• <tr id="yyy80"></tr>
  • <sup id="yyy80"></sup>
  • <tfoot id="yyy80"><noscript id="yyy80"></noscript></tfoot>
  • 99热精品在线国产_美女午夜性视频免费_国产精品国产高清国产av_av欧美777_自拍偷自拍亚洲精品老妇_亚洲熟女精品中文字幕_www日本黄色视频网_国产精品野战在线观看 ?

    GMLP-IDS:A Novel Deep Learning-Based Intrusion Detection System for Smart Agriculture

    2023-12-12 15:50:00AbdelwahedBerguigaAhlemHarchayAymanMassaoudiMossaadBenAyedandHafedhBelmabrouk
    Computers Materials&Continua 2023年10期

    Abdelwahed Berguiga,Ahlem Harchay,Ayman Massaoudi,Mossaad Ben Ayed and Hafedh Belmabrouk

    1Department of Computer Science,College of Science and Arts in Gurayat,Jouf University,Sakakah,Saudi Arabia

    2Olive Research Center,Jouf University,Sakakah,Saudi Arabia

    3Department of Electronic Industrial,ENISo,Sousse University,Sousse,Tunisia

    4Department of Physics,College of Science at Zulfi,Majmaah University,Majmaah,Saudi Arabia

    ABSTRACT Smart Agriculture,also known as Agricultural 5.0,is expected to be an integral part of our human lives to reduce the cost of agricultural inputs,increasing productivity and improving the quality of the final product.Indeed,the safety and ongoing maintenance of Smart Agriculture from cyber-attacks are vitally important.To provide more comprehensive protection against potential cyber-attacks,this paper proposes a new deep learning-based intrusion detection system for securing Smart Agriculture.The proposed Intrusion Detection System IDS,namely GMLPIDS,combines the feedforward neural network Multilayer Perceptron (MLP) and the Gaussian Mixture Model(GMM) that can better protect the Smart Agriculture system.GMLP-IDS is evaluated with the CIC-DDoS2019 dataset,which contains various Distributed Denial-of-Service(DDoS)attacks.The paper first uses the Pearson’s correlation coefficient approach to determine the correlation between the CIC-DDoS2019 dataset characteristics and their corresponding class labels.Then,the CIC-DDoS2019 dataset is divided randomly into two parts,i.e.,training and testing.75%of the data is used for training,and 25%is employed for testing.The performance of the newly proposed IDS has been compared to the traditional MLP model in terms of accuracy rating,loss rating,recall,and F1 score.Comparisons are handled on both binary and multi-class classification problems.The results revealed that the proposed GMLP-IDS system achieved more than 99.99%detection accuracy and a loss of 0.02%compared to traditional MLP.Furthermore,evaluation performance demonstrates that the proposed approach covers a more comprehensive range of security properties for Smart Agriculture and can be a promising solution for detecting unknown DDoS attacks.

    KEYWORDS Drones;DDoS attacks;Internet of Things;deep learning;multilayer perceptron;gaussian mixture model;Industry 5.0;Agricultural 5.0

    1 Introduction

    The emergence and development of the fourth industrial into the creation of smart and interconnected manufacturing systems that can optimize production,reduce costs,and increase efficiency have created new opportunities for businesses to improve their competitiveness,productivity,and sustainability [1].Indeed,the Smart Factory concept has been defined as the latest industrial trend,known as Industry 5.0[2–4].This concept integrates advanced technologies into production processes,such as big data,artificial intelligence,the Internet of Things(IoT),robotics,and cloud computing.

    Smart farming,also called intelligent farming or Agriculture 5.0,represents the adoption of new technologies for strengthening the efficiency of agro-industrial companies.In such regard,Agriculture 5.0 is anticipated to be transformed by Industry 5.0,leading to a radical transformation in such a sector [5].Industry 5.0 and Agriculture 5.0 are increasingly crucial to our lives since they simplify our daily tasks and interact with the world around us.Also,the application of IoT in Agriculture 5.0 ranges from family farming and plays a vital role in changing the overview of conventional agriculture scenarios.For example,the digitalization of agriculture has been brought about by the Internet of Things (IoT),cloud computing,and Artificial Intelligence (AI),where machines and production systems are connected,monitored,and controlled through a network of sensors and actuators [6].This can assist farmers in taking appropriate measures at the appropriate moment by monitoring soil moisture,temperature,and other environmental factors,consequently making the production process more scientific regarding fertilization,pest control,and improving crop yields.

    Internet of Things,remote sensing,machine learning,and deep learning revolutionize agricultural value chains as they are deployed for advanced data analysis.Additionally,machine learning enables algorithms to collect data from sensors,such as soil moisture sensors,weather stations,and satellite imagery.Collected data are then analyzed to identify and diagnose crop diseases,classify soil types,and monitor weather patterns,among other applications[7].Also,machine learning algorithms can predict the best time for planting,watering,and harvesting crops,allowing farmers to optimize their operations and increase productivity[8].

    Conversely,despite the benefits of implementing IoT in Agriculture 5.0 and Industry 5.0,several challenges could impact the agricultural and food sectors in the coming decades.The large volume of data generated by IoT devices can be overwhelming and require new data management and analysis approaches.In particular,the lack of intrinsic security measures increases the IoT’s vulnerability to cyber-attacks and data breaches[9].These data are subject to DDoS,Sybil attacks,and single-point failure.These attacks can cripple an entire agriculture system,causing significant financial losses and potentially risking lives[10].

    Nevertheless,there are many risks of integrating IoT in the agricultural sector.Indeed,IoT devices and sensors generate and transmit sensitive data,such as crop yield,soil moisture,humidity,light,and nutrient levels.However,if this data is not secured correctly,it can be easily targeted for unauthorized access,leading to loss of device connectivity or collecting sensitive personal information.Moreover,IoT devices are vulnerable to malware and ransomware attacks,which can disrupt operations and compromise data integrity.Another type of threat that could be encountered in the integration of IoT in the agricultural sector is Distributed Denial of Service(DDoS)attacks,where attacker overwhelms IoT devices with a tremendous amount of traffic,making the network congested,heavy,or even data loss,specifically at peak times.

    Moreover,IoT security-related are significant and must be carefully discussed and addressed appropriately.In this light,Intrusion Detection Systems(IDS)are becoming crucial for detecting and mitigating DDoS attacks in Agriculture 5.0[11].By leveraging machine learning algorithms,IDS can quickly detect unusual patterns in network traffic and identify potential DDoS attacks before they cause any harm.In addition,machine learning algorithms can also be adapted to detect new attack patterns,making IDS a powerful tool in ensuring the security of Agriculture 5.0.Overall,integrating IDS and machine learning is essential for protecting the agriculture industry against malicious attacks and ensuring that Agriculture 5.0 continues to provide safe and efficient farming practices.

    For example,Al Jouf City,a province in the Northern Region of Saudi Arabia,is the largest plant producing olive oil in the Middle East and has been identified as a potential key player in the production of the finest quality of olive oil.Al Jouf Agricultural Development Company(Saudi Arabia)has made tremendous efforts to modernize olive farming by introducing new technologies such as drip irrigation and precision farming.To achieve such goals,the company deploys drones,sensors,and satellite imagery to help farmers collect data,monitor their crops and make data-driven decisions about planting,harvesting,and irrigation.However,with an area of 7730 hectares and more than 5 million olive trees,it is hard to prevent and detect whether there are cyber-attacks on collected data.

