Abstract
For a better detection in Network information security monitoring system, the author proposes a method based on adaptive depth detection. A deep belief network (DBN) was designed and implemented, and the intrusion detection system model was combined with a support vector machine (SVM). The data set adopts the NSL-KDD network communication data set, and this data set is authoritative in the security field. Redundant cleaning, data type conversion, normalization, and other processing operations are performed on the data set. Using the data conversion method based on the probability mass function probability mass function coding, a standard data set with low redundancy and low dimensionality can be obtained. Research indicates that when the batch size reaches 64, the accuracy of the test set reaches its maximum value. As the batch size increases, the accuracy first increases and then decreases. When the batch size continues to increase, the model will inevitably fall into the local optimal state, resulting in the degradation of the detection performance of the system. In terms of the false alarm rate, the DBN-SVM model is also the highest; however, it is only 10.73%. Under the premise of increasing the detection rate, the false alarm rate is improved; for the overall detection performance of the model, it is within an acceptable range. In terms of accuracy, the DBN-SVM model also scored the highest. The accuracy rate is the ratio of normal and correct classification for intrusion detection. It can explain the detection ability of the model. In summary, the overall detection ability of the DBN-SVM model is the best. The good classification ability to use SVM is proved, and the classification of low-dimensional features is expected to increase the detection rate of the system.
1 Introduction
Previously, the security protection of industrial control systems is generally deployed in a layer-by-layer protection manner, that is, specific protective measures are set up at any position on the attack path (such as the entrance of the enterprise network, the entrance of the monitoring layer, and the entrance of the field layer), to block the attack, the protection of the field layer is realized at the highest security level [1]. Traditional firewalls are the experts who set up filtering rules based on experience. In order to realize the safety protection of the system, we need to define various bus protocols and protection rules in advance. It can be done by establishing a special agency service program for each application service, but only according to the predefined filtering rules it can monitor and analyze datagrams at the network layer and only provide a single-level static network security protection [2]. As shown in Figure 1, the filtering rules are set by network security experts based on experience and existing knowledge, and it is difficult to adapt to the ever-changing nature of the network. Although it can monitor and control the application layer communication flow, its speed is slower and consumes too many CPU resources [3]. When the flow volume at the junction is too large, the processing speed is too slow, the safety protection equipment is turned into a ring bottleneck, and the filtering and analysis conduction speed is too slow for slice data. There is a large time delay in data exchange, which is not suitable for the real-time requirements of the industrial control system. It is extremely difficult to control the network and is even a network paralysis in the current situation. A loss is created that cannot be remedied.
Network information security monitoring system.
2 Literature review
Lu proposed deep learning, and many fields have begun to introduce deep learning theories. In the field of visual recognition, audio recognition, as well as the field of natural language, deep learning has been widely used [4]. Rasyid et al. used deep neural networks to reduce the size of feature values and then classified different traffic. First, they used the deep belief network in the intrusion detection system [5]. Hasanikhah et al. introduced a support vector machine (SVM) in the intrusion detection system to deal with the system with a very large network data and which was unable to respond correctly in time thereby missing the best reported intrusion event. It results in problems such as system attacks or losses, however using SVM we can improve the detection efficiency of the system, filter the information that has a greater impact on the system, eliminate invalid information and greatly improve the real-time performance of the system [6]. Prather and Frazier introduced neural network technology with excellent performance in the intrusion detection system, to solve IDS’s low detection rate when detecting unknown attacks at that time, the problem of system losses caused by location attacks that cannot be effectively prevented, using neural network technology, and to predict and classify unknown cyber attacks, and at the same time, IDS based on the BP neural network [7] is proposed. Seo and Lee introduced the Elman neural network in IDS, taking advantage of its Lenovo function, and proposed IDS based on the Elman neural network to cope with the increase in the network bandwidth and the continuous decrease in the detection rate caused by the increase in network scale, and after experimental verification, the performance of intrusion detection, especially the detection rate, has indeed been improved [8]. Ahmed et al. proved that the discriminative restricted Boltzmann machine can be applied in different network environments; moreover, the learning ability of the system to incomplete data is realized [9]. Yu et al. introduced recurrent neural networks in intrusion detection, and it is found through experiments that the detection performance of the system is greatly improved compared with traditional algorithms. The integration between various fields is getting closer and closer, intrusion detection systems are also continuously incorporating