Intelligent Systems and Networks

Image thresholding is a simple but effective image segmentation technique that is widely used in medical image analysis to detect tumors. The input of the method consists of a grayscale enhancement image and a threshold. In conventional approaches, the image enhancement is computed by the CLAHE method, and the threshold is computed by the Otsu method. The purpose of this article is to improve the quality of an MRI brain tumor image by using the Bidimensional Empirical Mode Detection (BEMD) algorithm and morphological operations to create an enhanced image for segmentation without blurring the edges of objects, as well as a threshold derived from the Neighborhood Valley-Emphasis (NVE) method that is appropriate for the brain tumor image because the object is smaller than the background. Analysis of the computational results of the proposed method on five MRI brain tumors from the Figshare database shows that: (i) convolution of the first trend function of BEMD and the combination of the top-hat and bottom-hat transforms adequately improve the image quality for segmentation; the metric used to measure image quality shows that MSE = 0.009, PSNR = 68.6731, EME = 2.8339, and EMEE = 0.1957; (ii) segmentation using the proposed algorithm's image enhancement and the NVE method's threshold for an accuracy of 99.61% and a precision of 96.1%.

This paper proposes a new fuzzy nonsingular fast terminal sliding mode control (FNFTSMC) for pendubot with uncertainties. In the suggested method, the nonsingular fast terminal sliding mode controller (NFTSMC) is designed to get wonderful features namely speedy response, finite time converge, and singularity avoidance. Fuzzy logic control is utilized to optimize NFTSMC for rejecting chattering phenomenon and reducing errors. Moreover, the genetic algorithm is adopted to determine unknown constants of NFTSMC. The global stability of the system is guaranteed by using the Lyapunov synthesis method. Finally, some simulations are conducted to prove the efficiency of the FNFTSMC compared with NFTSMC.

This paper introduces a leak–localization method for fluid pipeline through monitoring bursts in acoustic emission signals. First, the algorithm seeks bursts in individual signals using the Neyman–Pearson signal detection theorem, groups adjacent bursts from two signals captured by two acoustic emission sensors mounted at two pipeline ends and localizes those burst sources through the time difference of arrival technique. Then, a coordinate histogram is constructed from a set of resulting burst source coordinates in a period to indicate a suspicious leaky position on the testing pipeline. The experimental results reveal that the proposed technique can determine single leaks with a mean relative location error of 1.2% while conventional methods return relative errors of greater than 7.5%. Furthermore, the proposed method can properly localize two simultaneous leaks whereas others cannot.

At the intersection of green and red lights, people often honk their horns to urge people. Besides, noise pollution in urban areas due to the abuse of car horns leads to high sound intensity that affects the health of people. Although many solutions have been given, this problem has not been completely solved. Therefore, we present a solution to measure the sound intensity at the intersection of red and green lights in this paper. When the sound intensity is too great (the threshold), the waiting time for the red light will increase. It will directly affect the perception of people and thus reduce the noise. The results show that the system has recovered noise with an accuracy of up to 96%.

This study proposes an ingredient-based recipe recommendation to make cooking easier and more convenient. Given the ingredients, the system can recommend classic recipes from the database or generate new recipes using the ViT5 model. We fine-tune ViT5 and train the model on our CookyVN-recipe dataset, consisting of 26,752 recipes in Vietnamese. The ROUGE-1, ROUGE-2, and ROUGE-L scores of our model are 64.45, 35.92, and 38.21, respectively.

The diagnostic imaging industry in many countries is currently severely under-resourced. The reason is that people know about it is not too much. Besides, the demand for radiologists has increased because of the impact of the Covid-19 epidemic. There are situations that the general practitioner or the clinician must diagnose x-rays. Therefore, the article proposes a solution to save time by assisting doctors in diagnosing problems by giving suggestions on abnormal areas and creating an AI application to recognize images that can be detected abnormal areas or not. The results show that although the accuracy of the proposed solution only achieves about 30% due to the small number of samples, it promises to be highly feasible when applied in practice.

This paper proposes the Hybrid Optical Fiber (OF)-Free Space Optics (FSO) based Inter-Optical Network Unit (ONU) transmission capable Wavelength Division Multiplexing (WDM)-Passive Optical Network (PON) architecture. This architecture can transmit up to 40 Gbps for downstream, upstream, and inter-ONU transmission. The downstream transmission is flowing from the center office (CO) to respective ONUs while Inter-ONU transmission is performed from one ONU to another ONU which are connected in a ring manner. The upstream transmission is performed from a particular ONU and travels towards the CO via a circular ring. The upstream signal can be transmitted on the same wavelengths which are used for the downstream transmission and therefore it utilizes the available bandwidth efficiently. In the proposed architecture, four 10 Gbps optical line terminals (OLT) transmit the signal through 20 km of single-mode fiber (SMF). Inter-ONU feature increases the reliability and survivability of the optical network. The bit error rate (BER) performance of the proposed architecture for the main and Inter-ONU transmission is reasonably good for the combination of 20 km feeder fiber (FF) and 200 m distributed fiber (DF)/FSO link. The architecture also provides better receiver sensitivity for a hybrid OF-FSO optical network with Inter-ONU transmission.

Blockchain is one of the key technologies today and is increasingly being applied in many different fields. Blockchain in the enterprise is emphasized due to its ability to overcome the inherent limitations of centralized applications widely used in enterprises. The modern workplace trend after the COVID-19 pandemic has paved the way for the deployment of enterprise blockchain networks on end devices (nodes) provided by enterprises to employees. However, according to the above approach, ensuring the availability of nodes is very difficult because it heavily depends on employees’ behaviors. To increase enterprise blockchain availability, we propose a novel approach that integrates a real-time supervisor with enterprise blockchain to monitor and ensure established policies.

In this paper, a proportional integral derivative (PID) controller is developed to apply to the dynamics model of the two-wheeled self-balancing (TWSB) robot. The proposed control strategy is created using a PID controller in conjunction with a genetic algorithm, PID-GA for short. The parameters are selected based on the genetic algorithm method to help the system achieve the fastest stability. Therefore, the controller has a simple structure and is easy to implement in practice. Based on the simulation result, it can be seen clearly that the TWSB robot control system’s output values closely match the target values and that the error is roughly zero, which demonstrates the proposed controller’s effectiveness and performance.

In this paper, we propose an automated deep learning Convolutional Neural Network (CNN) based detection method to distinguish cancerous tumors from benign tumors to identify pancreatic cancer in Computed Tomography (CT) images. We train and test the model using two datasets provided by The Cancer Imaging Archive (TCIA) and the Medical Segmentation Decathlon (MSD). The first dataset contains 18942 CT images of 82 patients with normal pancreas while the second dataset contains 15000 images of 280 patients with confirmed pancreatic cancer. Our proposed model is a modified extension of the current CNN network with the addition of the Densely Connected Convolutional layers. Our method has shown superior results compared to other research and deep learning method with an accuracy of 97.4%, sensitivity of 98.3%, and specificity of 96.6%.

