Proceedings of International Conference on Frontiers in Computing and Systems

The spike in challenges to security as well as information and resource management across the globe has equally borne the rising demand for a better system and technology to curb it. A news release from the International Committee of the Red Cross (ICRC) in 2020 revealed over 40,000 people were declared missing in Africa. A staggering percentage of that number, a little over 23,000, is documented in Nigeria alone. Despite the numerous factors surrounding missing persons globally, at more than 50% of the original figure, it is unsurprising that most of the cases in Nigeria are attributed to the insurgency and security mishap that has plagued the country for almost a decade. Some of the cases remain unsolved for years, causing the victims to remain untraceable, thereby taking up a different identity and existence, especially if they went missing. Current solutions to find missing persons in Nigeria revolve around word of mouth, media and print announcements, and more recently, social media. These solutions are inefficacious, slow, and do not adequately help find and identify missing persons, especially in situations where time is a determining factor. The use of a facial recognition system with deep learning functionality can help Nigerian law enforcement agencies, and other human rights organizations and friends and families of the missing person speed up the search and find process. Our experimental system combines facial recognition with deep learning using a convoluted neural network. In this study, the authors have used high-standard facial calibration and modeling for feature extraction. These extracted features form the face encodings that are after that compared to a given image.

Protein–protein interaction network (PPIN) plays a major role in information processing and decision making in cells. The PPIN works as a skeleton for cell signaling and functionality. Understanding the flow of information in a cell can enhance our understanding of functional outcomes and flow of signals. To utilize the whole potential of PPIN, we need to direct the edges of the networks. In recent years, a deluge of PPIN became available for analysis but to understand the full picture, these PPINs need to be oriented, or directed, as the direction of signal or information flow in human PPIN is still mostly unknown. In this paper, we propose MultiNet, a method based on the well-known diffusion-based approach which assigns directionality to PPIN created from eight different networks to cover the most of the human genome MultiNet achieves the highest AUC score of 0.94 over protein DNA interaction test set and performs better than the current state-of-the-art algorithms using networks from single sources.

In India, the agriculture sector uses about 84% of the total available water for irrigation. Most of the irrigation is surface-based, and nearly 40% of the irrigation water is wasted due to the unscientific practices adopted by the rural farmers to estimate the actual water demand of their crops. This practice leads to groundwater scarcity and high production costs of the crops. For precise irrigation, the estimation of water needed by a crop is the most significant factor. For proper estimation of irrigation water, the Food and Agriculture Organization recommends the Penman–Monteith method as the most scientific one. However, this method is very complex, involving several region-specific and climatic parameters. It is practically impossible for rural farmers to estimate the daily need for irrigation water using the Penman–Monteith method. As a practical alternative, an automated system is most desirable for rural farmers. This paper presents an irrigation support system to estimate the irrigation water required for a crop correctly. The system has been designed using an assembly of multiple regressors to get the best possible outcome. The outperformed regressor(s) is selected based on the performance evaluated using five standard metrics. This system will benefit rural farmers to estimate the exact quantity of irrigation water needed for a crop. The proposed system is an effective solution to encounter groundwater scarcity and lower the production cost of the crops.

India is a country with rich cultural diversities where festivals keep occurring. In this paper, we introduce the pandal hopping problem where travellers intend to acquire maximum rewards, which may be considered as the satisfaction attained by visiting a pandal. Visiting one pandal from another incurs some cost, but reward is awarded only when a new pandal is visited. The goal is to acquire maximum reward within the associated cost. In this paper, we compare different variations of random walk algorithms such as self-avoiding, exponential, log-normal, quadratic distribution and greedy algorithm. We finally establish that a self-avoiding random walk is best suited to solve such problems.

Structural variants (SVs) such as deletions, duplications, insertions, or inversions are alterations in the human genome that may be linked to the development of human diseases. A wide range of technologies are currently available to detect and analyze SVs, but the restrictions of each of the methods are resulting in lower total accuracy. Over the past few years, the need to develop a reliable computational pipeline has arisen to merge and compare the SVs from various tools to get accurate SVs for downstream analysis. In this study, we performed a detailed analysis of long-read sequencing of the human genome and compared it with short-read sequencing using Illumina technology in terms of the distribution of structural variants (SVs). The SVs were identified in two families with three members each (mother, father, son) using fifteen independent SV callers. Then we utilized ConsensuSV algorithm to merge the results of these SVs callers to identify the reliable list of SVs for each member of the families. Furthermore, we studied the influence of SVs on chromatin interaction-based paired-end tags (PETs). Finally, while we compared the length and number-wise distribution between long-read-based and short-read-based SVs and their respective mapping on PETs. We conclude that SVs detected by our algorithm over sequencing data using ONT are superior compared to Illumina across all SV sizes and lengths, as well as the number of mapped to PETs.