fresh deep neural networks, the effectiveness of this approach has been extensively proven by experiments, it is proved that the intrusion detection system can be combined with the neural network, and the development prospects of deep learning in intrusion detection systems have considerable development potential [10]. In the study by Hu et al., the remote sensing image classification based on deep learning uses deep learning to represent pixel data in a deep level, and a more abstract feature representation of the data can be extracted, which can improve the accuracy of classification [11]. Agrawal and Tapaswi proposed target detection algorithm based on deep learning, and it also includes the steps of candidate frame extraction, feature extraction, and classification detection, and compared with traditional methods, it is more powerful and intelligent in feature extraction. However, when the deep network is used for detection, generally, solidified classification and regression methods are used, but in actual tasks, the distribution of the targets to be detected is uneven, and during the training process, too much attention is paid to the definition of the positive sample candidate frame, but the importance of the negative sample is ignored [12]. Ee et al. proposed the method based on deep learning, which has attracted great attention. Considering spatial information and spectral information, the deep autoencoder is used in the HSI classification task. The stacked autoencoder network is used to learn features from the data after dimensionality reduction [13].
Based on the current research, the author proposes a method based on adaptive depth detection. An intrusion detection system model has been designed and implemented by combining the deep belief network and the support vector machine. The data set adopts the NSL-KDD network communication data set, which is authoritative in the security field. Redundant cleaning, data type conversion, normalization, and other processing operations on the data set have been performed. By using the data conversion method based on the probability mass function coding, a standard data set with low redundancy and low dimensionality can be obtained.
3 Target detection model and data analysis based on deep adaptive area suggestion network
The model proposed by the author is called DAP Net, which is a deep adaptive area proposal network. It consists of four main parts: VGG-16 backbone network, category prior network (CPN), fine-region proposal network (F-RPN), and detection network (accuracy Region CNN [A-RCNN]). The DAP Net method is a novel network designed, and it can be based on the distribution of various objects in the image, automatically adjusting the number of candidate frames [14]. First, the VGG-16 network is used to generate high-level convolution features for each image. Then, the feature is sent to three independent convolutional networks: CPN, F-RPN, and A-RCNN, and CPN is used to predict the prior information of the image category, including whether to include different types of targets, and predict the approximate number of targets.
3.1 Multilevel regression
CPN design proposes a new solution to the regression of multilevel category numbers. Due to the uneven distribution of objects in complex scenes and remote sensing images, the number of objects can appear at any value. For the advanced features of each image, each regression number of each category is predicted at the same time, and the regression of each category will not affect each other. In supervised batch learning, one of the key premises of the learning category prior networks is the definition of training samples [15]. For each image, nine levels (1, 2, 4, 8, 16, 24, 32, 48, and 64) are predicted as the reference base for regression. First, the actual number of objects in each category in each image is calculated, and the reference base of each level of each category is fixed. Then, the difference between the different levels of each category is calculated according to the following formula:
where c represents the category, and e represents the regression level within the E range.
3.2 Loss function
For the training of the CPN network, a new regression and classification loss are set. Similar to the multitask loss in the fast convolutional neural network, the loss of the CPN network is defined as follows:
where e is the number of levels of regression,
The classification and regression loss terms are normalized by
3.3 Training and testing
CPN is a small network [20,21,22,23,24,25,26,27,28] based on the advanced features of convolutional networks. As mentioned earlier, VGG-16 backbone network parameters [28,29,30,31,32] are initialized using the pretrained weights on the ImageNet1000 competition dataset [33,34,35]. For other convolutional layers in CPN, use the Xavier method to initialize the parameters. CPN can be trained end-to-end through backpropagation and stochastic gradient descent [36], and this includes forward iteration with labeled data as input [37,38,39]. According to the aforementioned multitask loss function, it can be optimized for each category prediction [40,41]. After training CPN, training images to the network are send, and the offset of the number of each image category relative to the base number of the level is generated. However, the purpose of CPN is to obtain the number of each category [42,43,44,45,46,47], not the offset of different levels of all categories. Therefore, during the CPN testing process, the scores of different levels of regression offset for each category are predicted. According to the output score of CPN, a threshold of 0.7 is fixed to filter the regression results of low confidence categories and low confidence levels. Only the regression results are kept at different levels for categories whose classification is higher than the 0.7 threshold, after calculating all the regression results of each category, and the average is the final predicted quantity result of the category [48].