    In such cases,machine learning and deep learning can play a more vital and robust role in analyzing vast amounts of data collected from various sources and detecting and preventing if there are cyber-attacks in farming systems[12–14].Attackers can exploit vulnerabilities in connected devices and other equipment,compromising data security and potentially causing significant harm to farming operations.More crucially,machine learning and deep learning algorithms can aid in analyzing large amounts of data to identify potential threats and detect whether there are cyber-attacks.As a result,farmers can take appropriate measures to prevent or mitigate cyber-attacks’impact on their operations[15].

    The remainder of the paper is carried out as follows: Section 2 discusses the relevant scientific literature on IDS models.Then in Section 3,we describe our proposed GMLP-IDS with its main components.The evaluation experiments and results outcomes are described in Section 4.The final part of this paper outlines in Section 5 the conclusion and future works.

    2 Related Works

    The paper [16] provided a novel approach for detecting assaults in the Internet of Things (IoT)using deep learning.The study focused on the security problems faced by IoT systems due to their distributed design and the enormous number of connected devices.The suggested solution comprises employing a distributed system of edge devices and a centralized server for the training and inference of a deep learning model.The edge devices collect and pre-process data from the IoT devices and transfer it to the centralized server for model training.The learned model is then transmitted back to the edge devices for inference,enabling real-time detection of assaults.

    The performance of the suggested strategy is assessed using the KDDCUP’99 dataset and compared to that of more established machine learning and rule-based methods.The findings demonstrate that the proposed deep learning methodology achieves good results for detection accuracy,very low false positive rate,and detection time compared to the alternative approaches.Overall,the study makes a significant contribution to the field of IoT security by outlining a distributed strategy for overcoming the difficulties associated with identifying attacks in complex IoT systems.However,the proposed scheme exchanges immense control messages between edge devices,fog nodes,and the central server,leading to large-scale IoT systems bottlenecks.Moreover,the technique depends on the edge devices’and fog nodes’dependability and security,which may only sometimes be the case in practical use.Finally,the study does not mention the suggested approach’s potential for expansion to handle larger datasets or more complex attacks.

    Thamilarasu et al.[17]designed an intelligent intrusion detection approach for IoT systems using deep learning algorithms.In their work,authors combined Convolutional Neural Networks(CNNs)and Long Short-Term Memory(LSTM)to extract features and classify network traffic anomalies in real-time.The approach is evaluated on the Bot-IoT dataset,and the results show that it can detect various types of attacks with a high detection rate and low false positive rate.However,the authors do not take into account the dynamic nature of IoT networks,where devices can join and leave the network at any time,resulting in changing network traffic patterns.Also,the approach relies on the availability of labeled training data for deep learning algorithms,which is difficult to be applied in practice.Finally,the proposed approach does not consider the impact of network latency and communication overhead on real-time attack detection.Overall,this paper provides a promising approach for detecting intruders in IoT systems,but further research is needed to address these weaknesses and evaluate the proposed approach in real IoT systems is required.

    In [18],the authors provided a comprehensive review of the use of Federated Learning (FL)for Intrusion Detection Systems (IDS) in the Internet of Things (IoT) context.They focus on Federated Learning (FL) which is proposed as a potential solution to address the challenges posed by these networks’distributed and heterogeneous nature.Several advantages and limitations of FLbased intrusion detection are discussed in the paper,including the need for secure and efficient communication protocols,the selection of appropriate machine learning models,and the effect of data privacy concerns on FL algorithms’performance.The authors conclude that FL has the potential to improve the effectiveness and efficiency of IDS in IoT,but further research is needed to address the challenges mentioned above and ensure the security and privacy of the data involved.

    A novel approach for anomaly detection in Internet of Things (IoT) networks was proposed by Ullah et al.[19].The authors used Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) networks instead of a standard RNN for anomaly detection in IoT networks.The proposed approach continuously analyzes the incoming network traffic and compares it with the learned standard patterns to detect anomalies in real-time.Compared to other proposed approaches[20],the output results show that the proposed deep learning model based on the recurrent neural approach outperforms previous deep learning implementations in terms of precision,recall,accuracy,and F1 score,as well as its ability to detect anomalies in real-time makes it a promising approach for intrusion detection in IoT environments.Nonetheless,even if authors explored that the proposed solution is based on a regular pattern of behavior in the IoT network and remains consistent over time,the optimal values for hyper-parameters for the Recurrent Neural Network(RNN)model,such as the number of layers,the number of neurons per layer,and the learning rate may depend on the characteristics of the specific IoT networks.They may require extensive experimentation to find the best configuration.

    In[14],the authors addressed how smart farming can improve Agriculture 4.0 and then increase efficiency,productivity,and sustainability.The authors emphasized using cutting-edge technologies,namely,cloud computing,edge computing,AI,IoT,big data,and 6G in agriculture.The authors discussed the benefits of “smart farming”,including precision farming,resource optimization,and predictive maintenance.Also,they explore the problems devoted to traditional farming,such as increasing food production,water scarcity,and the need for climate change.According to these characteristics,smart farming can help to solve resource optimization usage and lack of technical expertise.However,a comprehensive analysis of the challenges faced by Agriculture 4.0 technologies is not outlined by the authors.Also,a detailed discussion of the policy implications of implementing Agriculture 4.0 technologies in farming is not present.

    In another recent study,Friha et al.[21]developed a novel intrusion detection system(IDS)for the Agricultural Internet of Things (IoT) that uses Federated Learning (FL) as its core technology.The proposed scheme is based on a distributed architecture,where multiple IoT devices collect data and train a machine-learning model using the Federated Learning approach.This latter allows knowledge sharing with privacy maintenance and cost reduction.The model is then used to detect anomalies and intrusions in real-time.The paper also provided a detailed description of the FELIDS architecture and its components,including the data collection module,the FL module,and the IDS module.Consequently,FELIDS is then evaluated using a dataset of real-world attacks on IoT-based agriculture systems,and performance results are compared to other IDS systems.Authors demonstrate that the proposed scheme FELIDS achieves a higher detection rate and lower false positive rate than classic versions of machine learning,making it a promising solution for securing IoT-based agriculture systems.

    Gupta et al.[22]have used a hierarchical deep-learning model to analyze network traffic and detect anomalies.The model is trained on an extensive network traffic dataset to learn patterns and identify abnormal behavior.Then,it monitors network traffic in real-time and generates alerts when detecting anomalous activity.The proposed model aims to enhance the cybersecurity of multi-cloud healthcare systems.Authors have also shown that the proposed approach outperforms traditional approaches in terms of accuracy and detection rates.Overall,the proposed approach has the potential to improve the security of healthcare systems and protect sensitive patient information from cyber threats.

    Gudla et al.[23]designed a new detection scheme,named deep intelligent DDoS attack detection scheme(DI-ADS),for Distributed Denial of Service(DDoS)attacks in fog-based Internet of Things(IoT) applications.Fog defender utilizes a deep learning-based approach to allow only legitimate requests to be accessed to the cloud.DI-ADS employs a distributed architecture that uses edge devices located at the fog layer of the IoT network to collect and pre-process data before sending it to a central server for analysis.The system utilizes a Convolutional Neural Network(CNN)to classify network traffic as either normal or attack traffic.Traffic attacks are generated using the well-known Kali Linux tool,and the performance of DI-ADS is evaluated using various metrics,including behavior detection time,false positive rate,loss rate,and accuracy.

