing environment; meets students’ personalized learning needs; the campus, dormitory, and students and ensures the stable
cultivates students’ comprehensive quality and innovation operation of the whole system [18]. Compared with developed
Wireless Communications and Mobile Computing 3
countries, the domestic research on dormitory management nition speed. Face recognition is one of the important
system lags behind. Not only does the dormitory management technologies for the implementation of intelligent education
system and concept not meet the needs of university informa- management in colleges and universities. It can carry out face
tion construction, but also the technical means in the manage- recognition in a noncontact way, which is convenient and reli-
ment system are relatively backward. Student dormitory able. The principle of face recognition technology is shown in
management is a very important part of college student man- Figure 1. It mainly collects face related data and extracts image
agement. The level of student dormitory management reflects features and then stores the obtained information in the data-
the level of student work in a school, which will directly affect base as basic data information. When the camera captures the
all aspects of the school. For the current major universities, the face image in the specified area, it will digitize the image for
work of student dormitory involves a lot of information. If image processing, then compare and recognize it with the exist-
manual registration is adopted, it will consume a lot of time ing image eigenvalues in the database, and finally get the result.
and energy of administrators, and it is easy to cause informa- Therefore, the key link of face recognition technology is
tion errors, information loss, and other problems. Other face detection and image feature extraction and comparison.
scholars have combined artificial intelligence technology with The face detection algorithm used in this paper is to fuse the
university network service platform, so that students can extracted image gradient direction histogram, namely hog,
obtain corresponding service information through the net- with the image features extracted by convolution neural net-
work platform, such as waiting time in canteen and usage of work based on Yolo model. The fusion result is the final fea-
bathroom [19]. ture of the face.
At present, many domestic universities have applied artifi- The image gradient direction histogram feature extraction
cial intelligence technology in education management system. algorithm can extract the edge contour of the image in a short
Although it has improved the efficiency of university manage- time and overcome the influence of illumination and color on
ment and teaching quality, it still does not achieve the expected the detection results in the process of face recognition and
effect. There are three main problems. First of all, the digitiza- detection. After the image is input, it needs to be changed into
tion of the educational management process in colleges and a unified format file with the same pixel size, and then the
universities is low. University management contains a wide horizontal gradient and vertical gradient of each pixel con-
range of contents, and each piece of management content con- tained in the image are calculated accordingly and then calcu-
tains a large number of data. Although the introduction of lated through the convolution kernel, which is ½−1, 0, 1T and
artificial intelligence technology can improve the data statistics ½−1, 0, 1. The calculation is shown in formulas
and sort in various fields, the data sources of universities are
scattered and have a certain fluidity, which make the statistics gx ðx, yÞ = hðx + 1, yÞ − hðx − 1, yÞ, ð1Þ
of some data incomplete, resulting in the low authenticity of
the data managed by artificial intelligence [20]. Second, the gy ðx, yÞ = hðx, y + 1Þ − hðx, yÞ: ð2Þ
teaching in the process of college education management has
not been accurate. Artificial intelligence technology can The gradients of the coordinate points of the pixels in the
improve the quality and level of teaching to a certain extent, horizontal and vertical directions are expressed as gx ðx, yÞ and
but teachers do not fully grasp the learning situation and teach- gy ðx, yÞ, respectively, and the pixel values are expressed as h
ing effect of students. Relying too much on the evaluation of
ðx, yÞ. According to the gradient value obtained from formu-
human-computer interaction will also make teachers’ evalua-
las (1) and (2), the touch and direction of the gradient corre-
tion of students inaccurate and incomplete. Third, the idea of
sponding to the point can be solved, as shown in formulas
people-oriented education is ignored. Artificial intelligence
technology can standardize the educational management of qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
colleges and universities and make all departments of colleges gðx, yÞ = gx ðx, yÞ2 + gy ðx, yÞ2 , ð3Þ
and universities act according to rules. At the same time, these
standards are also a restriction on students and teachers, which
gy ðx, yÞ
are not conducive to the innovation and reform of teaching. d ðx, yÞ = tan−1 : ð4Þ
gx ðx, yÞ
3. Construction of Face Recognition Model in Through the calculated pixel gradient modulus and corre-
sponding direction, the picture can be transformed into gradi-
University Intelligent Education
ent histogram, and the gradient histogram of all units can be
Management System obtained. After that, the whole picture can be traversed arbi-
The research of face recognition system began in the 1960s trarily through the sliding window. In this process, the eigen-
and has been improved with the development of computer values contained in each block will be normalized, so as to
technology and optical imaging technology since the 1980s. improve the robustness of the image to shadow, illumination,
The real application stage is in the late 1990s, and it is mainly and edge changes. The normalization method is shown in
realized by the technology of the United States, Germany, and formula
Japan. The key to the success of face recognition system lies in vm,n
whether it has a cutting-edge core algorithm and makes the vnormed
m ,n = 9 4
: ð5Þ
recognition results have practical recognition rate and recog- ∑m=1 ∑n=1 ðvm,n Þ2 + ε2
4 Wireless Communications and Mobile Computing
Feature point
Image
location
Image Facial feature Face
Face detection
acquisition point location normalization
Face position
Normalized
face image
Processed
image
Feature
Feature Face image
End transformation
extraction preprocessing
and recognition
Identification Facial
results feature
The abscissa index of the image gradient direction histo- the corresponding size will become smaller. As shown in
gram is expressed as m, the cell index contained in the sliding formula (8), it is the sampling function:
window block is expressed as n, and the count value of the gra-
dient direction histogram corresponding to the index is
ðl Þ ðl−1Þ
y j = down y j : ð8Þ
expressed as vm,n .