4 Experimental results and analysis
4.1 Experimental environment and evaluation standards
Simulation experiment environment
Operating system: Windows: 10/64 bit, CPU: Inteli7-4700, RAM: 16.00GB, Development software: TensorFlow1.12.0 auxiliary tool: Excel
Evaluation standard of testing ability
The essence of intrusion detection system is to classify feature values. When dealing with classification problems, the confusion matrix is often used as the evaluation standard. Commonly used performance indicators of intrusion detection systems include accuracy rate, detection rate, and false alarm rate.
Confusion matrix
To verify the detection performance of the DBN-SVM model, that is, to identify the abnormal network data, the confusion matrix of the two categories is set as shown in Table 1. There are four situations, namely, to predict the normal normal record TN, predict the abnormal normal record FP, predict the normal abnormal record FN, and predict the abnormal abnormal record TP [49].
Confusion matrix
| Actual value/predictive value | Normal | Anomaly |
|---|---|---|
| Normal | TN | FP |
| Anomaly | FN | TP |
A high-performance intrusion detection system requires the detection rate and accuracy to be as high as possible. The false alarm rate is as low as possible
The accuracy rate (accuracy) calculation method is calculated as follows:
The calculation method of detection rate (DR) is calculated as follows:
The false alarm rate (FPR) is calculated as follows:
4.2 Experimental results
Adjusting parameters is a very critical step in deep learning. The quality of a model depends on its parameters. The DBN-SVM model of this experiment adjusts the corresponding parameters in terms of training rounds, number of neurons, batch size, and so on, and it is expected that good classification accuracy and speed can be obtained [50]. The impact of training rounds on the results is very important, and there are few training rounds, and it is difficult to learn the relationship between features; as there are too many training rounds, the model is prone to overfitting. At the same time, the training time will be longer. An early-stopping strategy is adopted to give the model a threshold, and when the accuracy difference between the two training results is less than this threshold, training is stopped. Hence, this can reduce training time and also prevent overfitting. As shown in Figures 2 and 3, when the training sequence reaches 20 times, the accuracy rate and the loss rate tend to be stable. To reduce the training time of the model, the training round is set to 20 times [51,52,53,54,55].
Changes in accuracy rate with rounds.
Changes in loss rate with rounds.
The learning ability of a neural network depends on the number of neurons. The greater the number, the stronger the learning ability, and it will also bring more training time and overfitting problems [56]. As shown in Figures 4 and 5, when the number of neurons is 20, 40, 60, 80, 100, and 120, the model’s accuracy rate and loss rate change the graph. As the number of neurons increases, the accuracy of the verification set shows a trend of increasing first and then decreasing, and the maximum value is taken where the independent variable is 100. Then, the model is overfitted, and increasing the number of neurons will only increase the training time and no longer increase the accuracy of the validation set. Therefore, the number of neuron nodes in the model is set to 100.
Changes in accuracy rate.
Changes in loss rate.
Choosing an appropriate batch size can speed up model optimization and prevent the missing of the optimal solution in the process of gradient descent.
The experimental results of varying batch size is depicted in Figures 6–7.
Changes in accuracy.
Changes in loss rate.