    The authors of[24]proposed a hybrid lightweight system to detect cyber-attacks in the Internet of Medical Things(IoMT)fog environment.The proposed system uses machine learning algorithms to detect network anomalies and abnormal behaviors,and it consists of two main components: a traffic profiler and a hybrid anomaly detection model.The traffic profiler collects network traffic data and extracts relevant features.The hybrid anomaly detection model combines two machine learning algorithms,a one-class Support Vector Machine(SVM)and a decision tree,to detect network anomalies and abnormal behaviors.Experimental results show that the proposed system can detect network anomalies and abnormal behaviors with high accuracy and low computational and memory overhead,making it suitable for use in resource-constrained IoMT environments.

    The study conducted by Jiang et al.[25] proposed a dynamic ensemble algorithm for anomaly detection in IoT-imbalanced data streams.The algorithm considers the dynamics,continuity,and data streams generated in IoT systems.To address the data imbalance problem,the proposed algorithm uses a combination of multiple classifiers to improve anomaly detection accuracy.Indeed,the algorithm uses the Borderline-SMOTE approach to generate minority samples in training sets.The Borderline-SMOTE dynamic weighted LightGBM (BSDWLGB) algorithm also uses LightGBM as base classifiers and adopts a chunk-based dynamic weighted ensemble mechanism,enabling classifiers to adapt to data distribution changes and remove the classifier whose accuracy performance is lower than a given threshold.The proposed algorithm is evaluated using several datasets,and the results show that it outperforms several state-of-the-art algorithms in terms of accuracy,precision,recall,and F1 score.Furthermore,the proposed algorithm can effectively detect anomalies in IoT-imbalanced data streams and can be used in various applications,such as intrusion detection,fault diagnosis,and predictive maintenance.

    The study conducted by [26] proposed a novel deep learning-based anomaly detection method called FlowADGAN for network intrusion detection systems(NIDS).FlowADGAN is inspired by the successes of Generative Adversarial Networks (GANs) for detecting anomalies in the area of Computer Vision and Images.It consists of two components: a generator and a discriminator.The generator learns to generate normal network traffic data,while the discriminator learns to distinguish between normal and abnormal network traffic data.Furthermore,a comprehensive evaluation of FlowADGAN demonstrates that the proposed scheme can detect known and unknown network intrusions and is a promising deep learning-based anomaly detection method for NIDS.

    Wang et al.[27] proposed a new method to classify malicious code families by merging convolutional neural networks (CNNs) with generative adversarial networks (GANs).The authors emphasized that conventional malware detection methods are not effective against the continuous evolution of malicious code variants.To deal with the associated challenges,new methods are crucial to enhance the efficiency and accuracy of malware detection.The proposed framework uses a code visualization method to convert malicious code into images,which can then be fed into a CNN for feature extraction and classification.According to the evaluation performance of the proposed method on a dataset of malicious code families,the authors show that the method achieves a classification accuracy of 97.78%and outperforms the conventional malware detection methods.The summary of deep learning approaches for network intrusion detection for IoT networks is presented in Table 1.

    3 Proposed Framework:GMLP-IDS

    3.1 Network Model

    This research deploys a network model with three layered architectures:The uppermost layer is the cloud layer,the middle layer is referred to as the fog layer,and the lower layer is composed of IoT or smart devices.The three-layered network model is a prevalent architecture in the industry 5.0 paradigm,which has gained significant global traction as a fundamental driver of industrial growth and transformation.

    The topmost layer,as shown in Fig.1,is the cloud layer,consisting of many cloud storage nodes and a cloud data center.The cloud layer provides scalability and resilience for storing and processing large amounts of data.It serves as a centralized storage unit for updated information on the behaviors of various IoT devices.Additionally,the IoT devices in the lower layer receive regular updates from the cloud layer,ensuring that their behaviors are up-to-date,whether normal or under DDoS status.

    Figure 1:System architecture for the proposed deep learning-based IDS for DDoS attack in Agriculture 5.0

    The intermediate layer is called the fog layer,consisting of fog nodes(FN1,FN2,...,FNn)for data processing and storage.It is designed to provide low latency and efficient processing power to meet the needs of applications running on the devices in the lower layer.These nodes offer services to nearby IoT devices and maintain timely records of their behavior.This intermediate layer acts as a bridge between different IoT systems and the cloud layer,thus facilitating the interoperability of IoT systems.Communication between fog nodes is carried out via wired or wireless links.The fog layer is connected to the upper layer via cloud gateway(CG)and base stations(BSs),and communication is carried out using wired or wireless connections.The IoT Device gateway facilitates Communication between the fog and the lower layer.

    The lower layer,also known as the IoT or smart devices layer,consists of drones and sensors.In this layer,drones detect changes or events in the environment and process substantial amount of data.By flying over fields,drones can communicate with other IoT devices (e.g.,sensors) and be placed directly in the soil to measure soil conditions:temperature,humidity,light,soil moisture,and nutrient levels.Also,drones equipped with cameras and sensors can monitor crop health,detect anomalies,and respond to all end-user requests.As mentioned earlier,these drones communicate with the fog and cloud layers to perform speedy computation and service.However,despite their essential role,IoT devices in this layer have limited storage and computational capacity.

    Using our proposed system for DDoS attacks has several potential ethical implications,and the system must handle data privacy issues.In fact,our system handles collected data privately to avoid unauthorized access and ensure that collected data is processed and stored transparently and securely.Secondly,sensitive information is often encrypted to protect it from unauthorized access.Finally,as illustrated in Fig.1,our IDS ensures that user data is used for its intended purpose and stored at the cloud data center.The cloud data center provides a centralized infrastructure for storing,processing,and managing large amounts of data sent from the agricultural sensors layer in a cloud computing environment.

    3.2 DDoS Attack Model

    In this section,we examine the system’s security issues where the Attacker(Ai)attempts to gain access to a communication channel and control targeting IoT devices such as drones,sensors,etc.,through various methods to process a large amount of data.

    To achieve this,as shown in Fig.2,the attacker compromises the IoT devices by replacing their normal functions with malicious code,causing them to behave like malicious nodes.The compromised IoT devices then use various techniques to launch attacks on fog nodes at the intermediate layer,such as sending excessive requests to jam the network and gain control.This study focuses on 11 DDoS attacks,including Network Time Protocol(NTP),Domain Name System(DNS),Lightweight Directory Access Protocol (LDAP),Microsoft SQL Server (MSSQL),Network Basic Input/Output System(NetBIOS),Simple Network Management Protocol(SNMP),Simple Service Discovery Protocol(SSDP),User Datagram Protocol(UDP),the delay that can occur when sending and receiving data over UDP (UDP-Lag),Synchronization (SYN),and Trivial File Transfer Protocol (TFTP),to evaluate our proposed work.To accomplish this,we utilized the CIC-DDoS2019 dataset for training and testing purposes[28].Indeed,as mentioned earlier,the proposed system model has three layered architectures:The uppermost layer is the cloud layer,the middle layer is referred to as the fog layer,and finally,the agricultural sensors layer comprises various IoT devices and drones that monitor agricultural environment data.The IoT and drones that monitor agricultural environment data make up the agricultural sensors layer.Actuators are activated in this layer,which also features a smart grid design and innovative energy technology for powering IoT devices.In each fog node,a deep learning-based intrusion detection system is placed.The IoT data are transmitted directly to the fog computing layer for analysis and machine learning algorithms.By outperforming nature in detecting DDoS attacks,GMLP-IDS is being implemented in this middle layer,the fog computing layer.The cloud layer provides scalability and resilience for storing and processing large amounts of data.It serves as a centralized storage unit for updated information on the behaviors of various IoT devices.

