ArticlePDF Available

High-Capacity Real-Time Face Retrieval Recognition Algorithm Based on Task Scheduling Model for the Treatment Area of Hospital

Authors:

Abstract and Figures

This paper presents an in-depth study of face detection, face feature extraction, and face classification from three important components of a high-capacity face recognition system for the treatment area of hospital and a study of a high-capacity real-time face retrieval and recognition algorithm for the treatment area of hospital based on a task scheduling model. Considering the real-time nature of our system, our face feature extraction network is modeled by DeepID, and the network is slightly improved by introducing a central loss verification signal to train a DeepID-like network model using central loss and use it to extract face features. To further investigate and optimize the schedulability analysis problem of the directed graph real-time task model, this paper proposes a rigorous and approximate response time analysis method for the directed graph real-time task model with an arbitrary time frame. Based on the theoretical results of the greatly additive algebra, it is shown that the coherent qualifying function is linearly periodic, i.e., the function can be represented by a finite nonperiodic part and an infinitely repeated periodic part, thus calculating the coherent qualifying function independent of the magnitude of the interval time. The algorithm for high-capacity real-time face retrieval and recognition in the treatment area of hospital based on the task scheduling model is further investigated, and a face database is established by using the PCA dimensionality reduction technique. Based on the internal architecture of the processor, image preprocessing and IP core packaging are implemented, and the hardware engineering of the high-capacity real-time face recognition system for hospital visits is built using the IP-based design concept. The performance tests of the face detection model and feature extraction network show that the face detection model has a significant reduction in false-positive rate, better fitting of border regression, and improved time performance. The face feature extraction network has no overfitting, and the features are highly discriminative with small feature extraction time consumption. The high-capacity real-time face recognition system for the treatment area of hospital combined with the optimized directed graph task scheduling model can approach 25 fps, which meets the real-time requirements, and the face recognition rate surpasses that of real people. It realizes the intelligence, self-help, and autonomy of medical services and satisfies the medical needs of users in all aspects. 1. Introduction In the past half-century, biometric identification technology has developed rapidly because each biological individual has unique biometric points that can be measured, identified, and verified, and biometric identification technology identifies and authenticates individuals based on these unique points. Unlike traditional identification which is easily damaged, easily lost, and easily falsified, biometric identification technology which utilizes the unique characteristics of biological individuals is more convenient, fast, safe, reliable, and accurate. Biometric identification technology can be divided into two categories, one is identification by inherent characteristics of biological individuals, and the other is identification by behavioral characteristics of biologicals [1, 2]. Inherent characteristic identification currently has features like fingerprint, hand type, iris, retina, pulse, and face and has been used as biometric identification, and behavioral characteristics like voice and signature have been used as biometric identification. As a kind of biometric technology, face recognition has its unique advantages such as nonintrusiveness, convenience, friendliness, noncontact, and scalability compared with other biometric technologies in practical use scenarios, which makes it have an important position in biometric identification [3, 4]. If the time constraints of the real-time system are met, the system is said to be schedulable, that is, each task is completed before its own time limit. Therefore, it is required to be able to determine in advance whether it is schedulable [5]. If the result is not schedulable, then you need to increase system resources or reduce task load so that the system becomes schedulable again. Face recognition has become the current research hotspot in the field of artificial intelligence due to its huge application prospect. With the development of science and technology and the progress of society, the technical demand for conducting fast, efficient, and automatic face recognition is becoming more and more urgent. Face recognition technology first determines whether there is a face in the image or not, and if there is a face, the features of the face are extracted and matched with the known face comparison, and then, the identity of the face is recognized [6]. The face recognition process is generally divided into four steps: face detection, face feature extraction, face matching, and recognition. Among them, face feature extraction is the most important part of face recognition, and it is good or bad directly affects the final result of our face recognition. From another perspective, face recognition is finding a way to describe faces, and this description needs to have strong interclass distinguishability and intraclass stability. There are neural network-based methods, statistical methods, geometric feature-based methods, model-based methods, and multiclassifier integration methods for face recognition, among which the vast majority of traditional face recognition feature expressions are human selected; however, in practice, human selection of features is a very time-consuming and energy-consuming matter, and the good or bad selection depends largely on experience and luck. However, with the increasing electronic, automated, and connected control systems, the functions and performance requirements undertaken by real-time systems are becoming more and more demanding. At the same time, the type, number, and a load of tasks in the system have increased dramatically, which poses a great challenge to the time verification of real-time systems [7]. Therefore, it is of great significance to research schedulability analysis and performance optimization of complex real-time task systems. This paper will focus on the direction of the task scheduling model for a high-capacity real-time face retrieval recognition algorithm for hospital visits. The existing algorithms for large-capacity real-time face retrieval and recognition inhospital visit sites match sparse datasets, which makes the large-capacity heterogeneous face recognition inhospital visit sites cannot achieve the performance of visible face recognition when using the same task scheduling model. At the same time, the existing face recognition task scheduling model is large and cannot meet the needs of offline inference computation of edge devices nowadays. By investigating the multitask scheduling method to make the large-capacity real-time face recognition model complete the task quickly and accurately on the embedded platform is the purpose of this paper. Chapter 1 is the introduction, which briefly introduces the background and significance of the study of the real-time face recognition system and finally summarizes the main work of this paper and the structure of the paper. The second chapter is related work, briefly describes the status of domestic and international research, and both face detection and feature extraction are obtained using task scheduling model, and finally, it also introduces how to improve task scheduling performance by adjusting hyperparameters. Chapter 3 firstly introduces the research on the recognition algorithm of large-capacity real-time face retrieval inhospital visit sites based on task scheduling model; then, we select the current multitask scheduling model with very high real time and performance, improve it, and introduce better recognition methods for large-capacity real-time face retrieval inhospital visit sites. Chapter 4 is the result analysis, through the hospital visit place high-capacity real-time face retrieval recognition algorithm scheduling analysis, hospital visit place high-capacity real-time face retrieval recognition algorithm performance analysis, and hospital visit place high-capacity real-time face retrieval recognition algorithm system analysis, which proves the better performance and efficiency of the algorithm researched in this paper, the relevant test of our face recognition system, and the analysis of our face recognition system performance and result presentation. Chapter 5 concludes with a summary of the research work in this paper, pointing out the shortcomings of the scheme, proposing corresponding improvement methods, and looking ahead. 2. Related Work Since individual task scheduling is computationally limited, most problems require multiple task schedulers to work together to solve the target problem, hence the emergence of multitasking scheduling systems. Multitask scheduling systems have considerable robustness and adaptability and are more efficient for solving practical application problems. Addagarla et al. combined binary particle swarm optimization with contract networks and used the synergy of task scheduling to improve the accuracy of high-capacity real-time face retrieval recognition and reduce the scheduling time in the treatment area of hospital. The dynamic path planning method is accomplished by using the dynamic and reactive nature of task scheduling and adding it to the ant colony algorithm [8]. Bartolini and Patella incorporated a microgrid composed of distributed power sources into a communication network formed by multitask scheduling technology, which enabled the system to ensure stable system power output even when the load changed drastically [9]. The multitask scheduling technique is used to coordinate and control the large-capacity real-time face retrieval and recognition in a hospital visit site, and the multitask scheduling technique is incorporated into it to enhance the communication function of the large-capacity real-time face retrieval and recognition in a hospital visit site and to maintain a trouble-free retrieval and recognition [10]. The information captured by task scheduling is processed by applying convolutional fusion. The task scheduling technique is incorporated into semantic Web services to make the discovery, invocation, and combination of Web services more efficient, and the multitask scheduling technique is introduced into deep reinforcement learning to solve the problem that the policy gradient increases with the number of task scheduling [11]. However, the rapid growth of face recognition technology in China in recent years and the booming of task scheduling have forced major research institutes, companies, and universities to invest heavily in face recognition [12]. Choi and Cha proposed a transformation algorithm using Eigen to reach the transformation from face sketch to face photo and then perform face matching on the face photo obtained from the change [13]. A method using locally linear embedding is proposed to transform the face sketch into a real photo image and the near-infrared light face image into a visible light face image, respectively, and then use a general method to match the face recognition [14]. To consider the relationship between face facial image regions and their neighboring regions in a comprehensive way, Markov random field model is added to the project of heterogeneous face recognition [15]. In this approach, since only the best candidate regions are selected for training, this leads to the deformation of facial images. Kumar et al. proposed a Markov weight field model, which can solve the problem of image deformation by selecting some selected regions to form a Markov network model [16]. Muhammad et al. proposed a direct push learning method using Markov random fields, which combines sketch images with real photo images of the test images into the learning process, thus reducing the error of the model on the test data [17]. This paper proposes and studies a task model for dynamically adjusting the switching speed, which is a strict extension of the static adaptive rate-of-change task model [18]. A schedulability analysis method was developed for such a new real-time task model. This method introduces a new directed graph task model to safely approximate the execution load of this type of rotation task. Compared with the existing static analysis methods, the proposed schedulability analysis method has higher accuracy and can effectively reduce the time complexity [19]. Hospital visit site high-capacity real-time face retrieval recognition is a process algorithm used to describe concurrent systems, and its simple and efficient algorithmic model makes hospital visit site high-capacity real-time face retrieval recognition the basis for modeling in several fields. The CCS is promoted for high-capacity real-time face retrieval recognition at hospital sites and can be used to represent communication topologies with dynamic change capability by transmitting channels between distributed nodes along other channels, which is very powerful for formal description. Since a change in the physical location of processors in a distributed system may bring about a change in the communication topology, it is particularly suitable to describe a concurrent programming model for task scheduling using high-capacity real-time face retrieval recognition inhospital visit sites. The goal of this paper is to combine static task scheduling algorithms with multitask scheduling techniques and high-capacity real-time face retrieval recognition inhospital visits, so that task scheduling in a distributed environment can better utilize environmental resources, not only to enhance concurrency and reduce communication overhead by using multitask scheduling techniques but also to combine the simulated communication capability of high-capacity real-time face retrieval recognition inhospital visits [20]. It defines a task scheduling model for uncertainty-oriented concurrent systems that reacts faster to network topology changes and is important for efficient utilization of resources in distributed systems. 3. Intelligent Analysis of Supply Chain Coordination Based on the Internet of Things 3.1. High-Capacity Real-Time Face Information Detection and Extraction for Hospital Consultation Places In the face of multipose face detection task, the organization of the classifier is an important issue, especially when the features of faces in different poses show great differences, which will bring a great challenge to the detector; to solve this problem, a partitioning strategy is usually used, and the classifiers are trained separately for faces in different poses and then combined to build a multipose face detector. The partitioning strategy is the most basic strategy to deal with the multipose face detection task, but it also needs to take into account the speed and classification accuracy, and the enhancement of classification capability will bring an increase in computational cost. We design Equation (1) by combining the localization error Dc and classification error Fc of the samples, the prediction probability of classification, and the intersection rate between the border and the actual target border and calculate a surrogate value cost for the border containing positive samples in the grid, based on which we judge whether the sample is selected to join the training. The positioning error Dc is defined as Equation (2). The classification error Fc is defined as Equation (3). In Equations (2) and (3), is a certain threshold value of the intersection rate. is a coefficient that controls the sensitivity of the localization error Dc to the threshold value ; the inputs and for both costs take values between 0 and 1. The purpose of DeepID2 for learning is to obtain the output vector, not to maximize the recognition rate; thus, the class spacing term is added to the paper. To facilitate the representation, we express the output vector obtained by DeepID2 layer extraction as a function, as in Equation (4), where denotes the extracted DeepID2 feature vector, the function denotes the feature extraction function, denotes the input face slice, and denotes the network parameters. The function for correct classification is the objective function of softmax, which is aimed at minimizing the cross-entropy and is defined to identify the signal, as in Equation (5), where is the DeepID2 feature vector of Equation (4), denotes the classification target, denotes the softmax layer parameter, denotes the target probability distribution, and denotes the classification prediction probability distribution. The objective function for maximizing the sample spacing is as follows, defined as the validation signal, as in Equation (6), where means that and belong to the same person, and this is when we need to minimize the interclass spacing; means that and belong to different people, and this is when we need to minimize the difference between and their distance values. The is a parameter that needs to be adjusted manually, and the purpose of proposing is that the objective function needs to be minimized, not maximized. Since the validation signal needs two samples to be calculated, the training process of the whole network needs to be changed accordingly. Two samples are randomly selected for each iteration during training and then trained. Our goal is to learn the parameter updated by stochastic gradient descent. 3.2. High-Capacity Real-Time Face Retrieval Recognition Algorithm Based on Task Scheduling Model for the Treatment Area of Hospital The multiframe task can be described by the minimum release interval , the execution time vector , and the relative time frame . A typical example of a multiframe task model is an MPEG video stream decoding application using multiple frame types, capable of simulating the periodic arrival of video frames and different frames with different decoding times. As shown in Figure 1, a multiframe task can be represented by a graph with a chain structure (Figure 1(a)), where the solid nodes represent the entry points for task execution. In addition, the execution time of the multiframe task is variable, but the release pattern and the relative time frame are fixed (Figure 1(b)). (a)
This content is subject to copyright. Terms and conditions apply.
Research Article
High-Capacity Real-Time Face Retrieval Recognition Algorithm
Based on Task Scheduling Model for the Treatment
Area of Hospital
Yi Zhou and Weili Xia
School of Management, Northwestern Polytechnical University, Xian 710072, China
Correspondence should be addressed to Yi Zhou; zhouyi1202@mail.nwpu.edu.cn
Received 29 October 2021; Revised 17 November 2021; Accepted 22 November 2021; Published 6 December 2021
Academic Editor: Miaochao Chen
Copyright © 2021 Yi Zhou and Weili Xia. 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.
This paper presents an in-depth study of face detection, face feature extraction, and face classication from three important
components of a high-capacity face recognition system for the treatment area of hospital and a study of a high-capacity real-
time face retrieval and recognition algorithm for the treatment area of hospital based on a task scheduling model. Considering
the real-time nature of our system, our face feature extraction network is modeled by DeepID, and the network is slightly
improved by introducing a central loss verication signal to train a DeepID-like network model using central loss and use it to
extract face features. To further investigate and optimize the schedulability analysis problem of the directed graph real-time
task model, this paper proposes a rigorous and approximate response time analysis method for the directed graph real-time
task model with an arbitrary time frame. Based on the theoretical results of the greatly additive algebra, it is shown that the
coherent qualifying function is linearly periodic, i.e., the function can be represented by a nite nonperiodic part and an
innitely repeated periodic part, thus calculating the coherent qualifying function independent of the magnitude of the interval
time. The algorithm for high-capacity real-time face retrieval and recognition in the treatment area of hospital based on the
task scheduling model is further investigated, and a face database is established by using the PCA dimensionality reduction
technique. Based on the internal architecture of the processor, image preprocessing and IP core packaging are implemented,
and the hardware engineering of the high-capacity real-time face recognition system for hospital visits is built using the IP-
based design concept. The performance tests of the face detection model and feature extraction network show that the face
detection model has a signicant reduction in false-positive rate, better tting of border regression, and improved time
performance. The face feature extraction network has no overtting, and the features are highly discriminative with small
feature extraction time consumption. The high-capacity real-time face recognition system for the treatment area of hospital
combined with the optimized directed graph task scheduling model can approach 25 fps, which meets the real-time
requirements, and the face recognition rate surpasses that of real people. It realizes the intelligence, self-help, and autonomy of
medical services and satises the medical needs of users in all aspects.