As shown in Figure 6, the accuracy of the test set reaches its maximum when the batch size reaches 64. It shows that in the process of the increasing batch size, the accuracy first increases and then decreases. When the batch size continues to increase, the model will inevitably fall into a local optimum, leading to a decrease in the detection performance of the system. As shown in Figure 7, the loss rate of the system increases with the increase of batch processing, and too large batch size will increase the loss rate of the system and reduce the classification ability of the system. To reduce training time, overfitting has to be prevented, the detection ability of the model has to be improved, and the batch size of the experiment is set to 64 [57,58,59,60].
4.3 Intelligent monitoring and analysis solution
Through diversified acquisition, big data storage and retrieval, and AI intelligent analysis, the system presents six main functions, including situation awareness, traffic perspective, backtracking analysis, performance monitoring, security detection, and asset management. It also combines all kinds of equipment and topology management, thematic analysis, machine learning modeling analysis, and attack countermeasures. Nondestructive detection, abnormal file identification and restoration, active measurement, industrial control network monitoring, and other functions create an integrated network comprehensive, intelligent monitoring, and analysis solutions.
Situational awareness. The virtual reality visualization displays the network trend, traffic composition and distribution, and performance and alarm information. The virtual reality visualization displays the security event tracks from the perspectives of the attacker and the victim and displays the attack type, severity level, serious geographical region, attack, and the most serious information of the attacked department, organization, and individual.
Traffic perspective. Monitors traffic in real time, displays the time granularity up to 1 s, and displays the applications, users, and external addresses of the current traffic in real time. The system supports virtual link as the basic object. Data among network traffic applications, application groups, users, user groups, countries, cities, external addresses, virtual links, and academic platforms should support correlation and multidimensional drill-down analysis.
Performance monitoring. The system can comprehensively evaluate and analyze mainstream service access services such as Web page, DNS, mail, database, and voice, including key performance indicators such as connection establishment time, response time, and connection success rate. Supports comprehensive service comparison and analysis by region and server to evaluate service status and locate performance faults and bottlenecks.
Retrospective analysis. Session record data contains more than 25 key indicators, such as recording time, port, MAC address, duration, total bytes, total packets, and rate. Security events can automatically trigger the preservation of original data packets, save attack, intrusion, and virus data packets, and decode, analyze, and download data packets online. You can use baseline alarms to trigger the correlation of service performance, network performance, network session, and security event data. You can view associated session records in service performance, associated network performance indicators in session records, and associated original data in security events
Package to view associated session records in security events.
Detect all kinds of attacks in the network. It includes 35 categories such as port scanning attack, code overflow attack, Trojan virus attack, worm attack, SQL injection attack, DOS attack, ActiveX control attack, mail attack, bad IP detection, personal Internet access detection, remote procedure call attack, system vulnerability attack, and other attack detection behaviors. User-defined attacks: Users define security events by setting the source/destination IP address, port, location of feature characters in the payload, packet flag, and packet frequency. Multilevel association analysis, multidimensional drill-down analysis according to any dimension (such as alarm type, alarm information, target country, target city, protocol, source address, destination address, suspect user, suspect user group, severity level), In addition, we can check the alarm trend, event number, flow pattern, and other details.
Asset management. Monitors the overall asset/virtual asset overview within a specific range and provides summary statistical classification to display the overall status and security alarms of selected assets. Supports type filtering, core setting, and legitimate device configuration. Supports periodic asset scanning and manual immediate asset scanning. Supports asset life-cycle management and asset safety status score, indicating high-risk assets, detection and discovery of lost assets, and provides security hardening suggestions.
5 Conclusion
The original data set NSL-KDD adopted a common processing method and eliminated redundancy, type conversion, and normalization. After completing the data preprocessing step, the experimental standard data set is obtained. The DBN-SVM model is trained, the processed experimental training set is inputted into the model, learning is forwarded to obtain the local optimal parameters, and then the top-level BP network is used to verify the results; if the threshold is exceeded, the error is reversed, and the gradient descent method is used to train the global optimal parameters layer by layer, until the model parameters with suitable results are trained. The detection performance of other models is compared. It shows that this model performs well in intrusion detection and has the highest detection rate and accuracy rate; although with a certain false alarm rate, but for the overall detection capability of the system, it is still the model with the best comprehensive performance. The DBN-SVM model to the intrusion detection system, has certain application value and can improve the detection rate and accuracy of the intrusion detection system.