    Figure 2:DDoS attack model

    3.3 Data Preprocessing

    3.3.1 Description of the Dataset and Study Area

    In this section,we discuss the effectiveness of our proposed Intrusion Detection System using the CIC-DDoS2019 dataset,which was created as part of a study by the Canadian Institute for Cybersecurity(CIC).It contains a wide range of the most recent and popular Distributed Denial of Service(DDoS)attacks based on real-world information such as LDAP,MSSQL,NetBIOS,PortMap,UDP,SYN,and UDP-Lag,all available in both PCAP and CSV files [28,29].The CIC-DDoS2019 dataset,shown in Table 2,comprises approximately 50,063,123 records,with 50,006,249 rows dedicated to DDoS attacks and 56,874 rows for benign traffic.Each row has 88 features,and the dataset is divided into two classes:benign and malicious.The performance of our proposed GMLP-IDS model on the CICDDoS2019 dataset is evaluated to detect and classify malicious traffic in real-world scenarios accurately.Moreover,this dataset can be regarded as a valuable resource for researchers and practitioners in the field of network security by providing a suitable basis for evaluating the system’s performance against a wide range of DDoS attacks[30].

    3.3.2 Binary Pre-Processing

    As mentioned earlier,the CIC-DDoS2019 dataset contains a wide range of the most recent and popular DDoS attacks based on real-world information[28].This dataset was divided into 11 subtypes of attacks to have a better-engineered and more diverse set of attacks to be used for DDoS attack detection.Eighty-eight features describe each record on the dataset,and the attack type is based on binary classification,such as normal or malignant traffic.The features included in the dataset include source and destination IP addresses,port numbers,packet sizes,timestamps,and other features.Each attack generates a different kind of malicious network flow.

    To further verify the effectiveness of our GMLP-IDS,CIC-DDoS2019 is used to perform binary classification,so we divided the dataset into two portions:75%for the training process.In contrast,25% is allocated for the testing process.Table 3 summarizes the flow count for training and testing processes in the CIC-DDoS2019 dataset.

    3.3.3 Multiclass Pre-Processing

    The main objective of our proposed model in this section is multiclass classification in the context of DDoS attacks.Each CSV file from CIC-DDoS2019 contains 88 features related to network traffic and is labeled with one of two classes:BENIGN or a specific type of network attack(DNS,MSSQL,UDP,etc.).First,each file belonging to the dataset is unified into a separate information instance(data frame).Next,we extract a random number of rows from each of these data frames.We then append the resulting data frames into a single data frame.Once we combined all the data into a single information instance,we removed specific unwanted attributes(NaN,and infinite)from the selected data.

    Additionally,incorporating a data pre-processing phase in the deployment process leads to a more robust training phase and,consequently a more accurate model,as set out by [31].Table 4 presents the statistics of occurrences that have been selected from each data frame.The final step of data preprocessing is a one-hot encoding process[32].Indeed,datasets containing categorical information and label classes that cannot be directly processed by machine learning or deep learning models.To enable accurate predictions,the data must be transformed into numerical values using one-hot encoding.This technique involves creating binary vectors to represent each possible category of categorical variables.Also,to further improve the classification process,all fields are normalized by rescaling their dynamic ranges.This normalization step ensures that no single feature dominates the training process[33,34].

    Table 4:Statistic of occurrences selected from CIC-DDoS2019 dataset for multiclass classification

    3.3.4 Feature Selection

    As mentioned earlier,each row on the CIC-DDoS2019 dataset has 88 features and 50,063,123 records.To properly evaluate these huge amounts of data,selecting weighted features with a significant linear relationship with the class label is essential.For this purpose,Pearson’s correlation coefficient method was performed in this work to select the 15 features that had the highest correlation with the class label [35].By selecting only the most important features,the problem’s dimensionality can be reduced,and then the model will be simpler and more efficient.Additionally,reducing the number of features can help prevent overfitting,which occurs when the model is too complex.Fig.3 depicts the correlation matrix for the top 15 features for each DDoS attack classification(DNS,LDAP,MSSQL,NetBIOS,NTP,SNMP,SSDP,UDP,SYN,UDP-Lag,TFTP).

    It was found from Fig.3 that each correlation matrix is characterized by its own set of 15 features identified as the most significant in detecting and classifying DDoS attacks.These matrices visually represent the relationship between features and show how they are correlated to identify patterns and anomalies in network traffic.By analyzing the correlation matrices,we can gain insights into the behavior of network traffic and develop more accurate and effective DDoS attack detection and prevention techniques.

    Figure 3:(Continued)

    Figure 3:Selecting features by using the Pearson’s correlation coefficient method

    3.4 GMLP-IDS Module

    The main contribution of this research work is to develop a novel intrusion detection system,named GMLP-IDS,to identify cyberattacks in the field of cybersecurity and Agriculture 5.0.Our system utilizes a hybrid approach that combines two different models,namely the Gaussian Mixture Model(GMM)and MLP classifier,to accurately classify input data into one of several possible output classes.GMM is an extension of the single Gaussian model that assumes data points are generated from a mixture of multiple Gaussian distributions.GMM extracts features from input data,which are then passed on to the MLP classifier for classification and,ultimately,for detecting cyberattacks.The functional steps of the proposed GMLP-IDS system are described as follows:

    1.Fitting a GMM to the training data:In this step,we have used the Expectation-Maximization(EM)algorithm to estimate the underlying probability distribution of the input features.The GMM is configured with a number of Gaussian components equal to 2,covariance type set to“full”,and we fix the number of iterations as 100.

    2.Data transformation:Once the GMM is trained,we transform the input data into a set of feature vectors.This set of feature vectors represents the probability of each input data point belonging to each Gaussian component in the Gaussian Mixture Model.

    3.Training the MLP classifier:Use the transformed feature vectors as input to train an MLP classifier using the backpropagation technique.This is done using the TensorFlow framework.

    4.Making predictions:This last step is used to make predictions on new data by first transforming the new data using the same GMM that was used to transform the training data.Then,the MLP classifier is used to predict the class label of the new data.

    As shown in Fig.4,supervised learning is utilized to distinguish between regular traffic and DDoS attacks using a GMLP-IDS.The GMM,based on unsupervised learning,is utilized to model the probability distribution of a dataset as a combination of multiple Gaussian distributions,each representing a subpopulation or cluster within the data.Then,the MLP takes the numeric and normalized values of the selected features prepared in the previous step to predict the class label of the new data.Traffic engineers identify and label each cluster,which is then used to update the GMLPIDS and enable it to identify previously unseen attacks.Expert analysis is utilized to label and identify unknown traffic or attacks that the Gaussian mixture model has classified.The deep learning model is updated through incremental learning.

    Figure 4:Functional diagram for the proposed deep learning-based IDS for DDoS attack in Agriculture 5.0

    4 Experiments

    We performed experiments on the Ubuntu 20.6 operating system using 12 GB of memory.To generate various MLP models for comparison purposes,we deployed two popular deep-learning opensource tools,namely Keras and TensorFlow.Also,Pearson’s correlation coefficient was utilized to determine the correlation between dataset characteristics and their corresponding class labels.

    To achieve high-performance results for detecting attacks on farm systems based on Agriculture 5.0,we selected the top 15 features for further processing using deep learning techniques.Given a large number of features,analyzing each one individually for feature selection is impractical.Therefore,Pearson’s correlation coefficient approach is called to reduce the number of features to a more manageable subset.Indeed,by focusing on the features with the highest correlation with the class label,Pearson’s correlation coefficient is still the most accurate and foolproof approach to retain the most important predictors of the target variable.