The input image also needs to be processed through the
convolution neural network based on the Yolo model. Each The nonlinear mapping capability of the network can be
convolution in the convolution neural network based on the enhanced in the full connection layer, as shown in formula
Yolo model will convolute, add bias, activate, and down sam-
ple the input image in the corresponding layer. The input !
n
ðl−1Þ ðl Þ ðlÞ
image needs to undergo six convolution operations and four olj =f 〠 xj ⋅ ω ji + bj : ð9Þ
pooling operations, and then the corresponding image eigen- i=1
values are extracted under the processing of the activation
function. The convolutional neural network based on Yolo In the formula, the number of neurons in the upper layer
model is an improvement based on the structure of convolu- is expressed as n, the connection strength between neurons in
tional neural network, as shown in Figure 2. Compared with the current layer and neurons in the upper layer is expressed
the traditional feature extraction and classification methods, as ω, the bias of neurons in the current layer is expressed as
the way of feature extraction of input image by convolutional b, and the activation function is expressed as f ðÞ.
neural network is to achieve automatic network learning Face features have great complexity and many catego-
through layer by layer convolution dimensionality reduction ries, so face features are classified by softmax regression
and multilayer nonlinear mapping to obtain the feature function in the network. If the number of samples is n and
extractor and classifier required for target recognition. There- the number of categories is s, the training set composed of
fore, after the input picture is convoluted through n × m con- s classes is expressed as fðxðtÞ , yðtÞ Þ, ⋯, ðxðsÞ , yðsÞ Þg, in which
volution kernels, biased and activated, the characteristic map
there is sample xðrÞ ∈ Rn+1 , and the corresponding class mark
of the corresponding layer can be obtained, the number of
which is n, and the activation function is shown in formulas is expressed as yðrÞ ∈ f1, 2, ⋯, mg. The regression function is
shown in formula
0 1
2 3
ðl Þ ðl Þ ðl−1Þ ðl Þ
yj = f @〠 〠 ω jiðu,vÞ × xi ðq + u, p + vÞ + b j A, exp θT1 xðrÞ
6 7
3 6 7
i∈M jðl=1Þðu,vÞ∈K ðlÞ
2 6 ⋯ 7
ð6Þ ðr Þ ðr Þ 4 5
p y = 1 x ; θ T ðr Þ
6 7 exp θm x
hθ xðrÞ = 6
6 ⋯ 7=
7 :
C ðlÞ = ðu, vÞ ∈ N 2 j0 < u < C x , 0 < v < C y : ð7Þ 4 5 ∑m exp θT xðrÞ
p y ð r Þ = m x ð r Þ ; θ k=1 k
Input
Softmax
regression
Full connection
5×5 2×2 5×5 2×2
Convolution Maximum sampling Convolution Maximum sampling
" #
−1 n m n ðrÞ o
T ðr Þ
eθ k x 49 2 49 2
J ðθÞ = 〠 〠 l y = k log m T ðrÞ : ð11Þ Lclass = 〠 〠 F face
ij ðci − ̂
ci Þ2 + 〠 〠 F noface ðci − ̂ci Þ2 :
n r=1 k=1 ∑l=1 eθk x i=1 j=1
ij
i=1 j=1
ð14Þ
The indicative function in the formula is expressed as lfg
. The positioning error is expressed as Lposition , the classi-
On the basis of convolution neural network, in the neu- fication error is expressed as Lclass , whether a face detected
ral network based on Yolo model, except that the last layer in the i network is expressed as F, the center point of the
adopts linear function, the activation function of the remain- corresponding grid boundary box is expressed as j, the coor-
ing volume base layer is shown in formula dinate value of the point is expressed as ðx, yÞ, the width and
height of the boundary box of the point are expressed as w
( and h, respectively, and the category of the detection target
xx > 0,
A ðx Þ = ð12Þ is expressed as ci .