This paper presents a method of synthesizing a robust adaptive control system for the industrial robot with six degrees of freedom. We describe the mathematical model of the robot in the form of nonlinear state equations that consider variable parameters and unmeasurable external disturbance. Uncertainty components in the dynamic robot model are identified based on adaptive control and the RBF neural network. We create a signal vector from the identification results to compensate for these components. The control law is built based on sliding mode control in which the chattering phenomenon is reduced to a minimum because the influence of the uncertainty components has been eliminated. The article’s proposed control system is adaptable, robust, anti-interference, and ensures good control quality.

Nowadays, vestibular is a common disease in Vietnam. The diagnosis of vestibular disorders is made using a variety of methods. In the content of the article, we research and detect vestibular disorders by testing balance. We use a data set to quantitatively measure the patient’s body angle by using a camera and computer. Then, data analysis and logistic regression model were built to classify patients with vestibular disorders. This makes sense in the process of testing and diagnosing diseases more accurately and efficiently.

This paper investigates simultaneous wireless information and power transfer (SWIPT) in a multiple-input single-output (MISO) system for the secure beamforming design. Under knowledge of statistical channel state information to eavesdroppers and energy receivers, the considered problem aims to minimize the transmit power under the probabilistic constraints of outage secrecy rate and harvested energy. The problem is intractable to solve due to its nonconvexity nature and the statistical channel errors. Exploiting the special structure of these problems, we first transform the probabilistic ones into tractable form by the Bernstein inequalities and the non-convex constraints are transformed as DC (Difference of Convex functions) ones. The general DCA (DC Algorithm) based algorithms are developed to solve the proposed DC programs. Numerical results illustrate the efficiency and robustness of our proposed algorithms.

The application of blockchain to centralized systems is one of the trends of improving the privacy, security, and transparency of the centralized approach. Blockchain is especially meaningful in trustless systems, where stakeholders are susceptible to tampering and consensus is required. In this work, we propose a practical framework for academic records management so that stakeholders can exploit records effectively. Based on the framework, we also conduct feasibility testing. The experiment is conducted at the computer lab I1.501 in the Thu Dau Mot University (TDMU). The results show that it is possible to apply blockchain to the academic records management system. The source code is available in Github with an open source license to make sure that institute can freely use this framework or modify to make it more suitable ( https://github.com/vanhuudhsp/AcademicRecordsBlockChain ).

This article nominates a new manner based upon the Neuman volume scattering model to get better the correctness of forest elevation calculation using polarized interference UAV-SAR images. The forest parameters are extracted by the suggested solution which are carried out through 3 steps. First, the contribution of the scattering components is determined based on the Neuman volume scattering model. Next, constraint conditions are added to determine the optimal complex polarimetry interferometry coherence (CPI) coefficient for ground and volume scattering components. Then the sum of least squares algorithm is used to calculate the ground phase. Finally, the forest parameters is directly restored based on the optimal iterative method. The efficiency of the newly mentioned manner is assessed with UAV-SAR data received from the AfiSAR project of NASA/JPL.

Air target threat assessment is an important issue in air defense operations, which is an uncertainty process to protect the valuable assets against potential attacks of the various hostile airborne objects such as aircraft, missiles, helicopters and drones/UAV. This paper proposes a method to solve the problem of threat assessment of air targets by presenting the process of air defense scenarios in the form of Markov decision process and using reinforcement learning with Deep Q-Learning to predict most dangerous enemy actions to provide more accurate threat assessment of air attacks. On the basis of information about typical air defense combat environment, parameters of binding target trajectory (speed limit, overload limit…) and capabilities of defensive units (number of target channels, fire zone limitation, burning time,…) a simulation environment is built to train and evaluate the optimal (most dangerous) trajectory model of the target based on the given environment. This optimal trajectory can provide input information that is closer to reality, such as real time of arrival; probability of aircraft being shot down by SAM; angle of attack… for threat assessment methods (fuzzy logic, Bayes network, neural network…). The proposed model has been tested on the OpenAI Gym tool using Python programming language. It was shown that the model is suitable to calculate the level of danger of the target with the object to be protected in the context of the general air attacking environment with dynamic and complex constrains.

Vessel detection, classification, and tracking are very important problems in maritime surveillance systems. In recent years, the field of computer vision has significantly developed, which allows its application to these systems. Accordingly, in this paper, a method based on a YOLOv5-based deep neural network combined with the Strong Simple Online Real-time Object Tracking (StrongSORT) algorithm is proposed for vessel detection, classification, and tracking. Specifically, the YOLOv5 model is trained by using a dataset of diverse images, which are collected from various public sources. The dataset contains several popular vessel types for the purpose of classification. Experimental results show that the proposed model gives high accuracy of vessel classification and high-speed tracking of approximately 16 frames per second, which is near real-time. The model has been embedded into a real small demonstrator to verify the potential implementation in maritime surveillance systems.

To reduce the DC microgrid (DCMG) system significantly, this paper proposes a power management for distributed DC microgrid system under minimum communication links. Only seven communication links are enough to achieve the power balance and voltage regulation in case of both the grid-connected and islanded modes for distributed DCMG system that consists of 5 agents. Several simulations have been implemented to verify the effectiveness of the proposed scheme under various conditions.

This paper proposes a method to diagnose atopic dermatitis based on deep learning network by analyzing image data of infected skin areas. We use deep learning network to analyze the layers and get the suitable layer for diseased and non-diseased images. These images are further feature extracted through HOG feature extraction and SVM classifier to classify disease and non-disease. The proposed method proves to be effective in precision and recall, that can be considered as an adjunct to traditional diagnostic methods, and the results obtained are equivalent to that of a diagnostician while limiting the heterogeneity between the predictors.

The weld bead geometry is the important information for determining the quality and mechanical properties of the weldment. The welding process parameters or variables that affect the weld bead geometry in the conventional arc welding process include the following: the welding voltage U, the welding current I, the wire feed speed WFS, the contact tip to work distance D, and the welding speed S. Modeling and predicting the weld bead geometry play an important role in welding process planning, to determine the optimal welding process parameters for achieving the improved weld quality. There have been lots of efforts and studies to develop modeling solutions and simulations to determine the weld bead geometry (Height H and Width W) from the welding process parameters (U, I, WFS, D, S) as the inputs. The welding process parameters can be determined based on the experiences, and the conventional analysis of variance (ANOVA); however, the high welding quality and accuracy are not always obtained. With the advancement of computer vision technologies, digital images from cameras and videos can be used for training the deep learning models, to accurately identify and classify objects. The digital images for evaluating the welding quality and the characteristics of welding objects can be captured via the use of the high-speed camera, and there are emerging data acquisition systems that can handle a huge dataset. In this paper, an adaptive neuro-fuzzy inference system (ANFIS) model is proposed to determine weld bead geometry from the main welding process parameters U, I and S. The proposed ANFIS model was successfully developed for the first basic investigations, as the foundation for further developments of innovative robotic welding systems which can be used for higher educations or research in Smart Manufacturing, with potentials for industrial applications.