1. Introduction
In the past half-century, biometric identication technology
has developed rapidly because each biological individual has
unique biometric points that can be measured, identied,
and veried, and biometric identication technology iden-
ties and authenticates individuals based on these unique
points. Unlike traditional identication which is easily dam-
aged, easily lost, and easily falsied, biometric identication
technology which utilizes the unique characteristics of bio-
logical individuals is more convenient, fast, safe, reliable,
and accurate. Biometric identication technology can be
divided into two categories, one is identication by inherent
characteristics of biological individuals, and the other is
identication by behavioral characteristics of biologicals [1,
2]. Inherent characteristic identication currently has fea-
tures like ngerprint, hand type, iris, retina, pulse, and face
and has been used as biometric identication, and
Hindawi
Advances in Mathematical Physics
Volume 2021, Article ID 1547025, 11 pages
https://doi.org/10.1155/2021/1547025
behavioral characteristics like voice and signature have been
used as biometric identication. As a kind of biometric tech-
nology, face recognition has its unique advantages such as
nonintrusiveness, convenience, friendliness, noncontact,
and scalability compared with other biometric technologies
in practical use scenarios, which makes it have an important
position in biometric identication [3, 4]. If the time con-
straints of the real-time system are met, the system is said
to be schedulable, that is, each task is completed before its
own time limit. Therefore, it is required to be able to deter-
mine in advance whether it is schedulable [5]. If the result is
not schedulable, then you need to increase system resources
or reduce task load so that the system becomes schedulable
again.
Face recognition has become the current research hotspot
in the eld of articial intelligence due to its huge application
prospect. With the development of science and technology
and the progress of society, the technical demand for con-
ducting fast, ecient, and automatic face recognition is
becoming more and more urgent. Face recognition technology
rst determines whether there is a face inthe image or not, and
if there is a face, the features of the face are extracted and
matched with the known face comparison, and then, the iden-
tity of the face is recognized [6]. The face recognition process
is generally divided into four steps: face detection, face feature
extraction, face matching, and recognition. Among them, face
feature extraction is the most important part of face recogni-
tion, and it is good or bad directly aects the nal result of
our face recognition. From another perspective, face recogni-
tion is nding a way to describe faces, and this description
needs to have strong interclass distinguishability and intraclass
stability. There are neural network-based methods, statistical
methods, geometric feature-based methods, model-based
methods, and multiclassier integration methods for face rec-
ognition, among which the vast majority of traditional face
recognition feature expressions are human selected; however,
in practice, human selection of features is a very time-
consuming and energy-consuming matter, and the good or
bad selection depends largely on experience and luck. How-
ever, with the increasing electronic, automated, and connected
control systems, the functions and performance requirements
undertaken by real-time systems are becoming more and
more demanding. At the same time, the type, number, and a
load of tasks in the system have increased dramatically, which
poses a great challenge to the time verication of real-time sys-
tems [7]. Therefore, it is of great signicance to research sche-
dulability analysis and performance optimization of complex
real-time task systems.
This paper will focus on the direction of the task sched-
uling model for a high-capacity real-time face retrieval rec-
ognition algorithm for hospital visits. The existing
algorithms for large-capacity real-time face retrieval and rec-
ognition inhospital visit sites match sparse datasets, which
makes the large-capacity heterogeneous face recognition
inhospital visit sites cannot achieve the performance of visi-
ble face recognition when using the same task scheduling
model. At the same time, the existing face recognition task
scheduling model is large and cannot meet the needs of o-
line inference computation of edge devices nowadays. By
investigating the multitask scheduling method to make the
large-capacity real-time face recognition model complete
the task quickly and accurately on the embedded platform
is the purpose of this paper. Chapter 1 is the introduction,
which briey introduces the background and signicance
of the study of the real-time face recognition system and
nally summarizes the main work of this paper and the
structure of the paper. The second chapter is related work,
briey describes the status of domestic and international
research, and both face detection and feature extraction are
obtained using task scheduling model, and nally, it also
introduces how to improve task scheduling performance by
adjusting hyperparameters. Chapter 3 rstly introduces the
research on the recognition algorithm of large-capacity
real-time face retrieval inhospital visit sites based on task
scheduling model; then, we select the current multitask
scheduling model with very high real time and performance,
improve it, and introduce better recognition methods for
large-capacity real-time face retrieval inhospital visit sites.
Chapter 4 is the result analysis, through the hospital visit
place high-capacity real-time face retrieval recognition algo-
rithm scheduling analysis, hospital visit place high-capacity
real-time face retrieval recognition algorithm performance
analysis, and hospital visit place high-capacity real-time face
retrieval recognition algorithm system analysis, which
proves the better performance and eciency of the algo-
rithm researched in this paper, the relevant test of our face
recognition system, and the analysis of our face recognition
system performance and result presentation. Chapter 5 con-
cludes with a summary of the research work in this paper,
pointing out the shortcomings of the scheme, proposing cor-
responding improvement methods, and looking ahead.
2. Related Work
Since individual task scheduling is computationally limited,
most problems require multiple task schedulers to work
together to solve the target problem, hence the emergence
of multitasking scheduling systems. Multitask scheduling
systems have considerable robustness and adaptability and
are more ecient for solving practical application problems.
Addagarla et al. combined binary particle swarm optimiza-
tion with contract networks and used the synergy of task
scheduling to improve the accuracy of high-capacity real-
time face retrieval recognition and reduce the scheduling
time in the treatment area of hospital. The dynamic path
planning method is accomplished by using the dynamic
and reactive nature of task scheduling and adding it to the
ant colony algorithm [8]. Bartolini and Patella incorporated
a microgrid composed of distributed power sources into a
communication network formed by multitask scheduling
technology, which enabled the system to ensure stable sys-
tem power output even when the load changed drastically
[9]. The multitask scheduling technique is used to coordi-
nate and control the large-capacity real-time face retrieval
and recognition in a hospital visit site, and the multitask
scheduling technique is incorporated into it to enhance the
communication function of the large-capacity real-time face
retrieval and recognition in a hospital visit site and to
2 Advances in Mathematical Physics
maintain a trouble-free retrieval and recognition [10]. The
information captured by task scheduling is processed by
applying convolutional fusion. The task scheduling tech-
nique is incorporated into semantic Web services to make
the discovery, invocation, and combination of Web services
more ecient, and the multitask scheduling technique is
introduced into deep reinforcement learning to solve the
problem that the policy gradient increases with the number
of task scheduling [11].
However, the rapid growth of face recognition technol-
ogy in China in recent years and the booming of task sched-
uling have forced major research institutes, companies, and
universities to invest heavily in face recognition [12]. Choi
and Cha proposed a transformation algorithm using Eigen
to reach the transformation from face sketch to face photo
and then perform face matching on the face photo obtained
from the change [13]. A method using locally linear embed-
ding is proposed to transform the face sketch into a real
photo image and the near-infrared light face image into a
visible light face image, respectively, and then use a general
method to match the face recognition [14]. To consider the
relationship between face facial image regions and their
neighboring regions in a comprehensive way, Markov ran-
dom eld model is added to the project of heterogeneous
face recognition [15]. In this approach, since only the best
candidate regions are selected for training, this leads to the
deformation of facial images. Kumar et al. proposed a Mar-
kov weight eld model, which can solve the problem of
image deformation by selecting some selected regions to
form a Markov network model [16]. Muhammad et al. pro-
posed a direct push learning method using Markov random
elds, which combines sketch images with real photo images
of the test images into the learning process, thus reducing
the error of the model on the test data [17].
