The Fourth Industrial Revolution and Beyond

Rice diseases are a prominent cause of reduced crop yield. Viruses, fungi, and bacteria might transmit all kinds of rice diseases. Early identification could have a significant impact in managing diseases and their spread. Automated rice disease identification systems could excel if they can have high predictive accuracy while requiring low computational resources. Nowadays, deep learning techniques are implemented in order to lessen human effort and time. In this paper, we present different experiments and findings related to the detection of three rice diseases: bacterial leaf blight, brown spot, and leaf smut. Multiple large pretrained models are fine-tuned on a relevant dataset. The fine-tuned models are then pruned and in one case quantized to achieve better performance. Most notably, the ResNet50-based model achieves 97% accuracy among all of them while requiring only 7200 FLOPS. In terms of correctness, this result is almost 4% better than previously published work while requiring a markedly lower number of FLOPS.

Bangladesh is among the few nations on this planet which will confront the lethal results of a worldwide temperature alteration. Global warming is the crisp product of climate change and it is real. To predict the global warming rate of Bangladesh, we have used eight machine learning algorithms with a dataset consisting of about five thousand data. We have collected data from the University of Dayton—Conversational Guard Organization from the Average Daily Temperature Library. We have proposed neural network systems for calculating the nonexistent data of the temperature dataset. We have also performed a correlation test for both test and train datasets and that outcome was significant. Our data period is from 1995 to 2010. The accuracy of our machine learning experiment shows that the Naive Bayes algorithm can predict the global warming consequences in Bangladesh with over 99 percent accuracy. Global warming is happening, sea level of the Bay of Bengal is rising, our experiment has showed that global warming is affecting Bangladesh, and the outcome of that will be immense. It is the right time we should start our awareness of climate change.

Marketers use different marketing strategies to elicit the desired response from the target customers. To measure customer engagement, they usually conduct one-on-one interviews, surveys, broad polls, and focus group discussions. However, these are costly and sometimes unreliable. On the other hand, neuromarketing measures customer response to marketing stimuli by measuring the electrical activity of the brain and has the potential to address these drawbacks. which can be overcome by neuromarketing. In this work, we proposed a prediction algorithm that can identify consumer affective attitude (AA) and purchase intention (PI) from EEG signals from the brain. At first, the raw EEG signals are initially preprocessed to remove noise. After that, three features are extracted: time, frequency, and time frequency domain features. Wavelet packet transform is used to separate the EEG bands in time-domain feature extraction. Then, for feature selection, wrapper-based SVM-RFE is utilized. Finally, we use SVM to distinguish the classes in AA and PI. Results show that for SVM with radial basis function kernel performs better to classify positive and negative AA with an accuracy of $$90\pm 4.33$$ 90 ± 4.33 and PI with an accuracy of $$75\pm 2.5$$ 75 ± 2.5 . So, EEG-based neuromarketing solutions can assist companies and organizations in accurately predicting future consumer preferences. As a result, neuromarketing based-solutions have the potential to increase sales by overcoming the constraints of traditional marketing.

This paper mainly focuses on the job requirements of the IT-Software industry during the COVID-19 pandemic situation using recent data on vacancy posting and job ads views on Bangladesh’s largest online job portal. The impact of COVID-19 in every sector in Bangladesh had started in March 2020. This pandemic made us adopt the online system almost in every aspect of our life. That is why the IT-Software job industry and the requirements in these jobs have reformed newly. This paper will show the major requirements and skills for the IT-Software job during this pandemic situation and help us to understand which skills we should develop in this pandemic situation.

At the beginning of the university, students often face confusion in career choices. Most of the students cannot understand what field they are interested. There are many existing systems where some researchers have worked with fresh graduates and some with selected tenth or twelfth-grade students to help them get an idea about one career option. However, those research works failed to help them choose a domain of interest from many opportunities in their field. Therefore, this research aims to propose a system for predicting interest of beginner-level engineering students for better career planning at an early age. This analysis gives a chance of finding students interest in four specific demanding fields (web development field, graphics design field, Android development field, and data science field). This research starts with collecting primary sources of data using structured questionnaires. The data are preprocessed for creating a well-formed dataset. Then, the research is done in two parts: statistical analysis (Chi-square test and binary logistic regression analysis) and machine learning analysis (decision tree, SVM, and multinomial logistic regression). In machine learning analysis, appropriate features are selected first, then build the models. The results from both parts are then evaluated using a confusion matrix and choose the best prediction. The research shows noteworthy results with respect to existing works.

Road accidents have become an unwanted phenomenon in our daily lives. This paper aims to give a clear concept of the huge number of road accidents in the districts of Bangladesh. It also provides a detailed summary of the total accidents finding how different features are contributing to the increment of accidents and pointing out the accident-prone regions so that the losses can be minimized in accidents with limited resources. This paper presents various charts and graphs that show variation in the number of accidents based on different factors, victims, injured-to-dead ratio, locations of accidents in terms of number. The paper analyzes data on road accidents in Bangladesh. The experimental results show that the number of accidents is related to date, time, locations, and vehicle types. The results also reveal that 66.4% are injured and 33.6% are dead. It is a notable turning point over the past decades for this country. The result of this analysis exhibits different features that are important to reduce road accidents and to manage the transportation system of Bangladesh including pedestrians to impact public health and the economy of the country.

Collaborative filtering is a sophisticated recommendation system strategy that efficiently manipulates recommendations corresponding to preferences of users. Computation of user similarities is one of the features of the collaborative filtering strategy. This paper has presented a user similarity computation strategy for collaborative filtering using word sense disambiguation technique. The staple contributions of this research are utilizing associative database model for representing searched words of users uniquely and word sense disambiguation technique for removing ambiguities for understanding the exact preferences of users. In addition to, a self-tuning constant and a computation model have been also developed to compute similarities of users. The proposed approach has overcome all the prominent challenges against the recommendation system and performed very promisingly on two distinct datasets with about 93.40 and 96.10% accuracy. Moreover, the performance results and comparison demonstrate that the proposed user similarity computation model for collaborative filtering is comparatively optimal than other existing collaborative filtering approaches.