-
Conflict of interest: The authors declare no conflict of interest.
-
Data availability statement: All data that support the findings of this study are included within the article.
References
[1] Santoso F, Redmond SJ. Indoor location-aware medical systems for smart homecare and telehealth monitoring: state-of-the-art. Physiolog Meas. 2015;36:R53–87.10.1088/0967-3334/36/10/R53Search in Google Scholar PubMed
[2] Gorbil G, Abdelrahman OH, Pavloski M, Gelenbe E. Modeling and analysis of rrc-based signalling storms in 3g networks. IEEE Trans Emerg Top Comput. 2015;4:113–27.10.1109/TETC.2015.2389662Search in Google Scholar
[3] Ha J. Preprocessor implementation of open ids snort for smart manufacturing industry network. J Korea Inst Inf Sec Cryptol. 2016;26:1313–22.Search in Google Scholar
[4] Lu WZ. Research and implementation of computer network user behavior forensics sys-tem based on system a log. J Inf Sec Res. 2018;9:23–40.Search in Google Scholar
[5] Rasyid M, Sukaridhoto S, Sudarsono A, Kaffah AN. Design and implementation of hypothermia symptoms early detection with smart jacket based on wireless body area network. IEEE Access. 2020;8:155260–74.10.1109/ACCESS.2020.3018793Search in Google Scholar
[6] Hasanikhah N, Amin-Nejad S, Darvish G, Moniri MR. Efficient implementation of space-time adaptive processing for adaptive weights calculation based on floating point fpgas. J Supercomputing. 2018;74:3193–210.10.1007/s11227-018-2369-7Search in Google Scholar
[7] Prather WE, Frazier WG. Method for effective implementation of an acoustic based hostile fire detection system on moving vehicles. J Acoustical Soc Am. 2016;139:2109.10.1121/1.4950273Search in Google Scholar
[8] Seo CK, Lee SG. A study on the implementation of geographic information system for an intelligent power distribution network with location informations of power line communication-based automatic meter reading system. Trans Korean Inst Electr Eng. 2015;64:365–9.10.5370/KIEE.2015.64.3.365Search in Google Scholar
[9] Ahmed H, Elkhatib M, Adly I, Ragai H. Design and implementation of fuzzy event-detection algorithm for border monitoring on fpga. Int J Fuzzy Syst. 2015;18:1–11.10.1007/s40815-015-0079-3Search in Google Scholar
[10] Yu F, Wan Q, Jin J, Li L, Tang Q. Design and fpga implementation of a pseudorandom number generator based on a four-wing memristive hyperchaotic system and bernoulli map. IEEE Access. 2019;7:181884–98.10.1109/ACCESS.2019.2956573Search in Google Scholar
[11] Hu X, Zhao J, Seet BC, Leung V, Chu T, Chan H. S-aframe: agent-based multilayer framework with context-aware semantic service for vehicular social networks. IEEE Trans Emerg Top Comput. 2015;3:44–63.10.1109/TETC.2015.2395959Search in Google Scholar
[12] Agrawal N, Tapaswi S. The performance analysis of honeypot based intrusion detection system for wireless network. Int J Wirel Inf Netw. 2017;24:14–26.10.1007/s10776-016-0330-3Search in Google Scholar
[13] Ee SHL, Kim YH, Ee JKL, Ee DGL. Hybrid app security protocol for high speed mobile communication. J Supercomput. 2016;72:1715–39.10.1007/s11227-014-1318-3Search in Google Scholar
[14] Jin W, Hyeun KD. Design and implementation of e-health system based on semantic sensor network using ietf yang. Sensors. 2018;18:629.10.3390/s18020629Search in Google Scholar PubMed PubMed Central
[15] Jia Y, Wang M, Wang Y. Network intrusion detection algorithm based on deep neural network. IET Inf Security. 2019;13:48–53.10.1049/iet-ifs.2018.5258Search in Google Scholar