This paper proposes a novel approach for balancing the neutral-point voltage (NPV) of the three-level sparse four-leg voltage source inverter (VSI) using a simplified model predictive current control (MPCC) method. By using a new cost function, the proposed method finds the best voltage vector without the need for current prediction calculations. The best switching state from the best voltage vector will be selected based on the deviation of two capacitor voltages (DTCV) to balance the NPV, without the use of weighting factor and capacitor voltage calculations. All results from simulation show that the proposed method significantly reduces computation time and eliminates the time-consuming weighting factor adjustment required by the conventional MPCC method.

Spiking neurons and spiking neuron networks (SNN) have recently been considered the third generation of neuron models, replacing the perceptron neuron, which has been the most popular model. Spiking neurons can be emulated by both analog and digital implementations. Because of the complexity of mathematical models, integrate-and-fire (IAF) and leaky integrate-and-fire (LIF) neurons, which are simpler, are the common models for digital and VLSI implementations. However, these models lack the dynamic properties of the Hodgkin-Huxley model. In order to overcome these issues, the resonate-and-fire (RAF) model, which has more dynamic properties, has been proposed by Izhikevich. This paper presents the first all-digital design of the RAF model. The design does not require floating-point operations and multipliers. Its RTL structure can be realized on a VLSI system. The design uses 54 ALM units on a DE10-nano board based on a Cyclone V SoC FPGA. It has been verified at 20 kHz of subthreshold oscillation.

The wheel slip controller (WSC) serves as the cornerstone of the anti-lock braking system (ABS). It is necessary to investigate and test a nonlinear robust WSC since the friction between the road and tire is a nonlinear function of wheel slip. In this research, a backstepping sliding mode controller (BSMC) is designed to control the wheel slip of a quarter-vehicle model. Extended state observer (ESO) is used in combination with the design of the BSMC in order to estimate the total uncertainty in the system. The viability of the proposed controllers is then verified in numerical simulation, and the performance of the proposed controllers is evaluated using three different key performance indicators (KPIs).

Among various technologies being applied for indoor localization, WiFi has become a common source of information to determine the pedestrian’s position due to the widespread of WiFi access points in indoor environments. However, the fluctuation of the WiFi signals makes it difficult to achieve a good localization result. In this paper, to handle this problem, the valued tolerance rough set and decision rules method (VTRS-DR), which is firstly registered to WiFi fingerprinting-based localization, will be implemented and evaluated in big and complicated environments using two public datasets. The first one was conducted by a subject using a smartphone at a multi-floor library for several months. Furthermore, to evaluate the localization accuracy when WiFi data was collected from different pedestrians as well as different smartphones, a crowdsourced WiFi fingerprinting dataset was utilized. From the deep analyses of localization results, the VTRS-DR method shows high accuracy and high robustness when testing in different environments with a mean error of 3.84 m, which is 27.87% lower than other compared methods.

In this work, we focus on controlling the 3-D overhead crane (3-DOC), the main problem when controlling the 3-DOC model is the problem of tracking the trajectory and reducing the load vibration. Therefore, the limitation of the system’s state variables must be made explicit. To solve this problem, firstly, the problem of orbital tracking is solved using the Sliding Mode Control (SMC), but the system’s states are not strictly controlled, so we propose a control method Lyapunov-based model predictive control (LMPC), which allows setting limits for state variables besides that, an auxiliary component is added based on the stability of the SMC to make the system globally stable. Finally, simulations are added to show the feasibility of the method for tracking, and anti-vibration for the payload.

This paper proposes a solution for the AGV (Autonomous Guided Vehicles) robot to effectively monitor a moving object using deep learning by enabling the robot to learn and recognize movement patterns. Using a model of a four-wheeled self-propelled robot vehicle, a highly adaptable and modifiable platform AGV was built. A customized development of the TensorFlowLite ESP32 module from the TensorFlow CoCo SSD model enables the ESP32-CAM camera module on the robot to self-identify objects and autonomously follow the human object in front. Using its built-in distance tracking algorithm, the robot can also detect and adjust its speed to safely follow the individual in front at an appropriate distance. It can function autonomously or manually via the local network. Even with a minimal configuration, the algorithm is appropriate for the automaton. The experimental findings demonstrate the method's precision and effectiveness including sensors, algorithms, and mapping technologies that enable the robot to identify obstacles and navigate around the AGV.

Supply chain gradually becomes a core factor to operate and develop for businesses. Using machine learning, especially with neural networks, to assess the risk in supply chain network has been attracted many research and become potential approaches. Via machine learning particular to Bayesian neural network, risk evaluation in supply chain network can be performed effectively to support supply chain partners to assess, identify, monitor, and mitigate risks. In detail, by using reliability theory, supply chain network’s risk is divided in alternative scales (from Very high risk to Very low risk). The Bayesian neural network allows to treat the weights and outputs as the variables in order to find their marginal distributions that best fit the data. By taking the advantage of Bayesian neural network in deep learning, the experiment in this paper shows a very high accuracy rate in supply chain risk prediction. This implicates the performance of using machine learning in supporting of managerial decision making in selecting suppliers.

In an era when digitization is on the rise, Vietnamese businesses are rapidly adopting technology to centralize business processes, support in system optimization, lessen time to solve problems, and scale up operations. The referred technology, or a software solution to be more specific, is called ERP (Enterprise resource planning). Among the top international ERP solution providers, Odoo has consistently become the standard software suite for many companies in Vietnam owing to its great adaptability, optimal interface, and inexpensive implementation and recurring costs. ERPViet is a prestigious Odoo solution supplier in Vietnam, not only do they provide implementation support, but they also develop several bespoke features like Overtime Management to meet the special requirements of Vietnamese firms. The goal of this paper is to decode how a new app such as Overtime Management is created by looking at the architecture, module structure, and the method to construct and install a new customized module on Odoo. The detailed procedure can be found in our Github repository.

Optimal pressure control to water leakage reduction can be accomplished by controlling pressure reducing valves (PRVs) installed in water distribution systems (WDSs). The optimal pressure control can be casted into a nonlinear program (NLP) where the model of PRVs is of important for proper operation of the systems. In this paper, at first, we reformulated the mathematical model of PRVs by using vanishing constraints which is suitable for the use in practice, then we applied an efficient relaxation approach for solving the mathematical program with vanishing constraints (MPVCs). The proposed relaxation approach has strong convergence. The application of MPVCs for optimal pressure management has been evaluated on one WDS in Vietnam showing that the MPVCs results in highly accurate solutions and it is suitable for the use in the decision support system for optimal pressure control.

The agriculture industry has undergone considerable changes in recent decades due to technological improvements and the introduction of new technologies such as the Internet of Things (IoT), artificial intelligence (AI), and secure communication protocols. Even in harsh weather, it is now possible to grow plants. Knowledge transfer continues to be a significant barrier to the adoption of AI technologies, specifically in the agriculture sector. The concept of “federated learning” (FL) was developed to safeguard data from various users or devices. FL is adequate for protecting data, but there are still several issues to be concerned about, including single-point failure, model inversion attacks, and data and model poisoning. For this study, a blockchain-enabled, FL-integrated framework is suggested as a solution to these problems. Additionally, the experimental setup and the Ethereum network are used to validate the suggested framework and the blockchain transactions.