Computational cost associated with Computational Fluid Dynamics (CFD) simulation is a major bottleneck even with advanced computational facilities. This study introduces a method to overcome this major challenge by using multi-fidelity physics informed neural network (MPINN). Mesh domain elements in a CFD model determines the fidelity of our provided multi-fidelity analysis. CFD simulations with a lower number of mesh domain elements are computationally inexpensive but produce results with lower accuracy levels. In contrast, a large number of domain elements will result in higher accuracy with high computational cost. Using MPINN, we can accurately forecast the relationship between low and high-fidelity findings, which allows us to extract high accuracy CFD results using computationally cheap low-fidelity simulations. We conducted a simple stationary benchmark analysis of laminar mixed convective heat transfer in a square-shaped lid-driven cavity to find the optimal amount of high-fidelity training data needed for accurate outcomes while reducing computational costs. The findings reveal that the heat transfer performance parameter may be estimated with high precision while saving up to 51% in computing expenditures using MPINN with only 20% high-fidelity training data. This study is the first implementation of MPINN merging with CFD simulations to predict the performance of thermo-fluid systems. This method will allow us to solve complex industrial and engineering problems with higher accuracy while being computationally efficient. Nonetheless, this approach eventually led to faster system/product design, evaluation, and quality enhancement with resource optimization through the synergistic integration of multidisciplinary key components of the 4th industrial revolution.

Due to the lack of traceability and proper authentication process, product counterfeiting is one of the main challenges in supply-chain management. In e-commerce business, merchants act as the central and only single controlling authority which leads to less trust and transparency. This creates a chance of product counterfeiting which can be reduced using blockchain with other supporting technologies. As products are delivered in multiple channels through many hands, there needs to be a way to identify original products and a proper product handover process. Many technologies are introduced to combat product counterfeiting such as barcode, QR code, NFC tag, hologram, etc. But these technologies are either expensive or lack full proof of product counterfeiting. Our research shows that a combination of destructible dynamic QR code, product location tracking along with blockchain technology can greatly reduce product counterfeiting in the e-commerce business. This paper proposes a novel approach of a product authentication system where a blockchain-based web application and a mobile application are introduced instead of a traditional product authentication system. Our evaluation shows that using this system, a consumer can easily identify whether a product is counterfeit or not.

The most Bangladeshi government and non-government scientific organizations have comprised both wired and wireless system network infrastructure. When such network infrastructure has been designed and developed at these organizations, ensuring network security has become a highly challenging issue. For this purpose, Next-Generation Hardware Firewall (NGFW) has been implemented by most scientific organizations. This hardware firewall is a very effective technology to perform network security. In this paper, we have chosen Atomic Energy Research Establishment (AERE), Savar under Bangladesh Atomic Energy Commission (BAEC) as an experimental zone. To facilitate research and services, Internet-based network infrastructure has been established at AERE, BAEC. A Next-Generation CISCO-based network firewall has been integrated with the implementation of AERE network infrastructure to enhance and ensure network security. We have experimented and collected data from October 09, 2020 to October 09, 2021 at the network system of AERE. By collecting and evaluating these statistical data in terms of different network security parameters, we have investigated AERE’s network data like controlling ingress and egress traffic, monitoring traffic flow, malware threat analysis, web content and application filtering, URL filtering, decrypting traffic results, etc. In this paper, we have illustrated how network and application security is being achieved by the NGFW firewall and as a resulting factor, the summarization of scale up the network security for a scientific organization is enlightened.

The upcoming sixth-generation (6G) mobile network introduces different communication technologies for efficient and controllable propagation. Backscatter communication (BackCom) is one of those which can transmit signals using the passive tag. This BackCom technology can harvest energy from the ambient signal sources (such as television and mobile phone tower) and is very energy efficient for 6G network and Internet of Things communication. For smooth and controllable propagation of signals, the reconfigurable intelligent surface (RIS) communication technology is very effective. With the help of its electronically controlled reflected path, the RIS technology improves the signal and network quality for better performance. This paper presents a crypto holographic RIS-assisted BackCom for secured and lossless communication in the upcoming network technologies. This combined communication technology will be energy efficient. It will increase the signal strength and data rate which are very important for the next generation mobile network. Research findings clearly show that when we use RIS-assisted BackCom, the channel gain, data rate as well as energy efficiency of signal reflecting surface are also increased significantly.

With the advent of wearable body cameras, human activity classification from First-Person Videos (FPV) has become a topic of increasing importance for various applications, including life-logging, law enforcement, sports, workplace, and health care. One of the challenging aspects of FPV is its exposure to potentially sensitive objects within the user’s field of view. In this work, we developed a visual privacy-aware activity classification system focusing on office videos. We utilized a Mask R-CNN with an Inception-ResNet hybrid as a feature extractor for detecting and later blurring out sensitive objects (e.g., digital screens, human face, paper) from the videos. We incorporate an ensemble of Recurrent Neural Networks (RNNs) with ResNet, ResNeXt, and DenseNet-based feature extractors for activity classification. The proposed system was trained and evaluated on the FPV office video dataset which is a subset of the BON [1] vision dataset for office activity recognition (2021) and includes 18 classes made available through the IEEE Video and Image Processing (VIP) Cup 2019 competition. On the original unprotected FPVs, the proposed activity classifier ensemble reached an accuracy of 85.078% with precision, recall, and F1 scores of 0.88, 0.85, and 0.86, respectively. The performances were slightly degraded on privacy-protected videos, with accuracy, precision, recall, and F1 scores at 73.68%, 0.79, 0.75, and 0.74, respectively.

In Bangladesh, scarcity of land and fast population growth is putting a strain on the land-man ratio. The property ownership registration system in Bangladesh is incomplete and insufficient. As a result, various government agencies handle various documents, and bureaucracy flaws enable organized crime. Considering these, we are proposing a method where we have come up with a Blockchain-based solution. We build a Blockchain-based system to ensure that individuals are not deceived under any circumstances. In our system, where land data will be safe and secure, data synchronization and transparency will be available, as well as ease of access, irreversible record management, and rapid and low-cost quick solutions. We designed a modern architecture to digitally store land records using Blockchain-based Hyperledger Fabric to ensure land security. This proposed model has b‘een presented by private Blockchain, taking into account the technological expertise and authority of the people and government. Finally, we compare our proposed architecture with the existing land records management model. Our system provides more security and data privacy than other models and saves both money and time in our daily lives. This will ensure that our system is transparent and acceptable in all directions.

Blockchain technology enables a distributed and decentralized environment with no more central authority. To increase the accuracy of electronic healthcare records (EHRs) and establish a secured patient-centric approach, Blockchain can be a smart solution in this case. Blockchain is a distributed ledger technology that allows for the secured transfer of medical records while a transaction is created in the network. As Blockchain is decentralized and transparent if one user tampers with medical records of transactions, all other nodes/participants would cross-reference each other and easily pinpoint the node/participants with the wrong information. In this paper, we proposed a smart contact-based Blockchain technology and implemented it in the Hyperledger framework to make the healthcare system more confidential and secure. Moreover, Smart contracts give us more security while transacting data and reduce the entire transaction cost, and also make faster transaction speed.