This paper proposes and studies a task model for
dynamically adjusting the switching speed, which is a strict
extension of the static adaptive rate-of-change task model
[18]. A schedulability analysis method was developed for
such a new real-time task model. This method introduces a
new directed graph task model to safely approximate the
execution load of this type of rotation task. Compared with
the existing static analysis methods, the proposed schedul-
ability analysis method has higher accuracy and can eec-
tively reduce the time complexity [19]. Hospital visit site
high-capacity real-time face retrieval recognition is a process
algorithm used to describe concurrent systems, and its sim-
ple and ecient algorithmic model makes hospital visit site
high-capacity real-time face retrieval recognition the basis
for modeling in several elds. The CCS is promoted for
high-capacity real-time face retrieval recognition at hospital
sites and can be used to represent communication topologies
with dynamic change capability by transmitting channels
between distributed nodes along other channels, which is
very powerful for formal description. Since a change in the
physical location of processors in a distributed system may
bring about a change in the communication topology, it is
particularly suitable to describe a concurrent programming
model for task scheduling using high-capacity real-time face
retrieval recognition inhospital visit sites. The goal of this
paper is to combine static task scheduling algorithms with
multitask scheduling techniques and high-capacity real-
time face retrieval recognition inhospital visits, so that task
scheduling in a distributed environment can better utilize
environmental resources, not only to enhance concurrency
and reduce communication overhead by using multitask
scheduling techniques but also to combine the simulated
communication capability of high-capacity real-time face
retrieval recognition inhospital visits [20]. It denes a task
scheduling model for uncertainty-oriented concurrent sys-
tems that reacts faster to network topology changes and is
important for ecient utilization of resources in distributed
systems.
3. Intelligent Analysis of Supply Chain
Coordination Based on the
Internet of Things
3.1. High-Capacity Real-Time Face Information Detection
and Extraction for Hospital Consultation Places. In the face
of multipose face detection task, the organization of the clas-
sier is an important issue, especially when the features of
faces in dierent poses show great dierences, which will
bring a great challenge to the detector; to solve this problem,
a partitioning strategy is usually used, and the classiers are
trained separately for faces in dierent poses and then com-
bined to build a multipose face detector. The partitioning
strategy is the most basic strategy to deal with the multipose
face detection task, but it also needs to take into account the
speed and classication accuracy, and the enhancement of
classication capability will bring an increase in computa-
tional cost.
We design Equation (1) by combining the localization
error Dc and classication error Fc of the samples, ythe pre-
diction probability of classication, and xthe intersection
rate between the border and the actual target border and cal-
culate a surrogate value cost for the border containing posi-
tive samples in the grid, based on which we judge whether
the sample is selected to join the training.
Cx,y
ðÞ
=Dc x
ðÞ
+Fc y
ðÞ
:ð1Þ
The positioning error Dc is dened as Equation (2).
Dc x
ðÞ
=1
1 + lim
SS
i=1ixt
ðÞ
x+t
ðÞ
:ð2Þ
The classication error Fc is dened as Equation (3).
Fc y
ðÞ
=1y
ðÞ
lim
S
S
i=1
iyt
ðÞ
y+t
ðÞ
:ð3Þ
In Equations (2) and (3), tis a certain threshold value of
the intersection rate. iis a coecient that controls the sensi-
tivity of the localization error Dc to the threshold value t; the
inputs xand yfor both costs take values between 0 and 1.
3Advances in Mathematical Physics
The purpose of DeepID2 for learning is to obtain the
output vector, not to maximize the recognition rate; thus,
the class spacing term is added to the paper. To facilitate
the representation, we express the output vector obtained
by DeepID2 layer extraction as a function, as in Equation
(4), where TðxÞdenotes the extracted DeepID2 feature vec-
tor, the tfunction denotes the feature extraction function, x
denotes the input face slice, and βdenotes the network
parameters.
Tx
ðÞ
=tx,β
ðÞ
:ð4Þ
The function for correct classication is the objective
function of softmax, which is aimed at minimizing the
cross-entropy and is dened to identify the signal, as in
Equation (5), where Tis the DeepID2 feature vector of
Equation (4), tdenotes the classication target, βidenotes
the softmax layer parameter, pðiÞdenotes the target proba-
bility distribution, and pðtÞdenotes the classication predic-
tion probability distribution.
ST,t,βi
ðÞ
= lim
N
N
i=1
pi
ðÞln pi
ðÞ=ln pt
ðÞ
:ð5Þ
The objective function for maximizing the sample spac-
ing is as follows, dened as the validation signal, as in Equa-
tion (6), where sij =1 means that fi and fj belong to the
same person, and this is when we need to minimize the
interclass spacing; sij =1means that Tiand Tjbelong to
dierent people, and this is when we need to minimize the
dierence between mand their distance values. The mis a
parameter that needs to be adjusted manually, and the pur-
pose of proposing mis that the objective function needs to
be minimized, not maximized. Since the validation signal
needs two samples to be calculated, the training process of
the whole network needs to be changed accordingly. Two
samples are randomly selected for each iteration during
training and then trained. Our goal is to learn the parameter
βij updated by stochastic gradient descent.
HT
i,Tj,sij,βij

=
max 0, TiTj
2/2, mTiTj
2

2:
ð6Þ
3.2. High-Capacity Real-Time Face Retrieval Recognition
Algorithm Based on Task Scheduling Model for the
Treatment Area of Hospital. The multiframe task ican be
described by the minimum release interval Ti, the execution
time vector Ci =ðCi,1;⋯⋯ ;Ci,miÞ, and the relative time
frame Di =Ti. A typical example of a multiframe task model
is an MPEG video stream decoding application using multi-
ple frame types, capable of simulating the periodic arrival of
video frames and dierent frames with dierent decoding
times. As shown in Figure 1, a multiframe task can be repre-
sented by a graph with a chain structure (Figure 1(a)), where
the solid nodes represent the entry points for task execution.
In addition, the execution time of the multiframe task is var-
iable, but the release pattern and the relative time frame are
xed (Figure 1(b)).
For nindependent multiframe tasks running on a pro-
cessor with a scheduling policy using the preemptive static
priority RM algorithm, the system is schedulable if Equation
(7) is satised.
Ux
ðÞ
=1+
1
x

1/n
1
"#
x+n
ðÞ
xn
ðÞ
:ð7Þ
Each node xiX(or task type) is represented by an
ordered pair heðxiÞ,dðxiÞi, where eðxiÞand dðxiÞare the
worst-case execution time and the relative time frame,
respectively. The edges represent the possible control ow
directions, i.e., the order of releasing tasks. An edge ðxi,yjÞ
Kis described by the parameter pðxi,yjÞ, which is the
A
B
C
DD
BDF
CE
AG
Graphical representation of multi-frame tasks
(a)
Release mode of multi-frame tasks
T1,C1 T2,C2 T3,C3
T2,C2 T3,C3T1,C1
Second
frame
First frame ird
frame
Second
frame
ird
frame
First frame
Task model one
Task model two
(b)
Figure 1: Graph representation and release pattern for multiframe tasks.