Federated learning (FL) is a new artificial intelligence concept that enables Internet-of-Things (IoT) devices to learn a collaborative model without sending the raw data to centralized nodes for processing. Despite numerous advantages, Federated Learning faces huge challenges from model overfitting due to the lack of data and statistical diversity among clients. To address these challenges, we propose a novel personalized federated learning method via momentum adaptation, the so-called pFLTI. Specifically, pFLTI generates the target model by adapting from the temporal ensemble of the meta-learner, i.e., the momentum network. This momentum network and its task-specific adaptations enjoy a favorable generalization performance, enabling self-improving of the meta-learner through knowlodels to find the across task relationsedge distillation. Moreover, we found that perturbing parameters of the meta-learner, e.g., dropout, further stabilize this self-improving process by preventing fast convergence of the distillation loss during meta-training. Our experimental results demonstrate that our algorithm is awesome.

This paper explores an advanced adaptive controller for Pneumatic Artificial Muscles (PAMs). PAMs offer lightweight, simple, and safe operation advantages but are difficult to model and control due to the non-linearity and hysteresis caused by their physical manufacturing. An adaptive controller based on a neural approximation is proposed to address these challenges, incorporating Radial Basis Function algorithms and adapting to model parameter uncertainty. The efficiency of the control approach is confirmed through a multi-scenario experiment with high-accuracy results, promising future developments in intelligent control of PAMs.

The presence of the Internet of Things (IoT) works with the assortment and diffusion of metropolitan information data. In any case, it can release clients’ very own security data in brilliant urban areas. Many works have been done and different security systems identifying with the precautions of distributed computing have been implemented from copious points of view. Be that as it may, they don’t propose a quantitative way to deal with dissecting and assessing protection and security in distributed computing frameworks. Accordingly, we propose another private data encryption strategy in IoT under distributed computing climate. Below IoT, as per the properties and securing time, protection data can be isolated into numerous subspaces. Because of the stream figure instrument, we plan an encryption framework model of data assortment. In the subspace, the security data is encoded and moved to the transfer hub. Subsequent to encoding, they are divided and rebuilt. At long last, we use stream figure and dual-key calculation to finish opportunity nondestructive change among plaintext and ciphertext to guarantee the trustworthiness of the encoded private data. Test outcomes illustrate that the presented strategy requires some investment in the encryption and unscrambling measure, which has superior ciphertext transformation yield impact and experiences fewer organization assaults in a similar encryption time. As far as computation cost, the proposed strategy diminishes by roughly 11%. Furthermore, it has greater security and works on the security and integrity of the protection data assortment procedure.

This paper presents the development of an autonomous mobile robot (AMR) system for hospital logistics in quarantine zones. The designed system includes a remote control station, five autonomous mobile robots Vibot-2, a wireless communication network and an observation system for reducing workload and avoiding the risk of infection for frontline health workers in quarantine zones. The overall architecture and hardware system are designed to meet all design requirements. The control software architecture of the robotic system is constructed in the robot operating system (ROS) framework. The navigation framework is developed based on the ROS Navigation Stack to move the robot safely and avoid the humans in its surrounding working area. In addition, the perception system is integrated a deep learning model for human detection and use the AprilTags to help the robots recognize and navigate to cart and charging stations. Finally, the low-cost solution based on the fusion of localization using AprilTags on the ceiling and the AMCL algorithm is developed for improving robustness and precision. The autonomous mobile robot system has been successfully built and deployed at COVID-19 field hospitals in Ha Nam, Bac Giang and Ho Chi Minh City in Vietnam in the year 2021.

Series Elastic Actuator (SEA) has attracted much attentions in recent years because of its high-performance torque control. However, the flexibility of SEA makes it difficult to position control. In this paper, we propose sliding mode-based position control for a single-joint SEA robot system. Three reaching laws including constant rate, exponential, and power rate reaching laws are considered. The simulations show the effectiveness of the considered control algorithms and a comparison has been done among them.

In this work, a high-speed and accurate face recognition system based on Field Programmable Gate Arrays (FPGAs) was completely developed. A complete pipeline that contains a sequence of processing steps, including preprocessing, face feature extraction, and matching, is proposed. For processing steps, lightweight deep neural models were developed and optimized so that they could be computationally accelerated on an FPGA. Besides the core processing pipeline, a database as well as a user application server were also developed to fully meet the requirements of readily commercialized applications. The experimental evaluation results show that our system has a very high accuracy based on the BLUFR benchmark, with a precision of 99.336%. Also, the system is very computationally efficient, as the computing time to recognize an ID in a dataset of 1000 IDs with 4000 images on the FPGA ZCU-104 is only 30.3 ms. For the critical case, the system can process 8 camera streams and simultaneously recognize a maximum of 80 IDs within a computing time of 342.3 ms for each ID. With its high-speed and accuracy characteristics, the developed system has a high potential for practical applications.

This study proposes a Lyapunov-based Model predictive control with an estimate disturbance radial neural network technique for an extremely complex system-a dual-arm robot. By using a radial neural network to estimate the disturbances, their effects on the system are reduced. This technique also ensures the cooperation between the two arms of the robot and guarantees stabilization through the Lyapunov inequality constraint. Simulations were performed to investigate controller qualities.

Surface plasmon resonance (SPR) is a vital, fast, and robust approach to exploring the real-time detection of molecular interactions. In recent decades, many innovative ways suggested making SPR-based sensors more sensitive and reliable. In order to make a Kretschmann-based SPR sensor more sensitive, many different versions of the sensor were investigated. We proposed a multi-layer structured (Glass-SF11/Ag-Au/ITO) SPR sensor, which is mathematically modelled and simulated via MATLAB environment. A numerical analysis was carried out to optimize the design parameters and characteristics of the proposed design. The results show that the sensor’s sensitivity (for RI ranges of 1.33–1.37) is affected by various design parameters, including the choice of metal, the metal layer’s thickness, and the supportive layer’s thickness. The highest sensitivity of 183°/RIU is achieved for the proposed sensor, which is significantly higher than the conventional SPR sensors. These findings suggest a novel method to design sensitive plasmonic refractive index-based SPR sensors for analyte detection.

This article studies formation control consideration of cooperative path following problem in a group of multiple Surface Vehicles (SVs). A proposed formation control protocol that contains optimal control problem in two subsystems of each SV with Model Predictive Control (MPC) and Approximate/Adaptive reinforcement Learning (ARL) Controller. The MPC is developed for nonlinear sub-system of SV with the tracking performance to be guaranteed by considering an appropriate optimization problem. Moreover, RL control design is carried out for time-varying sub-system by indirect method. Finally, the proposed control protocol is demonstrated by simulation result to show the effectiveness of this control protocol.

A new model for the multi-task of object detection and localization is developed in this paper. This model is used to control a robotic pick-and-place system for real-time applications. The model is developed from a deep neural network, namely RetinaFace with MobileNet as its backbone, with modifications to the output to allow simultaneous detection and localization of objects. A self-generated dataset was prepared for the training and testing processes. The error calibration of the camera is implemented. The robot's overall control algorithm is created. The results show that our model has a high accuracy of 97.4% for object detection and an error of less than 2.41 mm for object localization. The computation is also efficient, reaching 25 FPS with the very lightweight hardware of the Jetson Nano. Experiments with the pick-and-place method have a success rate of 100%.