Recently, intelligent reflecting surfaces (IRSs) have seen an upsurge of interest due to their ability to make the wireless environment programmable, which has historically been treated as an uncontrollable natural phenomenon. Even passive IRSs consisting of many reflecting units can autonomously adjust the reflection coefficients to alter the phase and amplitude of the incident signals. However, optimal reflection design for large IRSs are deemed to be impractical due to the underlying computational complexity. In this paper, we design IRS-assisted programmable wireless environment for secure communication using deep learning techniques. More specifically, we consider the k-nearest neighbor learning algorithm for significantly reducing the computational complexity in IRS design. Simulation results demonstrate the effectiveness of the proposed deep learning-based solutions as compared with traditional alternating optimization.

Internet of things (IoT) is a process of inter-networking of constitutional design, vehicles (also referred to as “connected devices” and “smart devices”), and various objects—included with software, electronics, actuators, sensors, and internetwork connectivity which is used to gather and interchange of data and information. IoT can control objective and gather information from distance across the network. Recently smart home systems attained more popularity to increase the quality of life where smartphone application plays an important role to monitor home apparatus using wireless communication. The target of this research is to develop an IoT-based smart home to control specific devices and monitoring house using android mobile devices. Features that are used here are temperature, gas detection, and door lock. A channel, alarm system, door lock can be visualized by the users from any mobile device. If any change happened in the monitored data, user will be notified.

A microcontroller-oriented trans-impedance instrumental system was designed and simulated. The system was considered as the alternative to pesticides for repelling weevil from stored rice in Bangladesh. The circuit system was simulated on the open-source TINA-TI SPICE-based analog simulation software. The advantages make the system efficient for repelling weevil. The system has been designed for easy installation, availability at an affordable cost. Effective radiation repelling the weevil ensures grain quality. The system had an application of driving the relevant frequency ultrasonic transducer (30–90 kHz). The radiation was considered with an effective beam angle to the direction of the rice stored area. Relevant low ultrasonic frequency radiation makes the insect embracement. The output signal was calculated as 24 mA and 15 V, and the signal contained 16 harmonics with a total phase change of -18.54 degrees. The system was constructed with the series oscillator and trans-impedance instrumental circuit. This configuration has turned into a quite stable signal output. To achieve the desired low frequency and low distorted ultrasonic signal, a narrow bandpass filter, frequency division, current to voltage conversion, and precision-level pre-amplification instrumentation circuit were introduced in this system. A low harmonic distortion was observed after the simulation of the circuit. The system could be generated any relevant low frequency by uploading a pre-program. The recommended radiation can repel the weevil during the grain infestation. The different low-frequency outputs could also be helpful for other relevant harmful insects.

This research work conveys the computational analysis of hafnium (IV) oxide and its doped crystal by Si, Ge and Sn replacing on the oxygen atom in HfO2, as hafnium (IV) oxide has been used in power-electronics devices of MOSFETs and electronics as RRAM due to wide band gap which makes a vast problems creating high resistances. Regarding this case, the hafnium (IV) oxide has selected inputs how the band gap has decreased later than doping by the large surface area atoms, such as Si, Ge and Sn. The first principle method in view of density functional theory (DFT) expresses the structural geometry, electronic structure and optical properties utilizing conformist calculations pertaining to HfO2 executing the computational avenue of the CASTAP regulation from material studio 8.0. The band gap was found in 4.340 eV, 2.033 eV, 1.686 eV and 3.210 eV for HfO2, Hf0.88Si0.12O2 Hf0.88Ge0.12O2 and Hf0.88Sn0.12O2 crystals through the Generalized Gradient Approximation (GGA) with Perdew Burke Ernzerhof (PBE), and the DFT and PDOS were simulated for evaluating the nature of 6s2, 5p6, 4f14, 5d2 orbital for a Hf atom, 3s2, 2p6 orbital for Si atom, 4s2,3p6, 3d10 orbital for Ge atom, 4d10, 5s2, 5p2 for Sn atom, 2s and 2p orbital for O atom of Hf0.88Ge0.12O2 and Hf0.88Sn0.12O2 crystals. The optical properties, for instance, absorption, reflection, refractive index, conductivity, dielectric function and loss function, were calculated.

With the advancement of wireless communications technology, 5G wireless mobile communication will be developed into a new generation. Multiple inputs multiple-output (MIMO) technology is predicted to be a useful element in the 5G communication area to enhance the performance that is yet unexplored in literature. This paper offers a suggested antenna model for minimizing return loss and a good efficiency factor for inset fed multiple-input-multiple-output (MIMO) microstrip patch antenna with a small size of mm is proposed for 38 GHz (Ka-band) which is in 5G frequency bands. Although many publications addressed this work, minimization of the system loss and increasing return loss, the gain is not yet found in the present literature. The proposed antenna consists of two antenna elements that are parallel to each other. Both antennas are constructed with the main substrate of Rogers RT 5880 and a superstate of polyimide film. The architecture of both patch antenna is 3.561 mm * 2.449 mm * 0.254 mm, return loss −66.009846 and −66.008594 dB, Voltage Standing Wave Ratio (VSWR) of 1.0010017 and 1.0010019, and gain 7.512 dB for both antennas. Considering all of the mentioned factors, the developed antenna is supposed to be suitable for or 5G communication technologies in the future.

On the verge of the 4th Industrial Revolution, the growth of electric vehicles (EVs)(e.g., Easy-bikes) all over the world, especially in developing countries, leads to grid overloading due to non-coordinated charging of large number of EVs. Coordinated charging of EVs is expected to mitigate this problem. This paper proposes a variant model of Coordinated charging, i.e., Charging Capacity Model (CCM), which sets a limit on the number of EVs allowed to be charged at different hours of the day. The proposed CCM uses a modified Particle Swarm Optimization (PSO) algorithm. For implementation, Internet of Things (IoT) and data analytics are also used. The proposed model is simulated for the case of Bangladesh, with a specific type of EV called the Easy-bike, which is a popular three-wheeler vehicle in the country. This model achieves peak shaving and valley filling for the EV load curve and calculates hourly charging capacities for each charging station for the average load curve of eleven months. These results enhance the policymakers to mitigate any overloading problem due to EV integration.

Marine life and environmental monitoring of deep sea have become a major field of interest for quite a long time because of the immeasurable region of the area of the ocean that comes with its own dynamics and vulnerabilities. Creating the Underwater-Internet of Things (U-IoT) model within Underwater Wireless Sensor Network (UWSN) provides the scope of ensuring proper marine life monitoring which supports the aspects of 4th Industrial Revolution. The U-IoT network model is designed for an automated, efficient, smart process of data transfer for both underwater and overwater communications through acoustic waves and Radio Frequency (RF) data transfer techniques, respectively. The proposed U-IoT network model is created with an optimum number of autonomous underwater vehicles (AUVs) and surface sinks in order to address Bangladesh’s overfishing problem (e.g., hilsa overfishing problem) which guarantees efficient management of the banning period by the authority. The network model is evaluated by comparing different deployment methods of AUVs and surface sinks taking the South Patch region of Bay of Bengal as the target area. The result shows that the proposed model transfers adequate data of marine life motion from the seafloor can enhance efficient administration of the overfishing problem.