4 Advances in Mathematical Physics
minimum release interval time for xiand yj. We assume that
all parameters are positive integers, and that the WCET of
each node is not greater than the minimum release interval
of all its outgoing edges, i.e., it satises Equation (8).
xi,yj

K,ex
i
ðÞ
px
i,yi
ðÞ
:ð8Þ
For the same path t=fx1,x2⋯⋯xngin task Tfunc-
tions f, there exists a vertical to slope one substitution rela-
tion dened as follows: for any vertical part of fas in
Equation (9).
fT,x
ðÞ
=Tft
ðÞ
+ψ:ð9Þ
For task T, the maximum cumulative execution request
of its task release in an interval of arbitrary length tis
dened as its request-limited function, as in Equation (10).
gT,x
ðÞ
= max
βDT
ðÞTf t
ðÞ1β
ðÞ
1+β
ðÞðÞ
:ð10Þ
For task T, its task release is dened as its coherence-
limited function in terms of the maximum number of execu-
tion loads in an interval of arbitrary length t, as in Equation
(11).
hT,x
ðÞ
= max
βDT
ðÞ Tg t
ðÞ 1β
ðÞ
1+β
ðÞ
lim
NN
i=1βi
0
@1
A
0
@1
A:ð11Þ
For task T, whose tasks are released and the time frame
is within interval t, its maximum cumulative execution
demand is dened as its demand-limited function, as in
Equation (12).
yT,x
ðÞ
= max
βDT
ðÞht
ðÞ1β
ðÞ
1+β
ðÞðÞ
:ð12Þ
Intuitively, the fðÞ and gðÞ functions quantify the maxi-
mum cumulative execution request and load, respectively,
while the yðÞ function measures the size of the execution
time that a task is released and must be completed within
a given interval. Like the fðÞ function, the gðÞ function is a
nondecreasing step function whose horizontal part is left-
open and right-closed. At the same time, the yðÞ function
inherits the properties of yand is also a continuous sloping
step function. Like the gðÞ function, the gðÞ function is a
step function whose horizontal part is left-closed and right-
open.
To reduce the complexity of computation and to achieve
the purpose of real-time processing, dimensionality reduc-
tion methods are used in face recognition. The commonly
used mainstream linear transformation method for dimen-
sionality reduction is principal component analysis (PCA),
which is derived from K-L transformation, and its essence
is to protect the data in the high-dimensional space into
the low-dimensional space through a linear transformation
while representing the original data as well as possible. In
the process of real-time face recognition, assuming that the
fps (frames per second) of the currently captured video is
30 Hz and the resolution of the video image is m. Then, at
a certain time, there will be nimages, and mis the dimen-
sion of each image. That is, in the m-dimensional space,
there exist nsamples as in Equation (13). The dimensional-
ity of the data is too high, it is time-consuming and laborious
to process, so I wonder if I can only process part of the
dimensions, and the results obtained are consistent with
the results of all dimensions. To put it simply, a picture
has 2000 feature dimensions, but in fact, only 100 of them
(or even less) have a huge impact on the result.
M= lim
n
n
i=1
xi
n
:ð13Þ
Assuming that 1 line is to be used to represent these n
samples and let edenote the unit direction vector of this line
through the mean, the equation of the line can be expressed
as Equation (14), where ais a real scalar representing the
distance of a point on the line from the point m.
dx
ðÞ
=dm
ðÞ
+αψ:ð14Þ
A linear projection of the nsamples in the direction of
the eigenvector corresponding to the maximum eigenvalue
of the walk matrix yields a 1-dimensional representation ia
in the least square error sense, and this projection transfor-
mation is the transformation of the bases. In the m-dimen-
sional space, the mbases are unit vectors ψiin the
direction of each coordinate axis ði=1,2nÞ, and a certain
sample xiðxi1,xi2ximÞin the space can be represented by
this set of bases as Equation (15).
xi= lim
m
m
i=1
xijψi+ lim
n
n
j=1
xijψi:ð15Þ
A nonlinear activation function with a segmented linear
activation function (Hard_Swish) is introduced in the Light-
facenet model, which can signicantly improve the accuracy
of the neural network for classication tasks when using
Hard_Swish instead of as ReLU, which is improved from
the Swish nonlinear activation function. Swish is dened in
Equation (16).
Swish = lim
M
M=mn
i=1
xixi1
ðÞ
2
:ð16Þ
The bottleneck is a bottleneck structure, which can be
used in the network to reduce the number of parameters
and computation of the network with no loss of accuracy.
The structure is shown in Table 1.
In the traditional method, the Euclidean distance is usu-
ally used to determine the identity similarity has a great lim-
itation. The characters classied by the softmax function will
have two dierent identities that are closer than the same
identity, thus failing to achieve the classication eect, and
adding intraclass constraints through the cross-entropy loss
5Advances in Mathematical Physics
function can make the network have a better classication
eect, and the improved formula is shown in Equation (17).
Y=ln ψWx
ðÞ β+1
ðÞ
β1
ðÞ
βlim
nn
i=1ψWx
ðÞ
+xi
:ð17Þ
By normalizing the weights and features, the prediction
of the network depends only on the learning of the weights
and feature angles ψ. All features are distributed on a hyper-
sphere of radius s. A hyperparameter is added to the cross-
entropy loss function for the edge penalty of the weights
and feature angles ψ. This is done to increase the edge angles
between categories and at the same time make the categories
more compact internally, resulting in a better performance
for facial recognition, as in Equation (18).
Y=ln ψsin xi+T
ðÞ
lim
nn
i=1ψWx
ðÞ
+xi+ lim
nn
i=1ψsin xi+T
ðÞ
:ð18Þ
If two task instance sequences have the same task instance
sequence, then as far as the coherence function is concerned,
one task instance sequence with the shortest release interval
will dominate the other task instance sequences. Therefore,
we can divide the release time area of the task instance so that
each node is xed in each release interval.
3.3. Design of High-Capacity Real-Time Face Retrieval
System for Hospital Consultation Places. The design of the
hospital visit place high-capacity real-time face retrieval sys-
tem architecture is oriented to all users, the overall goal of
the system as the basis of the design, standing in the global
overview perspective of the systems overall perspective for
the construction of the hospital visit place high-capacity
real-time face retrieval of the business aspects of the needs
and technical aspects of the needs from multiple manage-
ment and constraints, from a comprehensive perspective
requires the construction of a unied platform, the global
model with unied specications. The system is designed
for a ve-layer structure, as shown in Figure 2.
The planning analysis is mainly carried out in two
aspects: on the one hand, it is data-oriented and analyzes
the current situation of data application for each department
and each data source; the data quality analysis is mainly
focused on the usability of data and the usage of data. Based
on data application status and data quality analysis, the sys-
tems shared data model is constructed by combining appli-
cation analysis. From another aspect, planning analysis is
oriented to applications and processes, and planning analy-
sis is conducted for applications, processes, and data
exchange and sharing involving cross-department and
cross-system, and on top of that, an application process
model is constructed. Based on these two aspects, a feasible
planning and analysis report is formed and the outline
design and detailed solution design are written and loaded
into the integration and development and implementation
phase of the project.
The self-built face database contains the face data of 5
people in the lab, in which each persons face data contains
10 images with dierent angles and each face image has a
pixel size of 112 92. The self-built face database uses 9 face
images of each person as training and 1 face image as testing.
In this paper, the validation of the PCA algorithm dimen-
sionality reduction is implemented by Matlab. Firstly, the
data from the self-built face database is read, and the sample
matrix is generated; secondly, the dimensionality reduction
is implemented by the PCA algorithm, and nally, the prin-
cipal component faces are displayed. The main parameters
include the number of people read in, the number of samples
nFacePerPerson read in for each person, and the container
FaceContainer to store the sample matrix. In the process of
implementing the PCA algorithm, the diculty is to calcu-
late the sample covariance matrix of eigenvalues and eigen-
vectors. The custom function named CalculationPCA
calculates the eigenvectors and eigenvalues of the matrix by
calling the library function. The input parameters are the
stored sample matrix FaceContainer and the dimension K
to be reduced to; the output parameters are the sample
eigenvector matrix and the principal component vector after
reducing to Kdimensions; nally, the reduced dimensional
principal component face is displayed. The output of the
reduced dimensional data is saved. Since this paper adopts
the functions of OpenCV to realize the detection and recog-
nition function of the human face, the XML le is completed
by OpenCV to ensure uniformity.