A ball-balancing robot is a robot that can move and balance on a ball. This research comes from system modeling, finding the binding equations between state variables of the system through a differential equation. In order to construct the state-space model, Lagrange equations are used for each of the three planes. The system is linearized around the equilibrium position, to approximate the system to a linear system. This controller aims to minimize the influence of exogenous output, and in this paper, this control method is approached by linear matrix inequalities. Besides, the LQR controller is also presented to compare with the controller base on $$\mathbf {H_\infty }$$ method. For the xy plane, it does not affect the balance of the system, thus, the controller in this plane is not presented.

Wireless Underground Sensor Network (WUSN) is well known to monitor the soil quality for precision agriculture application. However, the uncertain property of the soil leads to many difficulties in designing a WUSN, one of them being the pathloss determination in different channels. In this study, we investigate the pathloss and the effect of soil moisture content on the pathloss in the Vietnam popular type of soil. From that foundation, we will construct a WUSN based on 920 MHz LoRa wireless technology. The experiment is implemented to validate underground wireless communication. The results show a good performance and stability of the network, where the connection between two underground sensor nodes could be extended up to 3 m with 5% soil moisture content.

Business Intelligence (BI) systems have been widely implemented in various industries to improve decision-making processes. In higher education institutions, BI systems have shown great potential in predicting student performance, which is crucial for academic and admission activities. This article begins by introducing the concept of BI and its applications in higher education institutions. A literature review is conducted to examine the existing studies on BI implementation in higher education institutions and predicting student performance. Based on the literature review, this article proposes the development of a BI system named to address the predicting student performance problems. The system includes an integrated data extraction module, a user-friendly interface for the school’s managers, and several methods of data analysis based on machine learning and statistical probabilities. Binary Logistics Regression and Neural Network are used to forecast student performance. The results of the study suggest that the BI system can provide practical effects in academic and admission activities .

The preferred way to establish diverse smart computing environments is by encapsulating functionalities as services and providing them through well-defined interfaces. However, such service environments are prone to constant changes and require proper change detection, which is essential for service providers and client applications. Current methods for detecting changes in services focus on structural and semantic changes, but they do not consider the actual behavior of a service by examining input and output values. This can result in the selection of inappropriate services for an autonomous replacement, as the behavior of the selected service may differ significantly from the replaced service. In this paper, we propose an architecture that captures and evaluates the behavior dimension of services to provide more reliable service replacements in IoT environments. We achieved this by using machine learning algorithms and a multi-agent architecture called EVA.

In this work, a daily action recognition system for smart-building applications is developed. The system consists of a processing pipeline to perform tasks of human detection, pose estimation, and action class classification. The Yolov7-Pose was used for the human detection and pose estimation task, while a trained model based on the CRT-GC method was used for the action classification. The prediction of the start-to-finish duration of an action in a sequence video is performed via the sliding window method. For the trained model and the evaluation, a self-generated dataset of six classes of daily actions with challenging conditions was created. The evaluation results show that the Yolov7-Pose outperforms others in terms of accuracy, robustness, and computational efficiency. The pose estimation reaches an AP50 of 89.1%, and the action recognition has an mAP50 of 85.6%, in which the highest accuracy reaches 95.7%. The total computing time for the overall processing pipeline is 14ms. The obtained results indicate that there is a high potential for practical applications.

The integration of Inertial Navigation System (INS) into Global Navigation Satellite System (GNSS) utilizing Inertial Measurement Unit (IMU) has become increasingly common in Mobile Mapping Systems (MMS) and navigation. It enables the accurate determination of the location, velocity, and attitude of mobile entites in a seamless manner. Besides, thanks to advantages such as compact light weight structures, low cost and energy consumption, the Micro-Electro-Mechanical System (MEMS) IMU and GPS transceivers have been an active research area. However, the quality of the small-cost INS/GPS systems remains low, specially in GNSS-noise and without-GNSS environments. To improve the system performance, this study applies analytical constraints, consisting of non-holonomic constraints and zero-velocity updates, to the data unification, such as the Enhanced Kalman Filter. Experiments and data analysis are used to validate the benefits of our proposal.

The large-scale cloud computing environment has raised great challenges for fault analysis in infrastructure. The openness of cloud computing makes it challenging to assess the state of the infrastructure, which affects the data center's continuing availability. In addition, the fault detection in which prior-knowledge of faults has not been defined yet can make mistakes since using supervise learning. To address this issues, the fault analysis model of physical machines in cloud infrastructure is proposed by three components, i.e., abnormal score, fault detection, and ranking suspicious metrics. The proposed model is validated by using a current Google cluster trace dataset.

Ultrasonic imaging is commonly known for its popular reconstruction methods such as B-mode, which is widely used in commercial ultrasound devices. However, B-mode measurement is still limited in quality and detailed information about the imaged object. Recently, the method of tomographic ultrasound has become interested due to the strong development of hardware and software for devices. This paper proposes the application of frequency compounding technique in beamformed DBIM tomographic imaging. Beamformed approach using several probe transmitting elements simultaneously to offer a narrow beam which has the ability to minimize the noise effect has been applied for DBIM-based density imaging. When applying frequency compounding, it promises to improve the convergence speed, spatial resolution of imaged objects, noise reduction, and can create images of high-contrast objects. The numerical simulation results show that the image recovery performance is significantly increased and the noise is considerably improved when applying the proposed method.

Human detection plays an important role in several fields (such as autonomous mobile robots, bio-medical applications, military applications, etc.) and has received considerable attention from researchers in recent years. Especially, human gesture recognition provides information to predict human behavior for collision avoidance of the robots. The present paper proposes an approach in which deep-learning method and machine-learning method are integrated to classify activities and movements for multiple human targets. The proposed recognition process involves three sequential steps, namely the YOLOv5 model for detecting targets, the Media Pipe for drawing the skeleton, and the LSTM network for recognizing activities. The proposed method is examined through different scenarios. In the case of detecting the target with YOLOv5, the experimental results show that the loss accuracy always maintains below 5% for both training and the validation processes, and the mean Average Precision (mAP) of the designed YOLOv5 model is always higher than 99% for all consideration case studies. Furthermore, the activity recognition performance of the proposed method also successfully detects and tracks the behavior of the defined target inspected via three case studies: sitting, standing, and hands up. The experimental results prove the stability and precision of the method and point out that this approach can be applied to further studies and applications.

This work proposes some optimization solutions and design methodologies for cascaded Bloom Filter design developed from the standard Bloom filter architecture presented in our previous study. These optimization solutions are all based on features extracted from the input data after passing through the first Bloom filter layer, which is used for the next Bloom layers in the entire filtering process. In addition, a solution using a small capacity of CAM instead of the Bloom filter in the last layer is also considered and evaluated in comparison with the solution using the pure Bloom filters. The CAM-assisted filter design could almost suppress the false positive rate with a trade-off in a small false negative rate.