A beam splitter focused on a dual-core photonic crystal fiber with a hexagonal lattice structure has been designed. The Finite Element Method (FEM) is used to create and characterize the polarization splitter. The splitter can split polarized light into two polarization states at the wavelength of 1.55 $$\upmu $$ μ m. This work presents a working wavelength of about 1300–1950 nm. An extinction ratio that values lower than 10 dB with a total length of 850 nm can ensure an ultra-broadband bandwidth of 650 nm covering all communication bands. This proposed splitter may be capable of playing a vital role in communications and sensing applications.

The mobility patterns of nodes significantly influence the performance of delay-tolerant network (DTN) routing protocols. Trace-based mobility is a class representing such movement patterns of nodes in DTN. This research analyzes the performance of DTN routing techniques on trace-based mobility regarding delivery ratio, average latency, and overhead ratio. Three real traces: MIT Reality, INFOCOM, and Cambridge Imotes are implemented on five DTN routing techniques: Epidemic, Spray and Wait, PRoPHET, MaxProp, and RAPID. For more explicit realization, Shortest Path Map-Based Movement from synthetic mobility model has also experimented with the traces. The Opportunistic Network Environment (ONE) simulator is used to simulate the considered protocols with these mobility models. Finally, this research presents a realistic study regarding the performance analysis of these DTN routing techniques on trace-based mobility along with Shortest Path Map-Based Movement by considering the variation of message generation intervals, message Time-To-Live (TTL), and buffer size, respectively.

This paper instigates the findings on switching and DC characteristics such as transfer characteristics and output characteristics as well as carrier concentration, electric potentials of the metal–oxide–semiconductor field-effect transistors (MOSFET) after changing its materials by silicon (Si), germanium (Ge) and silicon carbide (SiC) using COMSOL Multiphysics 5.5 software. Though there have been various researches on the MOSFET with many materials for years, the comparative study on MOSFET’s DC characteristics is presented here showing what changes happened with the materials of it. Among the three nMOS, the Ge-MOSFET notably shows the minimum threshold voltage along with maximum terminal current rating than the other Si- and SiC-based MOSFETs for exactly the same study setup, which means the Ge-based MOSFET as a voltage-controlled device needs less voltage to switch but give a higher range of saturation current as a current source, which all are discussed here.

This article represents the design and performance analysis of a compact (36.25 × 34.75 × 0.79 mm3) ultra-wideband (UWB) microstrip patch array antenna having parasitic elements whose operating frequency ranges from 2.77 to 12.85 GHz. The UWB antenna is mounted on Roger’s RT 5880 (lossy) substrate (2.2, 0.0009). The entire design process and simulation is performed by CST. The results of the designed UWB antenna show that the reflection coefficient is −32.8 dB at 3.45 GHz, −35.33 dB at 4.68 GHz, −22.524 dB at 7.78 GHz, and 18.23 dB at 12.09 GHz, gain is 3.92 dB at 4.68 GHz, directivity is 4.55 dBi at 4.68 GHz, and VSWR is 1.0348 at 4.68 GHz. Moreover, the maximum radiation efficiency is 98%. Due to compact size (995.153 mm3) and ultra wide operating band of 10.08 GHz, the designed antenna is a suitable model for many wireless applications including microwave imaging (3.1–10.6 GHz), WiMAX (3.4–3.6 GHz), WLAN, WiFi-5/6 (5.180−5.925 GHz, 5.940–7.075 GHz), and Sub-6 GHz (3.3–4.9 GHz) applications. The designed antenna has a stable radiation pattern with a good gain and efficiency.

Intelligent transport system (ITS) aims to launch constructive vehicular connectivity and synchronization techniques which notably improve travel efficiency by reducing traffic gridlock, irregularities, mishaps, etc. There are numerous applications of ITS innovations in first world countries where emergent nations are yet depending on the fragmented disorganized analog public transportation system. The key challenge in underdeveloped countries is the maintenance of efficient timetables and scheduling of public transportation systems. In this paper, we have proposed a bus transit framework that consists of a timetable and a bus-driver scheduling algorithm to resolve this predicament. A coherent timetable considering passenger load in traffic rush hours is generated for scheduling the fixed number of trips and lines. To schedule the buses and drivers, three selection procedures—First Serve Check First (FSCF), Equal Allocation (EA), Random Allocation (RA) have been implemented and analyzed with varying numbers of trips and lines. The experimental results depict that, FSCF schedules with a minimum number of vehicles and drivers, though the algorithm runtime is higher compared to other approaches.

Incongruity between news headlines and the body content is a common method of deception used to attract readers. Profitable headlines pique readers’ interest and encourage them to visit a specific website. This is usually done by adding an element of dishonesty, using enticements that do not precisely reflect the content being delivered. As a result, automatic detection of incongruent news between headline and body content using language analysis has gained the research community’s attention. However, various solutions are primarily being developed for English to address this problem, leaving low-resource languages out of the picture. Bangla is ranked 7th among the top 100 most widely spoken languages, which motivates us to pay special attention to the Bangla language. Furthermore, Bangla has a more complex syntactic structure and fewer natural language processing resources, so it becomes challenging to perform NLP tasks like incongruity detection and stance detection. To tackle this problem, for the Bangla language, we offer a graph-based hierarchical dual encoder (BGHDE) model that learns the content similarity and contradiction between Bangla news headlines and content paragraphs effectively. The experimental results show that the proposed Bangla graph-based neural network model achieves above 90% accuracy on various Bangla news datasets.

Waste management systems and their inherent problems are still a matter of great concern amid this cutting edge of science and technology. Pile of untreated waste in open environment aids to a wide range of problems such as air and water pollution, untidy and unhealthy surroundings, wastage of recyclable materials, potential health risk of waste management workers and many more. The root cause of this problem points to a single fact that is too much manual labour involved in the garbage treatment process (collection, separation and recycling)—cannot keep up to the pace with which garbage generation happens. An efficient recycling method is imperative to solve this problem, which can be achieved by a fast, real-time garbage detection and classification system. In this research, we will propose a novel deep learning-based approach for automatic detection and classification of five kinds of waste materials, namely, kitchen waste, glass waste, metal waste, paper waste, plastic waste, from the garbage dump for an efficient recycling process, which not only improves the efficiency of the current manual approach but also provides a scalable solution to the problem. The contributions of this paper include a fully human labelled data set consists of 2200 images of garbage dump with 135,541 annotated objects from aforementioned categories and a real-time garbage object localization and classification framework based on a lightning fast, fairly accurate and end-to-end deep learning algorithm. For the baseline, we have articulated a single-stage object detection and classification framework and initialised the detector training process with pre-trained weights (trained on MS COCO data set) of the feature extractor. Then, with some fine-tuning and employing a few transfer learning tricks, we have proposed a waste object detection framework, that yields a mAP of 66.08% at an IoU threshold of 0.35 with an inference speed of roughly 55–58 ms in a single-GPU environment on both images and videos. It surpasses the performances of all the contemporary frameworks which deal with waste separation task.