When a noisy image is acquired, the next step is to process
the image to achieve the desired eect. Second, to increase the
detectability of the object to be detected in the image and to
improve the image quality, grayscale processing, ltering,
noise reduction, and contrast enhancement operations are
required for the acquired image. The ARM architecture has
increased the clock speed by simplifying the instruction set
and also uses pipelining to increase the speed of the processor
instruction stream, but the nature of ARM for image process-
ing is serial, especially for embedded low-power products, due
to power constraints, the clock speed needs to be reduced,
which makes it dicult to ensure the real-time requirements.
This is where the signicant acceleration of many image pro-
cessing algorithms can be achieved. In this paper, we take
advantage of the high-speed parallelism of FPGAs to perform
grayscale processing, median ltering, and histogram equali-
zation on the acquired images through the PL part of Zynq
and package the module as an IP core.
4. Results and Analysis
4.1. Scheduling Analysis of High-Capacity Real-Time Face
Retrieval Recognition Algorithm for Hospital Consultation
Table 1: Bottleneck structure.
Serial number Input Operate Output
1h×w×dConv 1 × 1 h×w×d
2h×w× Ltd Conv 2 × 2 th × w× Ltd
3h×tdConv 3 × 3 th × tw × d
4th × w×dConv 4 × 4 th × tw × Ltd
6 Advances in Mathematical Physics
Places. As shown in Figure 3, the scheduling success rate of
both the grouped moderate algorithm and the improved
algorithm is higher than the saving algorithm in the same
feasibility check window. When the Wnd value is small,
the scheduling success rate of the improved algorithm is
not as good as that of the grouped moderate algorithm, but
as the Wnd value increases, the success rate of the improved
algorithm increases and exceeds that of the grouped
Infrastructure layer
Wechat users
System
administrator Doctor
Hospital decision
makers
Patients Department
Face retrieval
management
Hospital information
management
Face recognition
management
Number source
management
Doctor management Interactive data
management
Face data Permission dataBasic data
Multi-task model neural
network cloud
Task 1
Task 2
Training data
Training data
Scheduling model
User layer
Application layer
Network layer
Data layer
Figure 2: Overall system architecture diagram.
012345678910 11
Real-time acquisition size
300
280
260
240
220
200
Number of successful tasksAlgorithm success rate (%)
88
84
80
76
72
Saving algorithm success task
Group moderate algorithm success task
Improved algorithm success task
Saving algorithm success task
Improved algorithm success task
Moderate grouping algorithm success rate
Figure 3: Comparison of scheduling success rate of dierent scheduling algorithms.
7Advances in Mathematical Physics
moderate algorithm when Wnd is equal to 5. This is because
the improved algorithm needs to group the tasks in the win-
dow, and the grouping algorithm of the improved algorithm
is more sensitive to the size of the feasibility check window
because it takes into account the sharing of all resources of
the tasks compared to the grouping method in the grouping
moderate. When the window value is small, the dierence
between the improvement algorithm and the saving algo-
rithm is not too big, and the eect of grouping will slowly
appear when the window value becomes larger.
It can be seen from Figure 4 that among the processor
utilization rates of several algorithms, the average core utili-
zation rate of the improved algorithm is higher than that of
the other two, but the improved algorithm uctuates more
sharply than other algorithms. After calculation, the average
utilization rate of the saving algorithm, the appropriate
grouping algorithm, and the improved algorithm is
82.32%, 82.76%, and 86.12%, and the standard deviations
are 2.18, 1.64, and 3.25, respectively. One reason for this sit-
uation is that the improved algorithm will try to put tasks
with a high degree of correlation into the same core as much
as possible, but this will lead to a decrease in the degree of
load balancing.
4.2. Performance Analysis of High-Capacity Real-Time Face
Retrieval Recognition Algorithm for Hospital Consultation
Places. To validate the task scheduling-based high-capacity
real-time face retrieval and recognition algorithm for hospi-
tal visits, the neural network framework is used to imple-
ment the algorithm result analysis, and comparative
experiments are done in the training phase of the network
in terms of network structure, the number of parameters,
running speed, and recognition rate. The training samples
of the data set were divided by 11 : 2 to generate the training
set and test set for the experiments. The comparison results
of the experimental results are shown in Figure 5.
The time required by the task scheduling model-based
high-capacity real-time face retrieval recognition algorithm
and the traditional real-time face retrieval recognition algo-
rithm in the hospital visit site were tested separately. As
shown in Figure 6, it can be seen that the time required by
the traditional real-time face retrieval recognition algorithm
is about three times longer than that required by the task
scheduling model-based high-capacity real-time face
retrieval recognition algorithm for hospital visits, and it
can be seen that the speed of face detection is improved by
the task scheduling model, and the design objective is
achieved.
85
84
83
81
82
86
87
88
02468101214161820
Number of experiments
Processor utilization (%)
Saving algorithm
Moderate grouping algorithm
Improve algorithm
Figure 4: Processor utilization of the algorithm.
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Number of iterations
100
96
94
92
90
88
98
Retrieval and recognition accuracy (%)
Task scheduling face retrieval and recognition algorithm
Traditional face retrieval and recognition algorithm
Figure 5: Algorithm retrieval recognition eect.
8 Advances in Mathematical Physics
4.3. Analysis of High-Capacity Real-Time Face Retrieval
Recognition Algorithm System for Hospital Consultation
Places. The features extracted by the feature extraction net-
work are subjected to SVM classication to see the overall
recognition performance of the network, and we perform
two experiments to verify the network performance. We rst
extract the database of 1000 people and select the database
containing 5 to 100 a priori face images to extract face fea-
tures to train the SVM classier, leave 5 images for each per-
son as a test to test the classication performance, and test
the average value three times, respectively, and the results
are shown in Figure 7. As can be seen in Figure 7, our net-
work recognition performance increases in classication
accuracy in the early stage according to the increase of a
priori pictures, the highest reached 98.76%.
In the test set, we use CASIA Face dataset for testing, and
the network is based on task scheduling to do migration
learning. This experiment will be done by covering more
combinations randomly and uniformly before the training
starts, and the anchor points can be NIR or visible images.
The positive samples have the same id as the anchor points,
and the subsamples have dierent ids from the anchor
points, according to which about 160,000 sets of training
data can be obtained. Using these data, the trained visible
face recognition network is set up as a three-channel net-
work structure, and the network is trained by a triple loss
function, and the batch size is set to 256 and goes in for
training. The experimental results are shown in Figure 8.
From Figure 8, it can be seen that the scheme in this paper
has surpassed the existing mainstream algorithmic schemes.
When using multitask scheduling as the base network for
training, the recognition rate in the CASIA Face dataset
reaches a level of 99.2% and the verication rate reaches
98.9%. When using the traditional method as the base net-
work, the recognition rate reaches 98.8% and the verication
rate reaches 98.5% in the dataset. The task scheduling net-
work structure in this paper achieves similar performance
while using a fewer number of parameters and computations
than a standard convolutional network such as the tradi-
tional network. This results in a high-capacity real-time face
recognition network for hospital visits, which illustrates the
method of pretraining a perfect face recognition model by
task scheduling the face dataset and then using a triple loss
70
60
50
40
30
10
20
70
80
60
50
40
30
10
20
0
Retrieval recognition time (ms)Retrieval recognition time (ms)
123
4
5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Retrieval recognition times
Task scheduling face retrieval and recognition algorithm
Traditional face retrieval and recognition algorithm
Task scheduling face retrieval and recognition algorithm
Traditional face retrieval and recognition algorithm
Figure 6: Time required for face detection.
99.0
98.5
98.0
97.5
97.0
96.5
96.0
0 102030405060708090100
Number of face images
1.0
0.6
0.4
0.2
0.0
0.8
94.0
93.5
93.0
92.5
92.0
91.5
91.0
90.5
90.0
Task scheduling face retrieval and recognition algorithm (%)
Traditional face retrieval and recognition algorithm (%)
Figure 7: System recognition performance.