    4.1 Performance Metrics

    Selecting the appropriate performance criteria is vital for effectively evaluating machine learning and deep learning algorithms.In this section,we examine the overall performance of our IDS approach by testing its impact on prediction against attacks.The proposed GMLP-IDS model is evaluated and compared based on four performance indicators,including precision,F1 score,recall,and accuracy,as specified in Eqs.(1)–(4).Additionally,our study supports the use of confusion matrix analysis for classification accuracy when it makes predictions,as illustrated in Table 5.

    Table 5:Confusion matrix

    Where TP,TN,FP,and FN four outcomes are often used to create a confusion matrix.In the context of binary classification,these four outcomes are named true positive (TP),true negative(TN),false positive(FP),and false negative(FN).True positive(TP)occurs when the model correctly identifies a positive sample as positive(attack data that is accurately classified as an attack).The term True negative(TN)refers to a model that correctly identifies a negative sample as negative(benign data that is correctly identified as benign).False positive(FP)occurs when the model incorrectly identifies a negative sample as positive (benign data that was noted for any data that were wrongly identified as an attack).False negative (FN) occurs when the model incorrectly identifies a positive sample as negative(attack data that have been labeled as benign).

    4.2 Results and Discussion

    The performance results of deep learning techniques in terms of binary classification are shown in Table 6.Indeed,as mentioned earlier,our proposed intrusion detection approach and the traditional MLP are conducted on the CICDDoS2019 dataset to detect binary and multiclass attacks.We use Accuracy,Precision,Recall,and F1 score as performance indicators for each attack type.It can be observed clearly that GMLP-IDS,outperforms the traditional MLP in terms of accuracy and precision scores for most attack types.Table 6 depicts that the proposed classification model,GMLPIDS,achieved a high accuracy rating of 99.96%,99.61%,and 99.98%for DNS,NetBIOS,and SYN attacks,respectively.In comparison,the MLP model showed a correct accuracy rating of 95.94%for DNS attacks,99.48%for NetBIOS attacks,and reached 99.95%for SYN attacks.The above can be explained by the operation mode of the proposed model on making ethical decisions in diagnosing transformed feature vectors outputted from the Gaussian Mixture Model.Moreover,the GMLPIDS model obtained a higher precision rating for DNS and SYN attacks,achieving ratings of 98.64%and 99.99%,respectively,compared to the MLP model’s ratings of 86.80%and 77.77%.However,for MSSQL attacks,the MLP model achieved a higher precision score,with a score of 98.52%compared to our proposed approach model’s rating of 81.66%.

    Table 6:Performance results of GMLP-IDS and traditional MLP models to normal and various types of attacks with binary classification

    Overall,we can affirm that the proposed GMLP-IDS model is more effective in detecting attacks,namely,LDAP,NetBIOS,SYN,etc.These findings demonstrate the importance of choosing an appropriate model and evaluation metrics for detecting different types of attacks.As mentioned earlier,Pearson’s correlation coefficient approach deployed with GMM and MLP has retained the most important features of the target variable.Such a solution gives a more accurate and foolproof manageable subset of the trained process.

    On the other hand,Table 7 shows accuracy,precision,recall,and F1 score for each of the two compared machine learning models in the case of multi-class classification.The results show that GMLP-IDS gives a higher rating prediction in most attack types.Specifically,the proposed IDS has demonstrated optimistic results in detecting SYN-based attacks with a recall rating of 92.50%and NetBIOS-based attacks with a recall of 96.88%.Compared to the MLP model,which provides detection of SYN-based attacks with recall scores of 90.13%and NetBIOS-based attacks with a recall rating of 94.87%,we can identify that the GMLP-IDS model achieved near-perfect accuracy on multiclass classification and,therefore,improving the accuracy and effectiveness of intrusion detection systems in real-world network environments.

    Table 7:Performance results of GMLP-IDS and traditional MLP models to normal and various types of attacks with multi-class classification

    Fig.5 presents the training accuracy and loss convergence concerning epochs for the SYN-based attack.Indeed,a SYN-based attack is a type of network level where the attacker exploits the fact that the TCP protocol uses a three-way handshake to establish a connection between two systems.In such cases,the attacker sends a SYN packet to initiate the handshake with drones but does not complete it by sending the necessary ACK packet.This causes the drone system to keep waiting for the ACK packet,tying up its resources and preventing it from responding to legitimate requests.Moreover,the main disadvantage of IoT and drones is their high energy consumption,which can lead to the rapid depletion of batteries or other power sources.

    Drones,in particular,require a significant amount of energy to maintain flight and perform various tasks,such as data communication with soil sensors that continuously measure soil parameters such as moisture,pH levels,and nutrient content.However,if a drone runs out of battery mid-flight,it could crash and cause damage to the crops or other equipment.Furthermore,with such SYN-based attacks,IoT devices such as sensors and smart devices that operate on batteries can also quickly deplete their energy sources,especially when deployed in remote or harsh environments.Additionally,frequent battery replacements can be costly and time-consuming,reducing the efficiency and productivity of IoT and drone applications.

    From the profiling perspective of binary classification in Fig.5,we can see that the training accuracy of GMLP-IDS converges much faster than the traditional MLP model.For GMLP-IDS,the training accuracy curve begins at 99.98%and goes up to 99.99%,and after that,the convergence of training accuracy becomes stable at the 20thepoch.For training loss,Figs.5c and 5d depict the convergence of the loss with epochs.From Fig.5c,we can see that the GMLP-IDS achieves the lowest loss of 0.02%at the 25thepoch.

    On the other hand,concerning multiclass classification,Fig.6 illustrates that GMLP-IDS converges must be faster than MLP in the case of training accuracy.Indeed,the training begins at 99.89%and goes up to 99.94%at the 10thepoch.For training loss,Figs.6c and 6d depict the convergence of the loss with epochs.For multiclass classification,we can see the GMLP-IDS achieves the lowest loss of 0.2%at the 15thepoch,and MLP achieves the lowest loss of 0.25%at the fifth epoch.

    Figure 6:(Continued)

    Figure 6:Model accuracy and loss of MLP and GMLP-IDS models with multiclass classification

    5 Conclusions and Future Guidelines

    The present study proposes a novel Deep-intrusion detection system named GMLP-IDS for Smart Agriculture,which leverages a combination of the Gaussian Mixture Model and Multi-layer Perceptron.The CIC-DDoS2019 dataset is employed to train and validate the proposed solution,which is geared towards Agricultural 5.0.This approach encompasses the integration of IoT,drones,and sensors to optimize farming practices.By addressing the drawbacks of traditional agriculture,GMLPIDS offers an innovative solution to enhance the security of smart agriculture systems.The study compares GMLP-IDS with a conventional MLP approach based on four key performance indicators:precision,F1 score,recall,and accuracy.Furthermore,Pearson’s correlation coefficient approach is utilized to reduce the number of features to a more manageable subset.The findings highlight the superior performance of GMLP-IDS in intrusion detection accuracy,thus underscoring its potential to revolutionize the field of smart agriculture security.

    In future work,we plan to deploy drones equipped with cameras to capture images of olive leaves to identify different types of infections,including Aculus olearius,olive peacock spot,olive scab,and healthy leaves.Indeed,as mentioned earlier,such open olive farming at Al Jouf,with an area of 7730 hectares and over 5 million olive trees,it is essential a real-time control is achieved in such open-field farming.Additionally,given the massive amount of collected leaves images,data should be protected from unauthorized access or interception.On the other hand,by outperforming nature in detecting DDoS attacks,GMLP-IDS is considered in the proposed framework for being implemented at the fog layer.However,time and memory consumption are crucial considerations due to the limited resources of edge and fog devices.To deal with such issues,balancing the time and memory consumption in DDoS prevention at the fog layer involves optimizing the analysis and response mechanisms,employing memory-efficient strategies,ensuring scalability,and utilizing collaborative defense techniques.Moreover,in our work,we have deployed only one standard dataset,namely CIC-DDoS2019 dataset.The dataset contains a variety of DDoS attacks.However,some network traffic patterns may not be applied in a real agriculture environment,or the dataset might not fully represent the confusions and complexities of real-world network traffic,leading to the proposed IDS being less effective in detecting different types of attacks.Therefore,it is crucial to complement the use of the CIC-DDoS2019 dataset with other relevant datasets and regularly monitor the IDS to decrease false alarms and missed detection rates.