Nowadays, with the development of technology and the Internet, most households have surveillance cameras to observe everything around the house. Therefore, detecting abnormal human behaviors using videos generated by surveillance cameras has attracted much recent research. This paper focuses on applying the YOLO v4 to build the model detecting abnormal human behaviors, especially detecting fence climbing behaviors. Experimental results on the dataset, including 5340 images extracted from videos, showed that the model obtained the IoU measure of 71% and F1-score measure of 87%.

The network protection problem is one of the important issues in optical network design and belongs to the NP-hard problem. The problem becomes even more difficult in the next-generation optical networks - Elastic Optical Networks (EONs). In EONs, the protection problem must take into consideration extra elements such as power consumption, spectrum allocation requirements, and format levels. Furthermore, several recent studies focus on solving the problem of power-efficient network protection in EONs. Almost all prior research focused on heuristics because of scalability concerns. In this paper, we propose an ILP model for large-scale optimization that can solve huge instances. Experiments were successfully conducted on the NSFNET and USANET networks, with very reasonable computing times.

Educational technology has become an indispensable tool in modern education at all school levels today. With the number of K12 students in Vietnam accounting for nearly 20% of the national population, the educational technology sector has attracted startups developing many new EdTech products and investors interested in this field. Not to mention international products that developed in Vietnam market, domestic products have also reached hundreds. The rapidly increasing use of educational technology leads to user choice problems. With an extremely large market, it is become too difficult for users of choice, difficult for management as well as to develop standards to guide the development and circulation of products in Vietnam. There have been quite a few methods of evaluating E-learning systems, but the ranking of E-learning systems has not been mentioned much, it is a suggestion for efforts to perfect the production of the companies and attract investors and users. In this study, a method of evaluating and ranking eLearning products is proposed based on a set of criteria such as information quality, system quality, scalability, and user satisfaction. With 174 eLearning training systems in the Vietnamese market for K12 students, data are collected and measured according to the criteria from the reputable systems, then an appropriate ranking formula used to measure the specific value of the product is also suggested for users as well as support for experts in evaluating educational technology products at a later in-depth stage.

Elastography has emerged as a promising technique for non-invasive clinical diagnosis in recent years. This method estimates tissue elasticity by analyzing the speed of shear waves captured by an ultrasonic Doppler transducer, and offers several advantages such as high reliability, low cost, and non-invasiveness. However, the propagation of sliding waves in diseased tissue generates reflected waves that can affect the accuracy of the results. To overcome this limitation, this study introduces a novel technique that employs a directional filter to eliminate reflected waves and increase the reliability of the elastography results.

In many Vietnamese schools, grades are still being inputted into the database manually, which is not only inefficient but also prone to human error. Thus, the automation of this process is highly necessary, which can only be achieved if we can extract information from academic transcripts. In this paper, we test our improved CRNN model in extracting information from 126 transcripts, with 1008 vertical lines, 3859 horizontal lines, and 2139 handwritten test scores. Then, this model is compared to the Baseline model. The results show that our model significantly outperforms the Baseline model with an accuracy of 99.6% in recognizing vertical lines, 100% in recognizing horizontal lines, and 96.11% in recognizing handwritten test scores.

The existing works on in-vehicle infotainment (IVI) services have so far focused on only private vehicles, such as car. This paper proposes a framework of infotainment services for public vehicles (PVIS), such as bus or train. In the proposed PVIS scheme, an agent is employed to support a large number of users. From testbed experimentation, we see that the proposed PVIS scheme can provide scalable services for large number of users in public vehicle.

Global Navigation Satellite System with Realtime Kinematic Positioning (GNSS RTK) is now widely applied for land-based surveying to provide positioning solution at centimeter level. However, for highly dynamic surveying and mapping applications such as UAV photogrammetry, hydrographical surveying and mobile mapping that require a high frequency and continuous navigation solution, GNSS RTK only is insufficient. To overcome this issue, we propose a system that integrates GNSS RTK and Inertial measurement Unit (IMU) to provide navigation solution including position, velocity, and attitude. For this scheme, Extended Kalman Filter is used for data fusion. The conducted field test indicated that the proposed system and solution is enabled to provide navigation solution of frequency up to 50Hz with positional accuracy of centimeter in open sky view and decimeter in GNSS hostile environment.

Skin diseases, one of the common human diseases, could be life-threatening if not diagnosed and treated early. This study proposes a skin disease detection model based on some image processing techniques and deep learning architectures. First, we deploy a data pre-processing procedure to convert the input images to Hue-Saturation-Value (HSV) color space and remove their unnecessary information with a Hanning Window-based filter. After applying the Hanning Window-based filter, we downsize the image to 64 $$\,\times \,$$ 64 before fetching it into the learning model. Next, we train the Convolutional Neural Network (CNN) model on the processed image dataset and the original image dataset to compare the effectiveness of two approaches. The experimental results show that using HSV color space and Hanning Window-based filter can improve the performance in diagnosing six out of eight considered types of skin diseases.

Edge computing enables real-time intelligent services to be provided near the environment where the data is generated. However, providing intelligent services requires sufficient computing ability that is limited by edge devices. Knowledge transfer is an approach that transfers the trained model from an edge device to another edge device. In this paper, we propose a continuous deep-learning approach based on knowledge transfer to reduce the training epochs of a single-edge device without sharing the data in the edge computing environment. For training deep learning continuously in the network edge, each edge device includes a deep learning model and local dataset to fine-tune the reached model. The fine-tuning is completed, then, the updated model is transferred to the next edge device. For experimenting with the proposed approach, edge computing is comprised of multiple edge devices that are emulated by the virtual machines to operate the deep learning model and simulate the network communication. The deep learning model is developed for classification based on Convolutional Neural Network (CNN). As expected, the prediction accuracy is improved in multiple iterations of transferring in the edge computing environment.

The most frequent type of cancer among women is breast cancer. Patients with breast cancer have a considerably higher chance of survival when they receive early identification and treatment. The greatest technique for early breast cancer detection is screening mammography analysis. Because cancerous tissues and glands are close to the illumination, the results of analysis by radiologists are frequently limited; therefore, using computer analysis is more convenient. Image thresholding is one of the simplest methods for separating tumors; however, the results are only good if the mammograms have high contrast and the threshold value is correctly chosen. In this article, we present an approach for increasing image contrast, while preserving tumor edges by combining the stationary wavelet transform with morphological transforms. The tumor is then extracted using the adaptive multi-threshold. By using performance evaluation criteria for segmentation methods, such as accuracy, sensitivity, specificity, and Dice similarity coefficient, the results are compared to the ground truth data, demonstrating the method's accuracy.