9Advances in Mathematical Physics
function with a suitable triplet for small-scale datasets for
migration learning, which has a good eect for learning
small-scale datasets.
5. Conclusion
This paper investigates the task scheduling-based high-
capacity real-time face retrieval recognition algorithm for
hospital visit places, which is not convenient for the equip-
ment due to a large number of scheduling parameters. The
algorithm of large-capacity real-time face retrieval and rec-
ognition for hospital consultation places based on task
scheduling is proposed. Firstly, the weaknesses and short-
comings of the existing commonly used algorithms in doing
face recognition are analyzed, and they are improved in
terms of the multitask scheduling network structure. A mul-
titask scheduling approach is proposed to replace the exist-
ing high-capacity real-time face retrieval inhospital
consultation places to recognize facial information with dif-
ferent degrees of importance in dierent locations. For the
triple loss function training method, the impact of poor
training data selection and construction method on neural
network training is analyzed, and a reasonable method of
training data triad selection and construction is proposed.
A reasonable training data triad selection and construction
method are proposed to solve the problem that the network
cannot converge quickly due to the random selection of data
to build the training set and then do migration learning with
the triple angular loss function to improve the training eect
of the network. Further, the acceleration factors between
RPM divisions with dierent granularity are investigated
and abounded upper bound on the acceleration factors of
two eective RPM divisions with subset relations which are
given to demonstrate the bounded pessimism of the pro-
posed analysis method. The core of the algorithm is the
use of hybrid schedulability analysis techniques with dier-
ent accuracy and complexity. Experimental results show that
the optimization method proposed in this chapter provides
better and near-upper bound performance while signi-
cantly reducing the running time of the algorithm compared
to the best available related research results. Although the
system designed in this paper improves the speed of face
detection and combines soft and hard codesign to accom-
plish the face recognition function, there are still some short-
comings. Firstly, it is dicult to improve the recognition rate
of face recognition by relying on the simple task scheduling
algorithm, and the SVM algorithm will be added in the next
work to improve the recognition rate of face recognition.
These shortcomings will be improved in future work and
practice.
Data Availability
The data used to support the ndings of this study are avail-
able from the corresponding author upon request.
Conflicts of Interest
The authors declare that they have no known competing
nancial interests or personal relationships that could have
appeared to inuence the work reported in this paper.
References
[1] P. Porambage, J. Okwuibe, M. Liyanage, M. Ylianttila, and
T. Taleb, Survey on multi-access edge computing for internet
of things realization,IEEE Communications Surveys & Tuto-
rials, vol. 20, no. 4, pp. 29612991, 2018.
[2] Z. H. Ali and H. A. Ali, Towards sustainable smart IoT appli-
cations architectural elements and design: opportunities, chal-
lenges, and open directions,The Journal of Supercomputing,
vol. 77, no. 6, pp. 56685725, 2021.
[3] E. Indrawan, N. Jalinus, and S. Syahril, Project based learning
in vocational technology education study of literature,Inter-
national Journal of Scientic & Technology Research, vol. 9,
no. 2, pp. 28212825, 2020.
[4] S. Abirami and P. Chitra, Energy-ecient edge based real-
time healthcare support system,Advances in Computers,
vol. 117, no. 1, pp. 339368, 2020.
[5] W. M. Alenazy and A. S. Alqahtani, Gravitational search
algorithm based optimized deep learning model with diverse
set of features for facial expression recognition,Journal of
Ambient Intelligence and Humanized Computing, vol. 12,
no. 2, pp. 16311646, 2021.
[6] X. Zhang, W. Huang, X. Lin, L. Jiang, Y. Wu, and C. Wu,
Complex image recognition algorithm based on immune ran-
dom forest model,Soft Computing, vol. 24, no. 16, pp. 12643
12657, 2020.
[7] S. Baek, System integration for predictive process adjustment
and cloud computing-based real-time condition monitoring of
vibration sensor signals in automated storage and retrieval sys-
tems,The International Journal of Advanced Manufacturing
Technology, vol. 113, no. 3-4, pp. 955966, 2021.
[8] S. K. Addagarla, G. K. Chakravarthi, and P. Anitha, Real time
multi-scale facial mask detection and classication using deep
transfer learning techniques,International Journal, vol. 9,
no. 4, pp. 44024408, 2020.
[9] I. Bartolini and M. Patella, A general framework for real-time
analysis of massive multimedia streams,Multimedia Systems,
vol. 24, no. 4, pp. 391406, 2018.
[10] F. M. Talaat, S. H. Ali, A. I. Saleh, and H. A. Ali, Eective load
balancing strategy (ELBS) for real-time fog computing
180
180
160
160
140
140
120
120
100
100
80
80
60
60
40
40
20
20
092
91
93
94
95 80 81
82
83
84
85
86
87
88
89
90
Traditional system algorithm
e system algorithm of this paper
Figure 8: Face recognition experiment results.
10 Advances in Mathematical Physics
environment using fuzzy and probabilistic neural networks,
Journal of Network and Systems Management, vol. 27, no. 4,
pp. 883929, 2019.
[11] Y. Xu, S. Wu, M. Wang, and Y. Zou, Design and implementa-
tion of distributed RSA algorithm based on Hadoop,Journal
of Ambient Intelligence and Humanized Computing, vol. 11,
no. 3, pp. 10471053, 2020.
[12] Q. Zhang, H. Sun, X. Wu, and H. Zhong, Edge video analytics
for public safety: a review,Proceedings of the IEEE, vol. 107,
no. 8, pp. 16751696, 2019.
[13] W. Choi and Y. J. Cha, SDDNet: real-time crack segmenta-
tion,IEEE Transactions on Industrial Electronics, vol. 67,
no. 9, pp. 80168025, 2019.
[14] L. J. Rubini and E. Perumal, Hybrid kernel support vector
machine classier and grey wolf optimization algorithm based
intelligent classication algorithm for chronic kidney disease,
Journal of Medical Imaging and Health Informatics, vol. 10,
no. 10, pp. 22972307, 2020.
[15] C. Deng, R. Guo, C. Liu, R. Y. Zhong, and X. Xu, Data clean-
sing for energy-saving: a case of cyber-physical machine tools
health monitoring system,International Journal of Produc-
tion Research, vol. 56, no. 1-2, pp. 10001015, 2018.
[16] S. N. Kumar, A. L. Fred, and P. S. Varghese, An overview of
segmentation algorithms for the analysis of anomalies on med-
ical images,Journal of Intelligent Systems, vol. 29, no. 1,
pp. 612625, 2018.
[17] G. Muhammad, M. F. Alhamid, and X. Long, Computing and
processing on the edge: smart pathology detection for con-
nected healthcare,IEEE Network, vol. 33, no. 6, pp. 4449,
2019.
[18] S. Chen, Z. Zhang, J. Yang et al., Fangcang shelter hospitals: a
novel concept for responding to public health emergencies,
The Lancet, vol. 395, no. 10232, pp. 13051314, 2020.
[19] S. Jain, R. Khera, Z. Lin, J. S. Ross, and H. M. Krumholz,
Availability of telemedicine services across hospitals in the
United States in 2018: a cross-sectional study,Annals of Inter-
nal Medicine, vol. 173, no. 6, pp. 503505, 2020.
[20] P. Yang, K. M. Treurniet, L. Zhang et al., Direct intra-arterial
thrombectomy in order to revascularize AIS patients with
large vessel occlusion eciently in Chinese tertiary hospitals:
a multicenter randomized clinical trial (DIRECT-MT)pro-
tocol,International Journal of Stroke, vol. 15, no. 6,
pp. 689698, 2020.
11Advances in Mathematical Physics
ResearchGate has not been able to resolve any citations for this publication.