[16] Feng T, Yang L, Zhao X, Zhang H, Qiang J. Online identification of lithium-ion battery parameters based on an improved equivalent-circuit model and its implementation on battery state-of-power prediction. J Power Sources. 2015;281:192–203.10.1016/j.jpowsour.2015.01.154Search in Google Scholar
[17] Zhang J, Li F, Zhang H, Li R, Li Y. Intrusion detection system using deep learning for in-vehicle security. Ad Hoc Netw. 2019;95:101974.1–101974.12.10.1016/j.adhoc.2019.101974Search in Google Scholar
[18] Feng F, Liu X, Yong B, Zho R, Zhou Q. Anomaly detection in ad-hoc networks based on deep learning model: a plug and play device. Ad Hoc Netw. 2019;84:82–9.10.1016/j.adhoc.2018.09.014Search in Google Scholar
[19] Dong RH, Li XY, Zhang QY, Yuan H. Network intrusion detection model based on multivariate correlation analysis – long short-time memory network. IET Inf Sec. 2020;14:166–74.10.1049/iet-ifs.2019.0294Search in Google Scholar
[20] Amhoud E-M, Chafii M, Nimr A, Fettweis G. OFDM with index modulation in orbital angular momentum multiplexed free space optical links, 2021 IEEE 93rd Vehicular Technology Conference (VTC2021-Spring); 2021. p. 1–5.10.1109/VTC2021-Spring51267.2021.9448928Search in Google Scholar
[21] Gill HS, Singh T, Kaur B, Gaba GS, Masud M, Baz M. A metaheuristic approach to secure multimedia big data for IoT-based smart city applications. Wirel Commun Mob Comput. 2021;2021:1–10.10.1155/2021/7147940Search in Google Scholar
[22] Kumar A, Sehgal VK, Dhiman G, Vimal S, Sharma A, Park S. Mobile networks-on-chip mapping algorithms for optimization of latency and energy consumption. Mob Netw Appl. 2021;1–15.10.1007/s11036-021-01827-0Search in Google Scholar
[23] Boguszewicz C, Boguszewicz M, Iqbal Z, Khan S, Gaba GS, Suresh A, et al. The fourth industrial revolution-cyberspace mental wellbeing: Harnessing science & technology for humanity.Search in Google Scholar
[24] Amhoud E, Othman GR, Jaouën Y. Concatenation of space-time coding and fec for few-mode fiber systems. IEEE Photonics Technol Lett. 2017;29:603–6.10.1109/LPT.2017.2675919Search in Google Scholar
[25] Amhoud EM, Othman GRB, Bigot L, Song M, Andresen ER, Labroille G, et al. Experimental demonstration of space-time coding for MDL mitigation in few-mode fiber transmission systems, 2017 European Conference on Optical Communication (ECOC); 2017. p. 1–3.10.1109/ECOC.2017.8345841Search in Google Scholar
[26] Gaba GS, Privacy-preserving authentication and key exchange mechanisms in internet of things applications (Doctoral Dissertation, Lovely Professional University Punjab); 2021.Search in Google Scholar
[27] Choudhary K, Gaba GS. Artificial intelligence and machine learning aided blockchain systems to address security vulnerabilities and threats in the industrial Internet of things. Intell Wirel Commun. 2021;329:329–61.10.1049/PBTE094E_ch13Search in Google Scholar
[28] Zerhouni K, Amhoud EM, Chafii M. Filtered multicarrier waveforms classification: a deep learning-based approach. IEEE Access. 2021;9:69426–38.10.1109/ACCESS.2021.3078252Search in Google Scholar
[29] Gaba GS, Kumar G, Monga H, Kim TH, Liyanage M, Kumar P. Robust and lightweight key exchange (LKE) protocol for industry 4.0. IEEE Access. 2021;8:132808–24.10.1109/ACCESS.2020.3010302Search in Google Scholar
[30] Sharma A, Kumar N. Third eye: an intelligent and secure route planning scheme for critical services provisions in internet of vehicles environment. IEEE Syst J. 2021.10.1109/JSYST.2021.3052072Search in Google Scholar