There have been many studies that have shown that throwing margins have techniques to make the throw-in more effective. Therefore, this paper is based on the results of studies that determine the angle of the throw and the optimal velocity for conducting the pitching experiment, data processing, and assessment of the effect of height - weight - BMI on long-range, focusing on Asian people. The study was practiced with 70 members with different indicators, throwing the right technique based on experimental conditions equivalent to reality on the field. The results showed that weight had the greatest effect on the distance of the ball. The result shows that in the selection of male players to perform the throw-in situation, the following factors are prioritized: height over 1.6m, weight from 65–78 kg, and having a BMI classified normally or overweight. In the future, we hope to do more research to get throwing a throw-in situation with higher difficulty and use other factors to support athletes to achieve the best results from the proposed system showed that the hand acceleration and punch forces correlated strongly with an average acceleration of 28.3 m/s2 (without rotation) and 29.9 m/s2 (with rotation) producing an average force of 107.5 N and 139.9 N, respectively. These results show that the punching velocity had a great impact on the punching forces. The experiments also proved that the system can use to monitor the force, acceleration of the punch, and also a posture of practitioners when doing punches.

Based on the cooperation of neural network and extended state observer (ESO), in this paper, an approach for reinforcement control will be presented for Planar robot. By using the sliding surface as a state variable, the nominal system in quadratic form will be converted to first-order where the total uncertain component is estimated and remove by ESO. Then, a reinforcement algorithm will be added in collaboration to determine the nearly optimal solution of Hamilton-Jacobi-Bellman (HJB) equation. During the determination of the control signal, only one neural network is applied to reduce the computational complexity while still achieving the desired requirements. The simulation results of the algorithm will be examined on the Planar Robot with two degrees of freedom, thereby confirming the effectiveness of proposed control strategy.

In this article, we report on the results of the study on building a semantic tagging system using the English-Vietnamese bilingual corpus to create a lexical resource with lexical notes based on translation similarities, transfer of vocabulary, and classification schemes through bilingual connections. This tool plays an important role in building and developing natural languages processing systems such as automatic translation, text summary, text extraction, information retrieval, and question-answer automatic. In this context, our goal to see English as the source language is roughly translated into Vietnamese, we use an English-Vietnamese bilingual concept dictionary for the purpose of annotating the semantics of words. In our experiments, we used hand-annotated vocabulary sets, compared with the proposed model results, and the results were achieved based on an average vocabulary coverage of 83.05%, with accuracy reaching up to 82.40%. This test can be applied to build lexical la-bels for other languages than bilingual corpus whose source language is English.

Socially aware robot navigation has gathered more and more interest from research communities due to its promising applications. Recent breakthroughs in Deep Reinforcement Learning (DRL) have opened many approaches to archive this task. However, due to the data-hungry characteristic of DRL-based methods, many promising proposed works have only trained on simulation, making real life applications still an open question. In this paper, we propose (i) a new Synthetic Crowd Generation (SCG) framework along with (ii) a world model for generating valid synthetic data. As a data-generating framework, SCG can be easily integrated into existing DRL-based navigation models without changing it. According to evaluations on simulation as well as real life data, our SCG has successfully boosted the published state-of-the-art navigation policy in terms of sample efficiency.

Urban flooding is difficult to predict, and most cities lack the tools to track its evolution automatically. Surveillance camera systems are available in nearly every city, but they lack a smart function that would send an alert in the event of an emergency. To detect street flooding alarmingly, we suggest a highly scalable intelligent system. This system can simultaneously produce high-resolution data for future use and send out high-abstract warning signals. The chosen deep convolutional neural network model, U-Net with backbone MobileNetV2, achieved a classification accuracy of 89.58% and flood image segmentation accuracy of 95.33%. The demo prototype model is deployed on a cloud instance, serving up to 100 camera points. This method would create a highly-scalable measurement of street flood conditions without requiring the installation of new on-site infrastructure.

Nowadays, mobile robots have been popular not only in industrial applications such as materials transportation but also in non-industrial applications, e.g., human assistance. Among developed configurations, omnidirectional mobile robots have attracted great attention in recent times due to their superior maneuverability over their conventional counterparts. In this research, an application of a four mecanum-wheeled omnidirectional mobile robot (4-MWMR) in human assistance has been developed. By using image processing, the 4-MWMR is capable of following an authorized person, thereby assisting users in transporting large-size or heavy-weight materials. Good experimental results show the ability of the developed system to be used in practice.

This paper recommends intelligent control for mobile robots based on fuzzy logic controllers (FLC). This controller is designed with only two input state variables, such as position error, position deviation derivative of the robot, and one output variable, velocity. The robot is moved according to the trajectories set by fuzzy selection rules with an 9 × 9 matrix. The proposed FLC controller is compared with classical PID controller. The robot with the FLC controller moves to follow the trajectory with lower error and faster setup time than the PID controller. The efficiency of this controller is demonstrated by MATLAB/Simulink.

Path planning is one of the core research direction of mobile robot navigation. First, a grid-based method transfers the complex environment to a simple grid-based map. Hence, the mobile robot’s position is definitely determined in the grid map. To solve simultaneously the problems of the traditional A∗ path finding algorithm such as close distance from the obstacle, the shortest path, and path corners, the paper introduces the idea of the improved algorithm is to eliminate A* unnecessary nodes as for the purpose of reducing the computational scale. Then, the obtained path is smoothed by B-spline transition method. Hence, AMR’s optimal obstacle avoidance strategy based on A* algorithm will be completely constructed. Finally, simulated results are shown to prove the feasibility of the proposed method.

Deep learning models have proven effective in various computer vision tasks, including object detection and tracking. In this paper, we propose a method for detecting and tracking a specific person in a video stream using deep learning models, called DTTP model. Our approach utilizes a combination of object detection and re-identification techniques to accurately track the target person across multiple frames after recognizing the face of that one. We evaluate our method on publicly available datasets and demonstrate its effectiveness in tracking a specific person with high accuracy and improving the model’s speed by feature extraction of human pose and re-parameterization of deep learning models. Our DTTP model achieved 51.07 MOTA, 59.64 IDF1, and 47.31 HOTA on the test set of MOT17, which is higher accuracy and shows a favorable accuracy-speed trade-off compared to the state-of-the-art model like ByteTrack.

Developing very large scale distributed software systems is challenging from both functional and data management perspectives. Methods based on Microservices Architecture (MSA) have gained popularity for addressing the functional challenges. On the other hand, cloud-aware, very large scale NoSQL data management systems have also proved their effectiveness in tackling data management’s scalability challenges. Recent work have studied the combined approach for specific methods and systems. However, there has been no work that propose a complete method or study the underlying design principles. In this paper, we present the result of our initial research on this subject. We choose Cassandra as a case study as it is a popular system that supports cloud-aware, very-large-scale NoSQL data management. We propose the CaMSAndra software development method that combines the MSA and Cassandra methods. We define a UML metamodel for CaMSAndra and uses it as the basis for discussing the design principles. We analyse the relationship between bounded context and application workflow and, based on this, define a hierarchical service design that builds a service hierarchy by transforming an application workflow. We also discuss a data-driven cluster design in connection to the microservices. We demonstrate CaMSAndra with a well-known software domain called Hotel Reservation. We contend that our method is promising for developing very large microservice-based, NoSQL-based systems in general.

This article discusses the trajectory tracking control requirement in a Two-Wheeled Self-Balancing Robot (TWSBR) with linearization and discretization after utilizing optimal control consideration. A proposed optimal control system that employs states observer to estimate state variables and Approximate/Adaptive Dynamic Programming (ADP) controller. The unification between tracking problem and output feedback ADP based optimal control design is guaranteed. Finally, simulation studies are used to validate the proposed control structure and demonstrate the performance of this control strategy.