    Acknowledgement:The authors extend their appreciation to the Deanship of Scientific Research in Cooperation with Olive Research Center at Jouf University.

    Funding Statement:This work was funded by the Deanship of Scientific Research in Cooperation with Olive Research Center at Jouf University under Grant Number(DSR2022-RG-0163).

    Author Contributions:The authors confirm contribution to the paper as follows: study conception and design:Abdelwahed Berguiga,Ayman Massaoudi,Ahlem Harchay;data collection:Abdelwahed Berguiga,Mossaad Ben Ayed;analysis and interpretation of results: Abdelwahed Berguiga,Ahlem Harchay,Ayman Massaoudi,Mossaad Ben Ayed,Hafedh Belmabrouk;draft manuscript preparation:Abdelwahed Berguiga,Ahlem Harchay,Ayman Massaoudi,Mossaad Ben Ayed,Hafedh Belmabrouk.All authors reviewed the results and approved the final version of the manuscript.

    Availability of Data and Materials:The data used in this paper can be requested from the corresponding author upon request.

    Conflicts of Interest:The authors declare that they have no conflicts of interest to report regarding the present study.

    亚洲性久久影院| 嘟嘟电影网在线观看| 男男h啪啪无遮挡| 欧美成人a在线观看| 久久久久网色| 美女高潮的动态| 久久久久久久久久久丰满| 男人狂女人下面高潮的视频| 色哟哟·www| 我的女老师完整版在线观看| 久久人人爽av亚洲精品天堂 | 少妇熟女欧美另类| 精品人妻熟女av久视频| 成年人午夜在线观看视频| 成人无遮挡网站| 亚洲婷婷狠狠爱综合网| 六月丁香七月| 一区二区三区免费毛片| 国产成人a∨麻豆精品| 精品少妇久久久久久888优播| 我要看日韩黄色一级片| 成人影院久久| 男女国产视频网站| 黑人猛操日本美女一级片| 亚洲自偷自拍三级| av又黄又爽大尺度在线免费看| 最近最新中文字幕免费大全7| 久久99热这里只有精品18| 日韩伦理黄色片| 婷婷色综合大香蕉| 欧美bdsm另类| 26uuu在线亚洲综合色| 日本色播在线视频| 欧美日韩在线观看h| 成人黄色视频免费在线看| 国产极品天堂在线| 蜜臀久久99精品久久宅男| 中文字幕av成人在线电影| 久久热精品热| 男女啪啪激烈高潮av片| 午夜福利高清视频| 亚洲精品视频女| 亚洲精品自拍成人| 亚洲av中文av极速乱| 亚洲av国产av综合av卡| 亚洲av国产av综合av卡| 国产黄色免费在线视频| 亚洲av中文av极速乱| freevideosex欧美| 亚洲真实伦在线观看| 蜜臀久久99精品久久宅男| 午夜福利在线在线| 街头女战士在线观看网站| 亚洲美女黄色视频免费看| 中文欧美无线码| 国产精品无大码| 精品久久久久久久久亚洲| 99热国产这里只有精品6| 亚洲欧洲国产日韩| 搡女人真爽免费视频火全软件| 国产一区二区在线观看日韩| 视频区图区小说| 久久久久久伊人网av| 十分钟在线观看高清视频www | 国产av码专区亚洲av| 夜夜看夜夜爽夜夜摸| 全区人妻精品视频| 亚洲av日韩在线播放| 午夜日本视频在线| 街头女战士在线观看网站| 美女主播在线视频| 亚洲美女视频黄频| 十分钟在线观看高清视频www | 亚洲精品aⅴ在线观看| 国内少妇人妻偷人精品xxx网站| 亚洲精品,欧美精品| 黑人高潮一二区| 免费av中文字幕在线| 人人妻人人添人人爽欧美一区卜 | 人妻 亚洲 视频| 久久女婷五月综合色啪小说| 亚洲天堂av无毛| 在线观看国产h片| 久久人人爽人人片av| av专区在线播放| 熟女电影av网| 少妇人妻久久综合中文| 欧美精品国产亚洲| 国产av国产精品国产| 色视频www国产| 蜜桃在线观看..| 国产永久视频网站| 日韩中文字幕视频在线看片 | h视频一区二区三区| 免费大片黄手机在线观看| 久久99精品国语久久久| 久久97久久精品| 国产av国产精品国产| 亚洲欧美日韩另类电影网站 | 国国产精品蜜臀av免费| av又黄又爽大尺度在线免费看| 我要看日韩黄色一级片| 久久久久精品性色| 国产日韩欧美在线精品| 99re6热这里在线精品视频| 久久精品人妻少妇| 久久久成人免费电影| 18禁在线播放成人免费| 欧美成人a在线观看| 男人爽女人下面视频在线观看| 国产精品久久久久久精品古装| 精品人妻一区二区三区麻豆| 色5月婷婷丁香| 一个人看的www免费观看视频| 久久久亚洲精品成人影院| 亚洲精品色激情综合| 我要看日韩黄色一级片| 午夜福利影视在线免费观看| 在线免费观看不下载黄p国产| 精品亚洲成国产av| 成人综合一区亚洲| 国产熟女欧美一区二区| 成人毛片60女人毛片免费| 免费观看a级毛片全部| 国产国拍精品亚洲av在线观看| 五月玫瑰六月丁香| 免费观看无遮挡的男女| 在线 av 中文字幕| 欧美高清性xxxxhd video| 久久毛片免费看一区二区三区| 美女脱内裤让男人舔精品视频| 男人舔奶头视频| 精华霜和精华液先用哪个| 一级片'在线观看视频| 国产黄片视频在线免费观看| 国产在视频线精品| 欧美xxxx性猛交bbbb| 久久久久久久国产电影| 精品亚洲乱码少妇综合久久| 亚洲人成网站在线观看播放| 人妻制服诱惑在线中文字幕| 交换朋友夫妻互换小说| 亚洲激情五月婷婷啪啪| 精品国产乱码久久久久久小说| 精品午夜福利在线看| 人人妻人人添人人爽欧美一区卜 | 热re99久久精品国产66热6| 亚洲第一av免费看| 久久精品国产亚洲av涩爱| 国产成人精品久久久久久| 久久精品国产自在天天线| 国产视频首页在线观看| 国产片特级美女逼逼视频| 在线观看一区二区三区激情| 91精品一卡2卡3卡4卡| 身体一侧抽搐| 深夜a级毛片| 搡老乐熟女国产| 亚洲av免费高清在线观看| 久久人人爽人人片av| 国产伦精品一区二区三区四那| 日韩av在线免费看完整版不卡| 国产高清国产精品国产三级 | 亚洲av二区三区四区| 国产av精品麻豆| 亚洲精品自拍成人| www.