[31] Kumar P, Gaba GS. Biometric‐based robust access control model for industrial internet of things applications. IoT Security: Adv Authentication. 2020;133–42.10.1002/9781119527978.ch7Search in Google Scholar
[32] Hedabou M. Cryptography for addressing cloud computing security, privacy and trust issues. Book on computer and cyber security: principles, algorithm, applications and perspective. USA: CRC Press, Francis and Taylor Publisher; 2018.10.1201/9780429424878-11Search in Google Scholar
[33] Iggaramen Z, Hedabou M. FADETPM: Novel approach of file assured deletion based on trusted platform module. Lecture Notes Netw Syst. 2017;49:49–59.10.1007/978-3-319-97719-5_4Search in Google Scholar
[34] Azougaghe M, Hedabou M, Belkasmi M, An electronic voting system based on homomorphic encryption and prime numbers. In International Conference On Information Assurance and Security, 2015.10.1109/ISIAS.2015.7492759Search in Google Scholar
[35] Bentajer A, Hedabou M, Abouelmehdi K, Igarramen Z, Fezazi SE. AN IBE-based design for assured deletion in cloud storage. J Cryptolog. 2019;43:559–64.10.1080/01611194.2018.1549123Search in Google Scholar
[36] Gaba GS, Kumar G, Monga H, Kim TH, Kumar P. Robust and lightweight mutual authentication scheme in distributed smart environments. IEEE Access. 2020;8:69722–33.10.1109/ACCESS.2020.2986480Search in Google Scholar
[37] Hedabou M, Beneteau L, Pinel P. Some ways to secure elliptic curves cryptosystems. J Adv Cliford Algebras. 2008;18:677–88.10.1007/s00006-008-0093-8Search in Google Scholar
[38] Gaba GS, Kumar G, Monga H, Kim TH, Kumar P. Secure device-to-device communications for 5g enabled internet of things applications. Computer Commun. 2020;169:114–28.10.1016/j.comcom.2021.01.010Search in Google Scholar
[39] Sharma A, Podoplelova E, Shapovalov G, Tselykh A, Tselykh A. Sustainable smart cities: convergence of artificial intelligence and blockchain. Sustainability. 2021;13:13076–4967.10.3390/su132313076Search in Google Scholar
[40] Bentajer M, Hedabou M, Abouelmehdi K, Fezazi SE. CS-IBE: A data confidentiality system in public cloud storage system. Proc Computer Sci. 2018;141:559–64.10.1016/j.procs.2018.10.126Search in Google Scholar
[41] Azougaghe M, Hedabou M, Oualhaj O, Belkasmi M, Kobbane A, Many-to -One matching game towards secure virtual machine migrating in cloud computing, International Conference on Advanced Communication System and Information Security, 2016.10.1109/ACOSIS.2016.7843922Search in Google Scholar
[42] Masud M, Gaba GS, Choudhary K, Hossain MS, Alhamid MF, Muhammad G. Lightweight and anonymity-preserving user authentication scheme for IoT-based healthcare. IEEE Internet Things J. 2021.10.1109/JIOT.2021.3080461Search in Google Scholar
[43] Sharma A, Singh PK, Sharma A, Kumar R. An efficient architecture for the accurate detection and monitoring of an event through the sky. Computer Commun. 2019;148:115–28.10.1016/j.comcom.2019.09.009Search in Google Scholar
[44] Masud M, Gaba GS, Choudhary K, Alroobaea R, Hossain MS. A robust and lightweight secure access scheme for cloud based E-healthcare services. Peer-to-peer Netw Appl. 2021;14:3043–57.10.1007/s12083-021-01162-xSearch in Google Scholar PubMed PubMed Central
[45] Hedabou M, Frobenius A. Map approach for an efficient and secure multiplication on Koblitz curves. Int J Netw Security. 2006;3:233–7.Search in Google Scholar