Digital Twin (DT) is one of the important enabling technologies for Smart Manufacturing and Industry 4.0, with a huge potential for many impactful applications in healthcare and industries. This paper presents a conceptual model of a DT system, with a proof-of-concept (POC) prototype of a robot for demonstrations and further investigations of DT applications in telehealth and in-home healthcare. The successfully developed POC prototype were tested to evaluate time delay, and possible errors when operating and controlling the virtual and physical models of a robot. The proposed conceptual model of a DT system can be used for demonstrations about DT, with further developments for potential applications in healthcare and industries, especially when it is integrated with emerging technologies such as artificial intelligence, machine learning, big data analytics, smart sensors, augmented reality and virtual reality.

Fractional Frequency Reuse (FFR) is a common technique of 4G, 5G and beyond cellular systems to improve the utilization of radio spectrum and Cell-Edge User (CEU) performance. Conventionally, there are various algorithms to identify the CEU in the FFR technique. This paper studies three well-known algorithms, called SINR-based, SNR-based and distance-based ones. Specially, to determine CEU, two first algorithms utilize the signal strength while the distance-based one follows the position of the user. The simulation results indicates that although the distance-based algorithm requires less signaling messages than others, it can derive the highest coverage probability.

Insider threats pose a significant risk to organizational data security, and many organizations implement information security policies (ISPs) to reduce insider threats. This study used the unified theory of acceptance and use of technology 2 (UTAUT2) to examine factors that predict compliance among nonmalicious employees. A partial least squares structural equation modeling approach was used to examine survey data collected from N = 158 nonmalicious employees. The analysis indicated that social influence and facilitating conditions were the only UTAUT2 factors significantly predicting nonmalicious employees’ compliance. The study’s findings suggest that organizations should focus on building workplace cultures emphasizing ISP compliance’s social importance.

The punch is one of the major components of martial art that is related to kinematic indicators and impact forces. However, the impact forces of punching postures have not been fully investigated. Therefore, the aim of this research is to propose a new system to measure the acceleration and force created by the punch and monitor practitioners’ stances during the punching process. A study was conducted in which six participants (3 females and 3 males) participated with at least 1 year of experience. Each participant performed 5 straight punches with the fist rotation and 5 straight punches without the fist rotation. Force was measured from the load cell, acceleration from an accelerometer, and stance was monitored via a motion analysis module using a computer. The experimental results from the proposed system showed that the hand acceleration and punch forces correlated strongly with an average acceleration of 28.3 m/s2 (without rotation) and 29.9 m/s2 (with rotation) producing an average force of 107.5 N and 139.9 N, respectively. These results show that the punching velocity had a great impact on the punching forces. The experiments also proved that the system can use to monitor the force, acceleration of the punch, and also the posture of practitioners when doing punches.

Identifying and predicting nutritional deficiencies during the growing process of the tomato plant (Solanum Lycopersicum L.) is crucial since mineral nutrients are essential to plant growth. This paper aims to predict and recognize the nutrient deficiency occurring in tomato plants’ flowering and fruiting stages by using two deep learning models, Yolov7 and ResNet50. The study focuses on predicting and classifying tomato plants’ malnutrition stage with an essential mineral nutrient, calcium (Ca2+). ResNet50 and Yolov7 are used to classify three stages of calcium deficiency in tomato fruits by analyzing the captured images of the development of tomato plants under greenhouse conditions. The dataset includes a total of 189 captured images that cover the different levels of calcium deficiency in tomato fruits. Of these, 80% (153 captured images) were used for the training dataset, and 20% (36 captured images) were applied to validate the test dataset. The purpose of this study is to recognize the stage of nutritional deficiencies in order to increase crop yields and prevent nutrient deficiency-related tomato diseases. By analyzing the tomato fruit images captured during tomato plant growth, the performance of ResNet50 and Yolov7 was validated, with accuracy rates of 97.2% and 85,8%, respectively.

Machine learning is an essential tool for crop yield prediction. Crop yield prediction is a challenging task in the agriculture and agronomic field. In crop yield, many factors can impact crop yields such as soil quality, temperature, humidity, quality of the seeds, rainfall, and many more. To give an accurate yield prediction with the right machine learning algorithms we need to process a huge amount with the selections of impactful features. In this study, we performed a systematic literature review to select machine learning methods and features that can analyze large amounts of data and give more accurate results. We discuss the lacking’s of existing research and generated a comparative analysis to give a clear aspect of the better solutions. From a critical evaluation and specific search criteria, we found – papers from AGORA that contain many more different databases such as MDPI, Tylor and Francis, IEEE, etc. From 660 we selected 50 papers from that number that were used more efficiently and gives accurate results with a thorough investigation with the help of our selection criteria and generic research questions that can filter and bring out the more relevant papers regarding these fields. From the selected papers we evaluate the methods, and geographical areas that have been selected for acquiring data analyzed the features, and have a thorough inspection of the selected factors that have the most impact on yield prediction. This study will help future researchers to give a clear understanding of existing research and guide them to generate a more effective model.

This paper introduces a next-generation device for crop yield prediction that utilizes IoT and machine learning technologies. The device was implemented and tested, and it was found to have a high level of accuracy in predicting crop yields. It is a combination of three different machine learning models: Artificial Neural Network (ANN), Fuzzy Logic, and Support Vector Machine (SVM). The IoT sensors in the device gather data on various environmental and soil conditions such as temperature, humidity, and soil moisture, which is then fed into the machine learning models. The ANN is used to analyze the sensor data and extract features, the Fuzzy Logic model is used to handle uncertainty in the data and make predictions, and the SVM model is used for classification. The device was tested on various crops and it was observed that the accuracy of the predictions was good and the results were comparable to other state-of-the-art techniques. This technology has the potential to revolutionize the way farmers manage their crops and improve crop yields. It can also be used for crop forecasting, crop monitoring, and precision agriculture. By providing accurate and real-time information about crop yields, this device could help farmers make better decisions about their crops and increase their overall productivity and profitability.

The Efficient Manifold Ranking (EMR) is a scalable graph-based ranking algorithm that is applied widely in Content-based Image Retrieval (CBIR). However, the effectiveness of an EMR algorithm depends on (1) the feature extraction technique applied to images to extract feature vectors and (2) the relational graph architecture of anchor points built inside the EMR. To address the first problem, EfficientNet-B7 + is proposed in this article which is fine-tuned from a pre-trained model of EfficientNet and is used to extract deep feature vectors of images. Regarding the second problem, we adopt the relational graph architecture of lvdc-EMR, in which the anchor points of the graph are generated by a variant of the Fuzzy C-Mean (FCM) clustering algorithm that was developed by our research team. The experiments conducted on three benchmark datasets Logo2K +, VGGFACE2-S, and Corel30K bring the mean image retrieval accuracy to 88%, demonstrating the effectiveness of our proposed method. Comparing the average values while retrieving using lvdc-EMR, the proposed EfficientNet-B7 + obtains from 4% to 6% better than the original EfficientNet-B7.