With an aging population, the demand of transportation is increasing. As a result, the road safety has become a critical issue. The road condition may differ from one place to another. Early detection of road condition can prevent number of accidents and casualties. In this paper, a system is proposed to reduce difficulties of the rider. The system can exert the unknown crippled point, potholes of any road by the acceleration of a vehicle whenever it passes away. Besides the position tracking and monitoring system, it will locate the respective location automatically and update current condition in Google Map. This location tracking information will help the upcoming rider to get the overview of the road beforehand.

The mobile banking system is now very interested in blockchain technology because of its security mechanisms, data availability, increased transaction processing speed, and decentralized a database. Users of mobile financial institutions (MFIs) can use a mobile device to deposit money, withdraw cash, and send and receive money from their accounts. Mobile banking allows users to access account information, payments, deposits, withdrawals, and transfers, as well as investments, ATM assistance, and content services. The distributed ledger technology (DLT), often known as a blockchain, is a tamper-proof digital ledger that can store data and distribute it among connected nodes. Many scholars have contributed to the blockchain technology and consensus mechanism in their earlier studies. The process of integrating consortium blockchain technology and the Proof of Authority (POA) consensus protocol into Bangladesh’s mobile banking system is the subject of this research. A model of blockchain architecture with its classification and consensus protocol is also discussed. Moreover, the 51% security level of POA consensus protocol is justified. This initiative will have a huge impact on the financial industry, as well as improve the security of mobile banking.

Electricity usage is directly connected to the socio-economic growth of a society. As a developing country, the increased demand for electricity supply in Bangladesh has increased the emphasis on the need for accurate and reliable energy measurement methods. The estimation of the electricity consumption from households, institutions and industries is a must for a government to profile the country’s power capabilities and to plan for new installations if needed. From the perspective of the developing countries, mining authentic information about the ratio of the electricity consumption to its production is a matter of dilemma due to the conventional metering and billing system for the consumed electricity. In this paper, we have proposed a design which eradicates many of the drawbacks of the current system by integrating the energy measurement technology with IoT. Xtensa dual-core 32 bit microprocessor, Class 1 single phase energy measuring device, MODBUS, Wi-Fi, GSM/GPRS as communication protocols, MQTT as server, and MySQL as database system have been used in the proposed system. In this paper, real-time energy data monitoring, load control and emergency alert scheme have been demonstrated. Besides, the daily load-wise electricity consumption has been estimated throughout a week to analyze the fluctuations of three different loads considering their electricity consumption (kWh) and the balance credited (3.75BDT/kWh) on a daily basis. The output of this research has been found electrically, statically and economically very significant, which can be implemented as the next generation IoT based Smart Energy Metering System (SEMS) in the perspective of Bangladesh.

A methodology based on the use of finite-difference time-domain calculations and Genetic Algorithm is proposed that predicts aluminum nanoparticles (NPs) can be used for improving the opto-electronic performance of thin-film solar cells. Aluminum nanoparticles were coated with a thin silica shell layer that resulted in shifting the absorption properties of aluminum nanoparticles to longer wavelengths and aiding the chemical isolation of the highly reactive aluminum nanoparticle core. Silicon thin-film solar cells were then modified with various sizes of aluminum nanoparticles with varied shell thicknesses that were placed on top of the silicon substrate, embedded inside it, and placed in a “sandwich” configuration, with one particle on top of the substrate and another embedded inside, respectively. The results showed that Al-silica core–shell nanoparticles in a “sandwich” configuration demonstrated the most improved opto-electronic performance of solar cells when compared to the other configurations studied. These promising results can open the door for future collaborative use of Genetic Algorithm and FDTD calculations to design high sensitivity photovoltaic devices.

According to the United Nations (UN), the 2030 Agenda recognizes ending poverty in all of its forms and dimensions, including severe poverty which is the biggest global challenge and a necessity for sustainable development. Sustainable Development Goal (SDG) 1 aims to “end poverty in all its forms everywhere” but a global uniform approach has not been found rather than country-specific programs that tackle poverty with international initiatives to support them toward the SDG poverty eradication goal. With the opportunities generated by the 4th Industrial Revolution, nations can optimize their economic strengths to end poverty. Since the textile industry is one of the economic power engines of many developing countries, optimal strategic enhancements can feasibly ignite unexplored potentials with the emerging 4th Industrial Revolution technologies. With the rapid technological infrastructural developments, cloud computing has powered up lots of e-commerce businesses that enhance distribution of textile products. The problem is that there is no feedback channel for local and global customers to share their experience with the textile product developers and distributors, and this limits the marketing potentials of the textile producers and distributors locally and globally. In this work, we propose and demonstrate an explainable data-driven solution that optimizes 4th industrial Artificial Intelligence potentials and Natural Language Processing to boost textile product development and distribution. Our proposed Artificial Intelligence approach to sentiment analysis transparently explains evolving customer sentiments and preferences for textile products and provides insights for textile product development, personalized marketing, and distribution of textile garments for women.

Digitalization made the world remarkably agile; therefore, plenty of news is published on a related topic in different newspapers. An automatic text summarizer can contain all the primary information on a particular topic to understand the main essence of news or an article quickly and efficiently. We have developed a novel extractive-based multi-document summarizer for Bengali news. A very few summarizers have been introduced for the Bangla language. To the best of our knowledge, the word embedding technique has not been used in any of the prior summarizers. The selected sentences of the summaries’ semantic and syntactic relationship are maintained through the utility of word embedding. In this paper, we have used predictive-based embedding, which is continuous bag-of-words-based word2vec algorithm. Since it is a multi-document summarizer, redundant sentences are present in the dataset and are removed by a clustering technique. The proposed model gives a statistically significant result in terms of ROUGE-L F1-measure, which outperforms the baseline system on the same scale. The proposed model achieves state-of-the-art performance and can do a better readable and informative synopsis, which might help the reader gain vital information reliably fast.