0:1xother: To complete the detection network defined by the loss
function, we need to constantly modify and adjust the
parameters of the full connection layer through the feedback
The image feature values extracted by the convolution
information of the training results, so as to continuously
neural network based on Yolo model will be fused with the
improve the confidence of the recognition network, finally
feature images obtained by the gradient direction histogram
achieve the design goal, and output the relevant bounding
feature extraction algorithm in the last volume base, so as to
box parameters of the face target. The loss function in this
obtain the final feature image. Then the final face features
paper is shown in formula
will be obtained after the two-layer full connection layer pro-
cessing in the neural network, but it cannot be used to realize 49 2 49 2
the real face target recognition and detection, and the classi- L = α1 Lposition + 〠 〠 F face
ij ðci − ̂
ci Þ2 + α2 〠 〠 F noface
ij ðci − ̂ci Þ2 :
fier of the recognition network needs to be further set and i=1 j=1 i=1 j=1
trained. The loss function needs to be defined before setting
ð15Þ
and training the classifier. The error of the network is
divided into positioning error and classification error. Its
calculation is shown in formulas The trained network needs to carry out corresponding
detection to evaluate the face target position contained in
each unit with the set confidence parameters, as shown in
49 2 2
Lposition = 〠 〠 F face ̂ i Þ2 + hi − ĥi
ðxi − x̂i Þ2 + ðyi − ̂yi Þ2 + ðwi − w , formula
ij
i=1 j=1
pred
ð13Þ Confidence = Pð faceÞ ⋅ IOU truth : ð16Þ
6 Wireless Communications and Mobile Computing
448 HOG
7 7
7 36 4096
7
3 56 7
3
448 28 14 7 7
112 192 256 512 1024 1024 1024 1060 12
0.60
0.55
0.50
0.45
Error
0.40
0.35
0.30
0.25
0 10 20 30 40 50
Training times
The probability of the existence of human face is the structure diagram of YOLO model in convolutional neu-
expressed as PðfaceÞ, and the coincidence degree of the pre- ral network based on YOLO model.
dicted labeled bounding box is expressed as IOU. Its defini-
tion is shown in formula 4. Experimental Results of Face Recognition
Soveriapped Model in College Intelligent Education
pred
IOU truth = , ð17Þ Management System
Spredict + Struth − Soverlapped
4.1. Experimental Test Results of Recognition Module in Face
where the area is expressed as S. As shown in Figure 3, it is Recognition System. In order for the face recognition model
Wireless Communications and Mobile Computing 7
100
90
80
70
60
50
40
30
20
10
0
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Literature algorithm1 Paper algorithm3
Literature algorithm2 Literature algorithm3
Query statistics
Generate report
Picture upload
Data upload
Fuzzy query
Accurate query
User login
to run stably and effectively in the intelligent education set are cut uniformly, and the size of the sum is used as the
management system of colleges and universities, corre- standard to divide the samples into positive and negative,
sponding tests need to be carried out. In actual face recogni- that is, those with a size less than 0.3 are negative samples,
tion, the orientation, illumination, and expression of the face those with a size more than 0.65 are positive samples, and
to be recognized in the detection environment are different, those not within this range are not used temporarily.
so these factors also need to be considered in the process of Figure 4 shows the error comparison results obtained by fus-
training. In the training process, the existing faces in the data ing hog and Yolo model algorithm models.
8 Wireless Communications and Mobile Computing
1.1 8.0%
1.0 7.0%
0.8 6.0%
5.0%
0.6
4.0%
0.4
3.0%
0.2
2.0%
0 1.0%
Fastest recognition Slowest Average Face recognition
time recognition time recognition time error time
Literature algorithm
Paper algorithm
8 in 1 1 0 0 1 1 1 1 1 1 1 78 Late
9 in 1 1 1 1 1 1 1 1 1 1 1 100
10 in 1 1 1 1 1 1 1 1 1 1 1 100
11 on 0 0 0 0 0 0 0 0 0 0 0 0 Absenteeism
12 in 1 1 1 1 1 1 1 1 1 1 1 100
It can be seen from the results in the figure that the face with other face recognition algorithms through fddb data set.
detection algorithm model adopted in this paper only propa- As shown in Figure 5, the test results are compared.
gates complex samples, which can save computing resources The horizontal coordinate in Figure 5 is the recognition
on the basis of maintaining a high recognition rate. In order speed, and the vertical coordinate is the recognition accuracy.