Article
Full-text available
As automation and digitalization are being increasingly implemented in industrial applications, manufacturing systems comprising several functions are becoming more complex. Consequently, fault analysis (e.g., fault detection, diagnosis, and prediction) has attracted increased research attention. Investigations involving fault analysis are usually performed using real-time, online, or automated techniques for fault detection or alarming. Conversely, recovery of faulty states to their healthy forms is usually performed manually under offline conditions. However, the development of intelligent systems requires that appropriate feedback be provided automatically, to facilitate faulty-state recovery without the need for manual operator intervention and/or decision-making. To this end, this paper proposes a system integration technique for predictive process adjustment that determines appropriate recovery actions and performs them automatically by analyzing relevant sensor signals pertaining to the current situation of a manufacturing unit via cloud computing and machine learning. The proposed system corresponds to an automated predictive process adjustment module of an automated storage and retrieval system (ASRS). The said integrated module collects and analyzes the temperature and vibration signals of a product transporter using an internet-of-things-based programmable logic controller and cloud computing to identify the current states of the ASRS system. Upon detection of faulty states, the control program identifies corresponding process control variables and controls them to recover the system to its previous no-fault state. The proposed system will facilitate automatic prognostics and health management in complex manufacturing systems by providing automatic fault diagnosis and predictive recovery feedback.
Article
Full-text available
The convergence initiatives of cloud technology and the Internet of Things (IoT) have demonstrated a massive rise of the futuristic technologies which ensure sustainable production development of IoT applications such as intelligent transportation, smart cities, smart healthcare, and other applications. These real-time applications are employing the information and communication technology as an efficient utility to support a ubiquitous approach in the industry so that any apparatus can seamlessly interact with different cyber-physical systems and may also share its information effortlessly. Although the amalgamation of cloud and IoT offers robustness pervasive computing scenarios that can achieve the composite service with high quality and acceptable cost, it may face several challenges and issues not only in performance but also in architecture. The main aim of this study is to offer a stander framework to investigator and application designer that how sustainability can be achieved using different quality of service metrics. This is an overview of some crucial issues of IoT and the relation between it and other technologies based on various platforms. Focus is given in this study to three of the most important IoT applications that require highly successful sustainability procedures which are presented. A brief discussion and statistical categories of the research techniques published from 2000 to 2020, which highlight sustainable IoT applications, are also introduced. The contributions of this study are as follows: (i) summarizing the organization and methodology of the most important sustainable IoT applications, (ii) discussing the main challenges of the sustainable growth of IoT applications, (iii) highlighting future research directions in the field of IoT, and introducing guidelines for open business opportunities.
Article
Full-text available
In the era of deep learning, object detection plays an influential role for many industries. Detecting minute things are very much essential without human intervention especially at large scale industries. In this paper we have proposed multiple approaches for Multi-scale facial mask real time detection and classification for the hospital industry, crowd surveillance in the streets and malls are more useful in this COVID-19 Pandemic Situation. In our approach we have implemented two different detection models which are FMY3 using Yolov3 Algorithm and FMNMobile using NASNetMobile and Resnet_SSD300 Algorithms and used two different face mask dataset with 680 and 1400 images respectively. We have analyzed both the models by computing various probabilistic accuracy measures and achieved the 34% Mean Average Precision (mAP) and 91.7% Recall rate on FMY3 Model and achieved the 98% and 99% of accuracy and recall rate on FMNMobile Model. Finally we have shown results of various face mask detections from both the models.
Article
Full-text available
Facial expression recognition (FER) is an essential part of effective human–computer interaction and serves as a helpful medium for children and patients who have problems with communication. However, most of the previous studies focus on building a FER model based on supervised and unsupervised approaches. This paper is focused on a semi-supervised deep belief network (DBN) approach to predict the facial expressions from the CK+, Oulu CASIA, MMI, and JAFFE datasets. To achieve accurate classification of the facial expressions, a gravitational search algorithm (GSA) is applied to optimize some parameters in the DBN network. The Histogram oriented gradients (HOG) and 2D-Discrete Wavelet Transform (2D-DWT) are used for feature extraction from the lip, cheek, brow, eye, and furrow patches. The unwanted information present in the image is eliminated using a feature selection approach. The feature extraction is done by the Kernel-principal component analysis to obtain higher-order correlations between input variables and detect non-linear components. The HOG features extracted from the lip patch provides the best performance for accurate facial expression classification. Finally, a comparative analysis to compare the proposed model with different machine learning techniques based on the evaluation criteria. The results demonstrate that DBN-GSA based classifier is more accurate than the rest of the classifiers.
Article
Full-text available
The world of education, especially vocational technology education, is never finished to be discussed, ranging from skills-based to values-based education. One of the approaches in practical education is project-based learning. It is an educational approach that focuses on creative thinking, problem solving, and interaction between students and peers to create and use new knowledge. Through project-based learning, students will work in teams; discover skills in planning, organizing, negotiating, and making consensus on issues of assignment to be done; appoint who is responsible for each task; and decide how information will be collected and presented scientifically. Index Terms: project-based learning, vocational technology education, study of literature-
Article
Full-text available
With the continuous advancement of science and technology, the social network based on the Internet has gradually penetrated into people’s daily lives. The image data shared on social media using platforms such as mobile phones has exploded, and hundreds of millions of dollars are generated every day with picture information. In the past, the traditional ways of expressing textual information that people are familiar with have begun to be gradually replaced by image information that is not subject to regional culture such as language and script. In modern warfare, with the continuous development and equipment of highly sophisticated weapons and equipment, the amount of information that the entire combat system needs to process will also increase. In particular, the air defense system needs to quickly and accurately identify the aircraft targets that are coming. It mainly uses computer to extract the feature information of the acquired image and converts the content in the image into a feature expression that can be processed by the computer. After the appropriate classification algorithm, the image is the target object is classified by category. In this paper, we propose a complex image recognition algorithm based on immune random forest model. The experimental results show that the proposed algorithm has high recognition efficiency and higher robustness.
Article
In the present day, distributed algorithms become more popular due to their diversity in several applications. The prediction and reorganization of medical data required more practice and information. We propose a novel approach feature selection based on efficient chronic kidney disease (CKD) prediction and classification. Primarily, the pre-processing pace will be implemented over the input data. Then, the grey wolf optimization (GWO) algorithm gets executed to choose the optimal features from the pre-processed data. Next, the projected technique exploits the Hybrid Kernel Support Vector Machine (HKSVM) as a classification model to identify the presence of CKD or not. The simulation takes place in MATLAB. The validation of the presented model takes place using a benchmark CKD dataset as of machine learning repository such as UCI under the presence of several measures. New outcome specified that the planned categorization arrangement has surpassed by containing enhanced 97.26% accuracy for kidney chronic dataset when contrasted with existing SVM technique only accomplished 94.77% and fuzzy min–max GSO neural network (FMMGNN) classifier accomplished 93.78%.
Article
Fangcang shelter hospitals are a novel public health concept. They were implemented for the first time in China in February, 2020, to tackle the coronavirus disease 2019 (COVID-19) outbreak. The Fangcang shelter hospitals in China were large-scale, temporary hospitals, rapidly built by converting existing public venues, such as stadiums and exhibition centres, into health-care facilities. They served to isolate patients with mild to moderate COVID-19 from their families and communities, while providing medical care, disease monitoring, food, shelter, and social activities. We document the development of Fangcang shelter hospitals during the COVID-19 outbreak in China and explain their three key characteristics (rapid construction, massive scale, and low cost) and five essential functions (isolation, triage, basic medical care, frequent monitoring and rapid referral, and essential living and social engagement). Fangcang shelter hospitals could be powerful components of national responses to the COVID-19 pandemic, as well as future epidemics and public health emergencies.