色视频.com| 偷拍熟女少妇极品色| 最近手机中文字幕大全| 乱系列少妇在线播放| 亚洲人成网站在线播| 久久精品国产亚洲av涩爱| 两个人的视频大全免费| 91aial.com中文字幕在线观看| 男女国产视频网站| 日韩免费高清中文字幕av| 免费在线观看成人毛片| 青春草视频在线免费观看| 身体一侧抽搐| 精品国产三级普通话版| 日本av手机在线免费观看| 成年免费大片在线观看| 男人添女人高潮全过程视频| .国产精品久久| 日产精品乱码卡一卡2卡三| 亚洲精品色激情综合| 啦啦啦中文免费视频观看日本| 男人和女人高潮做爰伦理| 中文字幕制服av| 亚洲人成网站在线观看播放| 久久人妻熟女aⅴ| 99热网站在线观看| 亚洲精品久久午夜乱码| 亚洲人与动物交配视频| 观看免费一级毛片| 五月开心婷婷网| 国产乱人视频| 亚洲精品国产色婷婷电影| 免费在线观看成人毛片| a级一级毛片免费在线观看| 成人毛片a级毛片在线播放| 丰满人妻一区二区三区视频av| 久久久久人妻精品一区果冻| 欧美丝袜亚洲另类| 亚洲国产精品国产精品| 国产精品99久久久久久久久| 亚洲精品视频女| 免费av中文字幕在线| 国产精品一二三区在线看| 中国美白少妇内射xxxbb| 国产免费一区二区三区四区乱码| 亚洲国产精品一区三区| 老司机影院成人| 亚洲国产毛片av蜜桃av| 精品国产露脸久久av麻豆| 黄片wwwwww| 噜噜噜噜噜久久久久久91| 国产人妻一区二区三区在| 中文字幕制服av| 精品少妇黑人巨大在线播放| 日韩欧美 国产精品| 搡女人真爽免费视频火全软件| 在线观看国产h片| a 毛片基地| 亚洲第一av免费看| 国产精品不卡视频一区二区| 日韩av不卡免费在线播放| 少妇的逼好多水| 一区二区三区四区激情视频| 国产精品久久久久久久久免| 在线精品无人区一区二区三 | 成年人午夜在线观看视频| 免费人成在线观看视频色| 性高湖久久久久久久久免费观看| 国产亚洲5aaaaa淫片| 国产精品女同一区二区软件| 国产精品99久久99久久久不卡 | 欧美日韩在线观看h| 久久精品久久精品一区二区三区| 成年免费大片在线观看| 日韩成人伦理影院| 亚洲一区二区三区欧美精品| 一级毛片电影观看| 亚洲人与动物交配视频| 国产成人午夜福利电影在线观看| 国产黄色免费在线视频| 精品久久久久久久久av| 国产亚洲精品久久久com| 欧美97在线视频| 成人毛片60女人毛片免费| 亚洲av福利一区| 下体分泌物呈黄色| 中国美白少妇内射xxxbb| 亚洲精品日韩av片在线观看| 亚洲婷婷狠狠爱综合网| 欧美一级a爱片免费观看看| 男女免费视频国产| 久久午夜福利片| 伦精品一区二区三区| 亚洲一级一片aⅴ在线观看| 国产 精品1| 黄色欧美视频在线观看| 久久鲁丝午夜福利片| 亚洲精品亚洲一区二区| 有码 亚洲区| 美女内射精品一级片tv| 国产91av在线免费观看| 女人久久www免费人成看片| 又大又黄又爽视频免费| 在线观看免费高清a一片| 如何舔出高潮| 亚洲国产日韩一区二区| 视频区图区小说| 亚洲欧美日韩卡通动漫| 中文字幕人妻熟人妻熟丝袜美| 制服丝袜香蕉在线| 精品久久国产蜜桃| 在线免费观看不下载黄p国产| 日日摸夜夜添夜夜添av毛片| 国产黄频视频在线观看| 校园人妻丝袜中文字幕| 国产精品伦人一区二区| 国产一区二区三区综合在线观看 | 在线观看美女被高潮喷水网站| 99久久人妻综合| 高清黄色对白视频在线免费看 | 一本久久精品| 日韩欧美一区视频在线观看 | 最新中文字幕久久久久| 五月天丁香电影| 欧美日本视频| 亚洲国产精品专区欧美| 亚洲精品成人av观看孕妇| 99精国产麻豆久久婷婷| 欧美国产精品一级二级三级 | 久久久久久久亚洲中文字幕| 免费观看av网站的网址| 国产亚洲欧美精品永久| 国产亚洲一区二区精品| 看十八女毛片水多多多| a级毛片免费高清观看在线播放| 又黄又爽又刺激的免费视频.| 免费播放大片免费观看视频在线观看| 中文字幕亚洲精品专区| 国产一区二区三区av在线| 51国产日韩欧美| 大话2 男鬼变身卡| 国语对白做爰xxxⅹ性视频网站| 大香蕉久久网| 啦啦啦视频在线资源免费观看| 久久久欧美国产精品| 亚洲国产欧美在线一区| 亚洲第一av免费看| 777米奇影视久久| 中文字幕人妻熟人妻熟丝袜美| 久久精品国产自在天天线| 日本黄色片子视频| 自拍偷自拍亚洲精品老妇| 狂野欧美白嫩少妇大欣赏| 在线看a的网站| 成人无遮挡网站| 国产精品.久久久| 亚洲精品久久午夜乱码| 亚洲欧美日韩东京热| 偷拍熟女少妇极品色| 人人妻人人看人人澡| 中文精品一卡2卡3卡4更新| 精品人妻一区二区三区麻豆| 国产精品一及| 国产高清三级在线| 各种免费的搞黄视频| 极品少妇高潮喷水抽搐| 日本vs欧美在线观看视频 | h日本视频在线播放| 狂野欧美激情性xxxx在线观看| 久久这里有精品视频免费| 日韩三级伦理在线观看| 国产色婷婷99| 亚洲欧洲日产国产| 国产午夜精品一二区理论片| 一级a做视频免费观看| 亚洲人成网站高清观看| 国产精品无大码| 一区二区三区乱码不卡18| 成人免费观看视频高清| 午夜免费鲁丝| 国产黄色免费在线视频| 国产乱来视频区| 男人舔奶头视频| 亚洲欧美一区二区三区国产| 国产黄色免费在线视频| 性色av一级| 蜜臀久久99精品久久宅男| 少妇 在线观看| 男女边摸边吃奶| 亚洲人与动物交配视频| 蜜臀久久99精品久久宅男| 一级黄片播放器| 日韩一本色道免费dvd| 熟女人妻精品中文字幕| 男女啪啪激烈高潮av片| 日韩一区二区三区影片| 深夜a级毛片| 久久精品久久久久久久性| 欧美精品人与动牲交sv欧美| 午夜激情福利司机影院| 日本与韩国留学比较| 韩国av在线不卡| 国产极品天堂在线| 国产精品伦人一区二区| 视频区图区小说| 熟妇人妻不卡中文字幕| 国内揄拍国产精品人妻在线| 97在线人人人人妻| av免费在线看不卡| 精品亚洲成a人片在线观看 | 熟女av电影| 中国国产av一级| 日韩成人伦理影院| 啦啦啦中文免费视频观看日本| 亚洲欧美日韩东京热| 