In the last few years, an enormous amount of unstructured text documents has been added to the World Wide Web because of the availability of electronics gadgets and increases the usability of the Internet. Using text classification, this large amount of texts are appropriately organized, searched, and manipulated by the high resource language (e.g., English). Nevertheless, till now, it is a so-called issue for low-resource languages (like Bengali). There is no usable research and has conducted on Bengali text classification owing to the lack of standard corpora, shortage of hyperparameters tuning method of text embeddings and insufficiency of embedding model evaluations system (e.g., intrinsic and extrinsic). Text classification performance depends on embedding features, and the best embedding hyperparameter settings can produce the best embedding feature. The embedding model default hyperparameters values are developed for high resource language, and these hyperparameters settings are not well performed for low-resource languages. The low-resource hyperparameters tuning is a crucial task for the text classification domain. This study investigates the influence of embedding hyperparameters on Bengali text classification. The empirical analysis concludes that an automatic embedding hyperparameter tuning (AEHT) with convolutional neural networks (CNNs) attained the maximum text classification accuracy of 95.16 and 86.41% for BARD and IndicNLP datasets.

Feature engineering is treated as a crucial step in any intelligent system pipeline and practitioners often spend 70–80% of their time in this phase before modeling. Especially, in machine learning (ML), the performance of the classification systems deteriorates as irrelevant features are added, even when the features presented contain enough information about the problem. ML researchers often use feature selection and feature extraction techniques individually to filter out the redundant information from the data, but would not get desired results or performance due to choosing a single method only. In this article, we take an attempt to combine both feature selection and feature extraction methods that give an improvement in the prediction performance than using a single feature selection or feature extraction method. In the application of the proposed method, we consider the Indian liver disease patient dataset to classify liver disease using machine learning classifiers. In addition, ensemble classifiers also take into account reproducing and robustifying classification accuracy. The performance is evaluated and compared with different metrics such as accuracy, precision, recall, and F1-score. Using the proposed method, the KNN classifier can diagnosis 76.03% with a precision of 76.69%, recall 96.23%, and 85.36% F1-score which is much better than the individual feature selection and feature extraction approach. The proposed method can also be extended to medical datasets other than liver disease so that these types of deadly diseases can be correctly diagnosed for the betterment of humanity.

Bone is the most important part of our body which holds the whole structure of human body. The long bone situated in the upper arm of human body between the shoulder and elbow junction is known as “Humerus”. Humerus works as a structural support of the muscles and arms in the upper body which helps in the movement of the hand and elbow. Therefore, any fracture in humerus disrupts our daily lives. The manual fracture detection process where the doctors detect the fracture by analyzing X-ray images is quite time consuming and also error prone. Therefore, we have introduced an automated system to diagnose humerus fracture in an efficient way. In this study, we have focused on deep learning algorithm for fracture detection. In this purpose at first, 1266 X-ray images of humerus bone including fractured and non-fractured have been collected from a publicly available dataset called “MURA”. As a deep learning model has been used here, data augmentation has been applied to increase the dataset for reducing over-fitting problem. Finally, all the images are passed through CNN model to train the images and classify the fractured and non-fractured bone. Moreover, different pretrained model has also been applied in our dataset to find out the best accuracy. After implementation, it is observed that our model shows the best accuracy which is 80% training accuracy and 78% testing accuracy comparing with other models.

Obesity has become a worldwide problem that has rapidly increased with the advancement of technology. This has become one of the most important reasons which reduced life expectancy within the “modern” world. Overweight and obesity tendency continue to increase both in developed and in developing countries. From pediatric to geriatric individuals, it is common in every age group. In different studies around the world, it has been found that overweight and obesity are growing epidemic health concerns. Studies show obesity results in impaired health and premature death. Moreover, during the COVID-19 pandemic, sedentary lifestyle is leading a lot of individuals to obesity. The COVID-19 pandemic also has a negative impact on the mental health of individuals of the affected countries. In this study, we have evaluated 14 different predictive classification models to find the best precision rates to detect the obesity levels and the possibility of being overweight based on the data collected during the epidemic of COVID-19. In the research, apart from finding the best obesity detection model, we also explored the association between obesity and mental health disorders (anxiety and depression) among Bangladeshi people by analyzing their obesity level with their mental health condition, during the ongoing COVID-19 pandemic.

Skin disease is a common health condition of the human body that greatly affects people’s life. Early and accurate disease diagnosis can help the patients in applying timely treatment thus resulting in quick recovery. Recent developments in deep learning-based convolutional neural networks (CNN) have significantly improved the disease classification accuracy. Motivated from that, this study aimed to diagnose two types of skin diseases—Eczema and Psoriasis using deep CNN architectures. Five different state-of-the art CNN architectures have been used and their performance has been analyzed using 10 fold cross validation. A maximum validation accuracy of 97.1% has been achieved by Inception ResNet v2 architecture with Adam optimizer. The performance matrices result imply that, the model performs significantly well to diagnose skin diseases. Additionally, the study demonstrates two approaches for the practical application of the implemented model. (i) Smartphone oriented approach: It integrates the CNN models with the mobile application, (ii) Web server oriented approach: It integrates the CNN model with a web server for real-time skin disease classification.

Diabetic foot ulcers (DFU) are one of the major health complications for people with diabetes. It may cause limb amputation or lead to life-threatening situations if not detected and treated properly at an early stage. A diabetic patient has a 15–25% chance of developing DFU at a later stage in his or her life if proper foot care is not taken. Because of these high-risk factors, patients with diabetes need to have regular checkups and medications which cause a huge financial burden on both the patients and their families. Hence, the necessity of a cost-effective, re-mote, and fitting DFU diagnosis technique is imminent. This paper presents a convolutional neural network (CNN)-based approach for the automated detection of diabetic foot ulcers from the pictures of a patient’s feet. ResNet50 is used as the backbone of the Faster R-CNN which performed better than the original Faster R-CNN that uses VGG16. A total of 2000 images from the Diabetic Foot Ulcer Grand Challenge 2020 (DFUC2020) dataset have been used for the experiment. The proposed method obtained precision, recall, F1-score, and mean average precision of 77.3%, 89.0%, 82.7%, and 71.3%, respectively, in DFU detection which is better than results obtained by the original Faster R-CNN.

Corona is a super virus that is storming the world, and COVID-19 has already been recognized as one of the deadliest pandemics. Herd immunity is the only solution to stop such epidemic, and a vaccine is the fastest means to reach there. But we have to remember that, it is the vaccination, not the vaccine that can stop the virus. To control the disease, government has to take proper campaign to educate and make people aware about the necessity of taking the vaccine eradicating the misinformation about it. There are different opinions and judgements present among the general people. In our study, we have used social media comments from different posts and news as data to predict users’ sentiment in the context of Bangladeshi demographics. We have scrapped the comments from social media platforms and analyzed their sentiments. We have used KNN, Gaussian, and Naive Bayes classifiers to check which one gives better result. From the study, we could find out that 38.81% people show positive, 27.44% people show negative, and around 33.74% people show neutral sentiment toward the vaccination. The Bangladeshi government has to take proper steps to remove the fear of vaccine before implementing the mass vaccination program as the number of people with positive sentiment is not very high.