[46] Sharma A, Georgi M, Tregubenko M, Tselykh A, Tselykh A. Enabling smart agriculture by implementing artificial intelligence and embedded sensing. Computers Ind Eng. 2022;165:10793610.1016/j.cie.2022.107936Search in Google Scholar
[47] Boukhriss H, Hedabou M, Azougaghe A, New technique of localization a targeted virtual. In Proceedings of the 5th International Workshop on Codes, Cryptography and Communication Systems; 2014. p. 27–8.10.1109/WCCCS.2014.7107907Search in Google Scholar
[48] Yang H, Zhan K, Kadoch M, Liang Y, Cheriet M. Blcs: brain-like distributed control security in cyber physical systems. IEEE Netw. 2020;34:8–15.10.1109/MNET.011.1900275Search in Google Scholar
[49] Li H, Duan H, Zheng Y, Wang Q, Wang Y. A ctr prediction model based on user interest via attention mechanism. Appl Intell. 2020;50:1192–203.10.1007/s10489-019-01571-9Search in Google Scholar
[50] Vartouni AM, Teshnehlab M, Kashi SS. Leveraging deep neural networks for anomaly-based web application firewall. IET Inf Security. 2019;13:352–61.10.1049/iet-ifs.2018.5404Search in Google Scholar
[51] Huang S, Dai R, Huang J, Yao Y, Feng Z. Automatic modulation classification using gated recurrent residual network. IEEE Internet Things J. 2020;7:7795–807.10.1109/JIOT.2020.2991052Search in Google Scholar
[52] Dong Z, Wei J, Chen X, Zheng P. Face detection in security monitoring based on artificial intelligence video retrieval technology. IEEE Access. 2020;8:63421–33.10.1109/ACCESS.2020.2982779Search in Google Scholar
[53] Wang X, Yin S, Li H, Wang J, Teng L. A network intrusion detection method based on deep multi-scale convolutional neural network. Int J Wirel Inf Netw. 2020;27:503–17.10.1007/s10776-020-00495-3Search in Google Scholar
[54] Ye W, Chen H. Human activity classification based on micro-doppler signatures by multiscale and multitask fourier convolutional neural network. IEEE Sens J. 2020;20:5473–9.10.1109/JSEN.2020.2971626Search in Google Scholar
[55] Kansal L, Sharma V, Singh J. Multiuser massive MIMO-OFDM system incorporated with diverse transformation for 5G applications. Wirel Personal Commun. 2019;109:2741–56.10.1007/s11277-019-06707-1Search in Google Scholar
[56] Alyami H, Alosaimi W, Krichen M, Alroobaea R. Monitoring social distancing using artificial intelligence for fighting COVID-19 virus spread. Int J Open Source Softw Process (IJOSSP). 2021;12:48–63.10.4018/IJOSSP.2021070104Search in Google Scholar
[57] Arul R, Alroobaea R, Tariq U, Almulihi AH, Alharithi FS, Shoaib U. IoT-enabled healthcare systems using block chain-dependent adaptable services. Personal Ubiquitous Comput. 2021;1–15.10.1007/s00779-021-01584-7Search in Google Scholar
[58] Mukhtar H, Rubaiee S, Krichen M, Alroobaea R. An IoT framework for screening of COVID-19 using real-time data from wearable sensors. Int J Env Res Public Health. 2021;18:4022.10.3390/ijerph18084022Search in Google Scholar PubMed PubMed Central
[59] Bourouis S, Channoufi I, Alroobaea R, Rubaiee S, Andejany M, Bouguila N. Color object segmentation and tracking using flexible statistical model and level-set. Multimed Tools Appl. 2021;80:5809–31.10.1007/s11042-020-09809-2Search in Google Scholar
[60] Sodhi G, Kaur S, Gaba GS, Kansal L, Sharma A, Dhiman G. COVID-19: Role of Robotics, Artificial Intelligence, and Machine learning during pandemic. Curr Med Imaging. 2021;79:330–41.10.2174/1573405617666210224115722Search in Google Scholar PubMed
© 2022 Jing Niu et al., published by De Gruyter
This work is licensed under the Creative Commons Attribution 4.0 International License.