to further test the performance of the recognition network in It can be seen from the results in the figure that the algorithm
this paper, the algorithm in this paper is tested and compared in this paper is not the best algorithm among the four
Wireless Communications and Mobile Computing 9
algorithms in terms of recognition speed or recognition accu- To sum up, the algorithm of the face recognition system
racy, but the algorithm in this paper has the advantages that in this paper has good real-time performance, can realize
the other three algorithms do not have in terms of recognition face recognition in the expected time, improve the efficiency
speed and recognition accuracy. This is mainly because the of face recognition and reduce the error rate of face recogni-
fusion of Yolo model and hog fast extraction method shows tion, and can meet the needs of face recognition in different
good timeliness, and the final accuracy of this algorithm can environments. In the application test of dormitory manage-
reach 91.5%. Therefore, this algorithm can be applied to the ment and classroom attendance model, the application of
environment of face recognition in the intelligent education face recognition technology can improve the efficiency of
management system of colleges and universities. students’ education and management and achieve the pur-
As shown in Figure 6, the results of each test example of pose of effectively managing students.
the face recognition algorithm in the university intelligent
education management system are shown. As can be seen
from the figure, the main test examples include user login, 5. Conclusion
accurate query, fuzzy query, data and picture upload, report
With the popularization of higher education, the number of
generation, and query statistics. It can be seen from the test
college students and the scale of university campus are
results that the test response time displayed by each example
expanding. With the development of higher education, the
is within the expected range, which meets the expected
problem of campus education management is becoming
requirements of face recognition test.
more and more prominent. With the development of the
Internet technology and information technology, the con-
4.2. Application Test Results of Face Recognition System. In
struction of intelligent education management system and
the intelligent education management system of colleges
platform has become an inevitable development trend.
and universities, the learning management and safety man-
Therefore, this paper puts forward the application research
agement of college students are the focus. Therefore, this
of artificial intelligence technology in college intelligent edu-
paper will test the application of face recognition system in
cation management and applies face recognition technology
two aspects: attendance system and dormitory management.
in college intelligent education management system, dormi-
As shown in Figure 7, the face recognition test results of the
tory management system, and classroom attendance system.
dormitory environment of the face recognition algorithm in
The experimental results show that the convolution neural
this paper and the traditional algorithm in the literature are
network algorithm based on Yolo model combined with
compared. It can be seen from the demerit in the figure that
hog algorithm can effectively improve the extraction of fea-
the fastest recognition time of this face recognition algo-
ture information of face recognition, improve the recogni-
rithm is slightly longer than that of the traditional algorithm,
tion efficiency, and reduce the error rate of face
while the slowest recognition time is shorter than that of the
recognition. In addition, the face recognition algorithm in
traditional algorithm. Finally, in the average recognition
this paper has more advantages in the accuracy of face rec-
time, the recognition time of this face recognition algorithm
ognition than the traditional algorithm in the dormitory
has a little advantage over that of the traditional face recog-
management experiment. It can meet the real-time perfor-
nition algorithm, but it is not obvious. However, from the
mance of face recognition and improve the accuracy of face
perspective of the error rate of face recognition, the recogni-
recognition. In the classroom attendance experiment, the
tion error rate of this face recognition algorithm is less than
face recognition algorithm in this paper can identify and
that of the traditional algorithm, which is reduced by 3.27%.
deal with the problems such as students’ late and early
Overall, the face recognition algorithm in this paper has cer-
departure, absenteeism, and substitute classes. However,
tain advantages in application testing, which can meet the
there are also problems that cannot be recognized correctly
requirements of real-time face recognition and improve the
due to students’ sitting posture and face angle. Therefore,
accuracy of face recognition.
this aspect needs further improvement and research.
Figure 8 shows the test results of face recognition in
classroom attendance. In the classroom attendance test,
twelve college students are mainly selected as the test objects Data Availability
of face recognition. From the results in the figure, it can be
seen that the classroom attendance module based on face The figures used to support the findings of this study are
recognition system can identify and deal with the situations included in the article.
of late and early leave, substitute and absenteeism in the
classroom, and can meet the expected needs of classroom
attendance recognition. In the test process, it is found that Conflicts of Interest
the efficiency of face detection and recognition is relatively
The authors declare that they have no conflicts of interest.
high when college students maintain correct posture and
face is illuminated evenly. If the student’s posture angle is
too biased or the face deflection angle is large, there will be Acknowledgments
a recognition blind area, and face recognition will fail, result-
ing in the system judging that there is no corresponding face The authors would like to show sincere thanks to those tech-
in the image, and the recognition result score is zero. niques that have contributed to this research.
10 Wireless Communications and Mobile Computing