Metabolism, gene regulation, and biological processes of the human body happen in molecular pathways, and any disruption in the pathways may directly lead to diseases. Molecular signatures can help get a closer look into cell biology and the mechanisms of human diseases. Therefore, the aim of the research is to discover molecular signatures and pathways which are active unusually in the NPC tissues to outline a few of the essential pathogenesis involved. Thus, genome-wide expression profiling of the NPC tissue is analyzed to identify pathways and distinguish them based on their functionality. Gene-disease association data is used to describe how molecular pathways of nasopharynx cancer are linked to different diseases. By analyzing the gene-disease associations, 392 genes are found out that are common with NPC and other diseases. This study also finds out that neuronal disorders and cancer diseases categories are closely associated with NPC. Protein–protein interaction (PPI) networks which are crucial for understanding cell physiology in both normal and disease states are formed to identify the shared protein groups of various diseases. Overall, this study identifies biomarkers (e.g., hub proteins, TFs, and miRNAs) that regulate gene expression and control important biological processes and molecular pathways of NPC with other diseases. This study can assist further studies to identify the NPC biomarkers and potential drug targets. Determination of the right targets will be helpful for a combined therapeutic approach.

Health monitoring systems in hospitals, clinics, and many other health centers have experienced a significant amount of growth in the recent times. In this current pandemic situation, a lot of people are suffering from COVID-19, lungs disease, chronic disease, and various kinds of common flu, etc. The health of COVID and other common flu affected people as well as elderly people needs continuous and regular monitoring in order to avoid any disastrous situation. It is quite difficult task for health professionals or the patients to be present in person for continuous health monitoring. In this paper, we designed and implemented an IoT-based smart health monitoring system for patients, especially, the elderly, COVID-affected people and patients with chronic diseases. This healthcare monitoring system monitors the body temperature, blood oxygen levels, heart rate, and electrocardiogram (ECG) and uploads the real-time data to an open source Mosquitto (MQTT) server via Wi-Fi or a GSM modem for remote monitoring which can be accessed via a website or a mobile application. This low-cost and efficient device will help both the patient and the health professional in monitoring and diagnose of various kinds of health problems. An emergency alert system has been implemented into the system which will send text message alert to the health expert and the patient itself in case of abnormality in health data. Furthermore, health data analyzing system will help the medical professional in determining the critical patients who need special attention.

The amount of arable land is shrinking in the world as the world’s population is increasing. Producing more crops on less land now holds a significant importance in ensuring food security for this growing population. Soil parameters like soil pH, soil moisture, temperature, and humidity play a significant role in precision agriculture. Wrong land selection, over rainfall, and ignorance about maintaining appropriate soil parameters are the main obstacles in achieving a high yield of crops. A proper low-cost soil parameters monitoring system is yet to be proposed to the best of our knowledge. Therefore, automation of soil parameters monitoring can help in achieving an adequate yield of crops by making appropriate decisions. The objective of this paper is to propose a low-cost soil parameters monitoring system that monitors soil moisture, soil pH, along with environmental temperature wirelessly. Our proposed device collects soil parameters data from the agricultural field and then transfers these data to clients’ device using a wireless network. To evaluate our device, we collect data from different plants and analyze the sensor data.

This paper reports a system called “DIIGMCS—design and implementation of IoT-based greenhouse monitoring and controlling system” designed and implemented for monitoring and controlling the environmental parameters to provide a required environment inside a greenhouse for small plants. The heater bulb, cooling fan, and pump motor are used to adjust the environment inside the greenhouse. IoT is implemented to monitor and control temperature, humidity, and soil moisture remotely by the Website. The project consists of two parts, namely hardware and software. For the hardware part, a system is modeled with several sensors that can measure the value of the moisture from the moisture sensor. The DHT11 sensor contains two sensors; one is temperature, and the other is humidity sensor. The DHT11 sensor gives us both temperature values and humidity values. These values which we get from the sensors, all are values. The proposed system is controlled by an Arduino, which is in turn interfaced with an LCD display as well as a Wi-Fi connection in order to transmit data. The system shows data of sensors over IoT in real time. This system has two different modes such as automatic and server mode to control the environmental parameters inside the greenhouse automatically by sensors threshold value or manually by using an IoT-based Website. Thus, the IoT-based greenhouse monitoring system effectively uses the internet to monitor sensor status and save plants live on time.

Rice is the staple food in Sri Lanka and the widely cultivated crop among farmers. Proper assessment of crop nitrogen and rice crop condition in on-going cultivation helps farmers to effectively manage the crop inputs. Conventional methods of assessing rice crop condition, i.e., crop greenness by human eyes and leaf color chart method, are subjective and have limitations in converting in to quantitative decisions. Thus, assessing the nitrogen level, leaf greenness and leaf chlorophyll concentration has become challenging with traditional methods in Sri Lankan rice industry. This study evaluates the performance of applying remote sensing technique using unmanned aerial vehicle (UAV) for non-destructive, measurement of crop nitrogen level, leaf greenness and chlorophyll concentration. Multispectral UAV images were acquired using multispectral drone from a controlled rice field (Bg 300, at booting stage) in the rice research station with four blocks of treated nitrogen application levels. On-ground measurements were taken from quadrant size (1 m × 1 m) sample areas in each block. From the UAV-derived RGB orthomossaic and reflectance maps, normalized vegetative index (NDVI) was calculated. Finally, the averaged NDVI values extracted from the quadrant areas of the rice crops were compared against the ground measured values using Pearson correlation fit analysis. The results proved that NDVI strongly correlated with leaf chlorophyll, leaf greenness and nitrogen level with R2 = 97%, 96.8% and 96.4%, respectively. Findings strongly suggest the possibility of remotely measuring of nitrogen and chlorophyll level of the rice crop field.

Life in the modern world and technology are inextricably linked. The present era is known as the “Age of Science” because of scientific advancements. The Internet of Things (IoT) is a network of interconnected computing devices, mechanical and digital machinery, and items transporting data without requiring human-to-human or human-to-computer contact. An IoT-based automated sanitization system could be designed to stop the spread of COVID-19, the current global pandemic. In this modern era, combining IoT with alternative power sources to produce something positive for people would dramatically accelerate society. Combining an IoT-based sanitization system with an Earth-Battery, the entire prototype can be powered continuously without relying on an external power source. Earth-Battery is a renewable energy source where the soil cells generate electricity in the presence of water. As a result, it’s also known as a water-activated battery. A pre-programmed NodeMCU will control disinfector machines and automated doors in a prototype automated sanitization system. The fully automated process can be accessed globally using a smart device and the Internet. This research aims to develop an automated sanitization system that might be applied to safeguard human health by cleaning the interior surface of the atmosphere by getting power from the Sustainable Earth-Battery. A creative solution and incredible potential are designed in this paper to address the energy crisis in society and, with the low installation cost, ensure safety against COVID-19.