国精品久久久久久国模美| 免费看光身美女| 一区二区三区四区激情视频| 亚洲经典国产精华液单| 春色校园在线视频观看| 边亲边吃奶的免费视频| 51国产日韩欧美| 午夜福利在线在线| 嘟嘟电影网在线观看| 深夜a级毛片| 女的被弄到高潮叫床怎么办| 草草在线视频免费看| 精品亚洲乱码少妇综合久久| 亚洲自偷自拍三级| 精品视频人人做人人爽| 免费av中文字幕在线| 国产又色又爽无遮挡免| 国产亚洲最大av| 免费看不卡的av| 国产视频首页在线观看| 99re6热这里在线精品视频| 国产黄片视频在线免费观看| 欧美最新免费一区二区三区| 夜夜骑夜夜射夜夜干| 日本黄大片高清| 国产人妻一区二区三区在| 日韩av不卡免费在线播放| 久久久久性生活片| 成人黄色视频免费在线看| 国产精品久久久久久精品古装| 美女主播在线视频| 伊人久久国产一区二区| 亚州av有码| 亚洲丝袜综合中文字幕| 99九九线精品视频在线观看视频| 国产在线视频一区二区| 97热精品久久久久久| 成年美女黄网站色视频大全免费 | 能在线免费看毛片的网站| 久久国产精品大桥未久av | 欧美97在线视频| 欧美另类一区| 一个人看视频在线观看www免费| av国产精品久久久久影院| 国产黄片美女视频| 插阴视频在线观看视频| 国产精品一区二区性色av| 日韩 亚洲 欧美在线| 国产精品偷伦视频观看了| 18禁动态无遮挡网站| 日本wwww免费看| 欧美成人午夜免费资源| 我的女老师完整版在线观看| 日韩一区二区视频免费看| 精品一区在线观看国产| 在线精品无人区一区二区三 | 国产女主播在线喷水免费视频网站| 水蜜桃什么品种好| 深夜a级毛片| 久久精品国产亚洲av天美| 国产精品国产三级国产av玫瑰| 精品视频人人做人人爽| 欧美一区二区亚洲| 黄色日韩在线| 麻豆成人午夜福利视频| 国产精品成人在线| 国产在线一区二区三区精| 日韩中文字幕视频在线看片 | 亚洲aⅴ乱码一区二区在线播放| 尾随美女入室| 日韩强制内射视频| 成人美女网站在线观看视频| 久久久久久伊人网av| 又爽又黄a免费视频| 少妇人妻久久综合中文| 狠狠精品人妻久久久久久综合| 22中文网久久字幕| 免费观看的影片在线观看| 女人久久www免费人成看片| 天美传媒精品一区二区| 亚洲人成网站在线观看播放| 大码成人一级视频| 高清在线视频一区二区三区| 亚洲成人中文字幕在线播放| 青春草国产在线视频| 日韩av不卡免费在线播放| 韩国高清视频一区二区三区| 久久久亚洲精品成人影院| 大香蕉97超碰在线| 色网站视频免费| 免费观看性生交大片5| 国产欧美日韩精品一区二区| av又黄又爽大尺度在线免费看| 久久久久久人妻| 五月伊人婷婷丁香| 国产精品国产三级国产专区5o| 国产免费福利视频在线观看| 在现免费观看毛片| 两个人的视频大全免费| 国产精品嫩草影院av在线观看| 在线观看av片永久免费下载| 蜜桃亚洲精品一区二区三区| 爱豆传媒免费全集在线观看| 国产伦理片在线播放av一区| 男男h啪啪无遮挡| 在线看a的网站| 国产亚洲一区二区精品| 99热网站在线观看| 久久久欧美国产精品| 高清不卡的av网站| 成人毛片60女人毛片免费| 啦啦啦啦在线视频资源| 一级毛片 在线播放| 欧美变态另类bdsm刘玥| 免费观看无遮挡的男女| 高清黄色对白视频在线免费看 | 18+在线观看网站| 一二三四中文在线观看免费高清| 国产熟女欧美一区二区| 久久精品国产亚洲网站| 99久久精品热视频| 小蜜桃在线观看免费完整版高清| 男男h啪啪无遮挡| 国产高潮美女av| 亚洲精品456在线播放app| 我的女老师完整版在线观看| 欧美老熟妇乱子伦牲交| 性色avwww在线观看| 日韩人妻高清精品专区| 欧美高清性xxxxhd video| 免费久久久久久久精品成人欧美视频 | 久久久久久久亚洲中文字幕| 欧美性感艳星| 夜夜看夜夜爽夜夜摸| a级毛色黄片| 美女国产视频在线观看| 亚洲欧美成人精品一区二区| 午夜老司机福利剧场| freevideosex欧美| 高清在线视频一区二区三区| 噜噜噜噜噜久久久久久91| .国产精品久久| 51国产日韩欧美| 亚洲美女视频黄频| 久久毛片免费看一区二区三区| 国产成人a∨麻豆精品| 午夜福利高清视频| 久久久久久人妻| 国产亚洲午夜精品一区二区久久| 国产在线免费精品| 乱系列少妇在线播放| 又大又黄又爽视频免费| 99久久精品一区二区三区| 亚洲aⅴ乱码一区二区在线播放| 亚洲精品aⅴ在线观看| 中文字幕亚洲精品专区| 麻豆国产97在线/欧美| 晚上一个人看的免费电影| 国产精品人妻久久久影院| 国产探花极品一区二区| 色婷婷久久久亚洲欧美| 观看av在线不卡| 精品一区二区三卡| 亚洲欧美日韩卡通动漫| 22中文网久久字幕| 99久久人妻综合| 夫妻午夜视频| 国产白丝娇喘喷水9色精品| 在线免费十八禁| 在线观看一区二区三区| h视频一区二区三区| 91精品伊人久久大香线蕉| 精品人妻偷拍中文字幕| 亚洲精品一区蜜桃| 日韩强制内射视频| 精品国产一区二区三区久久久樱花 | 国产欧美日韩精品一区二区| 少妇猛男粗大的猛烈进出视频| 久久热精品热| 亚洲精品一区蜜桃| 国产精品精品国产色婷婷| av.在线天堂| 国产精品久久久久久久久免| 精品久久久久久久久av| .国产精品久久| 亚洲人与动物交配视频| 99热网站在线观看| 久久久久久久久久成人| 精品一区二区三区视频在线| 亚洲av成人精品一二三区| 80岁老熟妇乱子伦牲交| 最后的刺客免费高清国语| 亚洲美女黄色视频免费看| 日韩视频在线欧美| 美女福利国产在线 | 在现免费观看毛片| 少妇被粗大猛烈的视频| 久久毛片免费看一区二区三区| 一级毛片我不卡| 国产精品偷伦视频观看了| 久久国产精品男人的天堂亚洲 | 三级国产精品欧美在线观看| 精品99又大又爽又粗少妇毛片| 高清av免费在线| 成人亚洲精品一区在线观看 | 国产精品一区二区性色av| 狠狠精品人妻久久久久久综合| 成人特级av手机在线观看| 精品国产露脸久久av麻豆| 91精品国产九色| 欧美少妇被猛烈插入视频| 国产有黄有色有爽视频| 特大巨黑吊av在线直播| av一本久久久久| 欧美三级亚洲精品| 亚洲精品国产av蜜桃| 成人影院久久| 三级国产精品欧美在线观看| 亚洲精品乱码久久久v下载方式| 成人18禁高潮啪啪吃奶动态图 | 亚洲三级黄色毛片| 99久久综合免费| 欧美精品一区二区免费开放| 精品视频人人做人人爽| 欧美xxxx黑人xx丫x性爽| 国产午夜精品久久久久久一区二区三区| 亚洲三级黄色毛片| 好男人视频免费观看在线| 欧美精品一区二区大全| 国产一级毛片在线| 亚州av有码| 欧美97在线视频| 国产伦理片在线播放av一区| 国产男人的电影天堂91|