With the current advancement of IoT and information technology, there are many solutions related to agriculture and soil condition monitoring. It is very important that these solutions need to be tailored as per the area and climate that the farmers are working in. The government of Bangladesh has also initiated many projects with specific goals and objectives to make agriculture more advanced in order to achieve the sustainable development of the country. But, there are limitations to getting efficient outcomes from these projects, as many services are not interconnected yet. This paper demonstrates a model for monitoring the soil condition of farming land, where a farmer can easily know the need for irrigation or fertilizer usage in his field. This model focuses on the affordability of the device for the farmers, and it also minimizes the utilization of devices.

The goal of this project is to create a complete agricultural system that will enable our farmers to use the least expensive technologies. Sensors are utilized for the perception of the environmental conditions encompassing the crop whose outputs are obtained on Associate in nursing humanoid primarily based mobile application likewise as uploaded on the cloud. A record of this information will be maintained that may well be used for the long-run reference, i.e., within the next cropping season, enhancing the development of crop production. This paper shows the prudent utilization of the Internet of Things for typical farming. It demonstrates the utilization of NodeMCU ESP8266-based observed and controlled shrewd water system that is both cost effective and simple. Farmers may irrigate their fields more easily with the help of an automatic irrigation system. A cation concentration detector, water flow detector, temperature locator, and soil condition identifier all exist independently in this savvy water system, and these sensors microcontroller drives the pump motor. Remotely on the Internet through the ESP8266 Wi-Fi unit, the information was obtained by NodeMCU, which then passed. These are scanned using sent data and board IoT. This empowers the far-off instrument through a safe web association with the client or user. The actual time values and reference values of several parameters required by crops have been generated for an Internet website. Users can use the Internet service to control water pumps as well as keep an eye on preset reference values that might help farmers enhance productivity with higher-quality crops.

Bangladesh is the land of agriculture, known for growing varieties of crops, also the main source of employment, employing around half of the workforce. Plant diseases could have an impact on farmers’ livelihoods. In order to solve this major concern, a robot is introduced that can detect plant disease using convolutional neural network (CNN), which is a sophisticated image processing algorithm. The one and only method for preventing deficits in agricultural goods quality and amount is to determine actual disease. Plant health monitoring and disease detection are essential for the long agriculture viability. The implemented robot has a solar connection for the battery package, assuring that it has a consistent source of power. The acquired image will be processed, and the classes of the plant are displayed with a local network-based website as an outcome. The website provides detected plant image with the information of disease. Due to the focus on the agriculture, renewable power, and robotics in our work, the robot is named as “AGRENOBOT.”

As a consequence of rising urbanization and industrial growth, water contamination and degradation are developing at an alarming rate. The water scarcity around the world necessitates a long-term strategy to make the most of it. Traditionally, water quality is assessed by collecting water samples by hand and then testing and analyzing them in a laboratory setting. This paper examines how cutting-edge technology, such as the Internet of Things (IoT), can provide a sustainable and cost-effective method of monitoring multiple water parameters in real time. The proposed system was used to calculate the turbidity, TDS, pH level, and temperature of 30 different water samples with success. The turbidity level was measured in nephelometric turbidity units (NTU) and then transmitted via wireless fidelity (Wi-Fi) networks to an Internet of Things—cloud computing platform, where it could be viewed using an Android smartphone or PC. The experiments demonstrated that the monitoring system was capable of continuously monitoring the pH level, total dissolved solids (TDS), and temperature of water from various sources at different times, thereby providing safe water for industrial, agricultural, and commercial purposes. The cost and complexity of implementation are minimal due to the use of sensors and the Arduino Nano microcontroller, making it simple to validate the efficacy of the built system.

The growing world population has placed increased pressure on the agricultural sector on a global scale, and all over the world, efforts are being made to increase food production. Smart farming is an Internet of Things (IoT)-based approach that optimizes productivity in terms of quality and quantity without compromising the farmers’ economical circumstances or adding to their workloads. In this paper, the scope of smart farming has been considered in the perspective of Bangladesh where the Internet coverage is still not very reliable, and majority of field workforce are victims of poverty and illiteracy. Despite such barriers in the realization of smart farming, there have been several public and private projects aimed at gradually transforming the agricultural sector through the adaptation of sensor usage, IoT-based monitoring and satellite tracking. The feasibility of such endeavors has been reviewed in this paper, and the current implementation challenges have been discussed with some suggestions on possible solutions. The effect of such digital agriculture on the economy and the environment has also been linked to multiple Sustainable Development Goals (SDG) defined by the United Nations General Assembly. Toward the end, a conceptual framework for a low-cost smart farming system has been proposed that mainly comprises a number of ESP32 microcontrollers for collecting sensor data, a Raspberry Pi for hosting the database of sensor readings and the web application for viewing them from any device connected to the same network as the Pi. The wireless communication is performed over Wi-Fi, LoRa and GSM protocols.

Alzheimer’s disease is a neurocognitive disease that results from the brain shrinking and brain tissue dying over time. It gradually erodes memory, thinking skills, and the ability to carry out the most basic tasks. The use of an MRI to evaluate brain atrophy is thought to be a reliable way to diagnose and track the progression of Alzheimer’s disease. In such studies, deep learning architecture provides outstanding results. One drawback of deep learning is that it necessitates a large number of datasets to train the model. Another drawback is the black-box nature, and due to this nature, doctors, patients, and the general public are doubtful of the model’s prediction results. To remove these problems, this study proposes a novel Gradient-weighted class activation mapping (Grad-CAM)-based explanation of Alzheimer’s disease prediction, using the EfficientNet-B7 convolutional neural network architecture. Here, the data augmentation technique is used to make the small dataset suitable for the model. In this work, we have performed a classification between different stages of Alzheimer’s disease which are Cognitively Normal (CN), Early Mild Cognitive Impairment (EMCI), Late Mild Cognitive Impairment (LMCI), and Alzheimer’s disease (AD). Using the base dataset and augmented dataset, the achieved accuracy for four-class classification is 73.47% and 91.76%, respectively. The resulting accuracy for AD versus CN, CN versus EMCI, CN versus LMCI, AD versus EMCI, AD versus LMCI, and EMCI versus LMCI using augmented dataset is 97.72%, 97.70%, 98.01%, 95.12%, 96.70%, and 96.40%, respectively.