As performance of convolutional neural networks in image classification experiments largely depends on appropriate feature extraction, resizing and/or padding scenarios need to be studied as it affects feature extraction directly. While direct resizing may alter the texture of an image, appropriate padding is not straightforward. In this paper, possible resizing and/or padding scenarios have been empirically studied to have more insights into preprocessing in image classification experiments. It is found that appropriate padding while maintaining the aspect ratio appears to be reasonably prospective than direct resizing. Specifically, padding with estimated background appears to be well-performing in experiments where textures are significant. In all the experiments carried out, resizing with some kind of padding is found to yield the highest classification accuracy. Fairly significant improvement in classification accuracy, i.e., 2–3% in the range of over 90% is obtained with suitable padding.

The world’s largest mangrove region, Sundarban, plays a vital role in sustaining the ecological balance in the Ganga–Brahmaputra deltic region. It spans two neighboring countries, namely India and Bangladesh. In this article, we strive to assess the vegetation health of the Sundarbans region. We rely on the satellite images from Landsat 8 from January 2014 to January 2020. The vegetation health assessment is based on two indices, normalized difference vegetation index (NDVI) and forest canopy density (FCD) analysis of the interest area. NDVI consists of two bands: red and infrared band, showing a negative trend of 0.0085 per year. The image difference technique is employed further to study the vegetation health of the area of interest. NDVI difference marks the coastal regions with a higher depletion rate of vegetation than the regions away from the sea coasts. The histogram of the NDVI difference shows a negative shift from the central zero axis, which is the zone of no change in the vegetation cover. Being an early model to study vegetation health, NDVI has some disadvantages. It tends to quickly saturates at high biomass content and blurs the difference between moderately high plant cover from very high plant cover. So, we have taken another model, FCD, to compare the results of NDVI with it. It highlights the sharp depletion of highly forested cover. Nearly, 80% of the highly forested region has been depleted and joins the medium forested area. The overall mangrove cover is likely to dip more in the ensuing decades if governments of the two countries do not take necessary measures.

In this work, we present our study on the influence of intra-class variations and seed selection on image classification using temporal ensembling. Through our experiments, we observe that for Fashion-MNIST and Kuzushiji-MNIST datasets with medium and high intra-class variations, (a) classification accuracy declines by 20% and 30%, (b) raising seed images by 5x improves accuracy by 15% and 8% respectively. Additionally, preliminary investigation on the Fashion-MNIST dataset reveals that with diversified seed selection class, level accuracy improves by 6%. In due process, our study exhibits convergence issues for unsupervised loss when training on datasets with medium and high intra-class variations.

We present a novel and efficient semantic segmentation architecture termed Light U-Net in this paper. Light U-Net is less complex with regard to the parameter and model size than the original U-Net. We also propose a network and training strategy based on geometric data augmentation for efficient use of the annotated images from small and limited datasets. Following improvements, we show the use of Light U-Net for semantic segmentation of outdoor scenes. Results reveal that our improvements and data augmentation remarkably enhance the proposed network’s performance. Results from experiments demonstrate that the proposed architecture outperforms some recent segmentation models achieving mIoU scores of 70.06% and 79.84% on the Camvid and Stanford Background datasets respectively.

Intelligent transportation is indeed a prominent aspect of smart city development. Intelligent transportation targets to handle several issues by taking into consideration traffic or human mobility, sensory data, and geographical data generated in cities. It brings together the different concepts of urban sensing, data management and analytics along with various service providing mechanisms into a recurrent process for an unobtrusive and continuous improvement of an individual’s transit experience and operations of the city transport authority. In this paper, we first introduce the concept of an intelligent transportation system, along with discussion about the overall framework and key challenges with respect to the perspective of each step of the framework. Second, we classify the data acquisition methods. Third, summarize the typical methodologies needed in intelligent transportation systems for knowledge fusion across heterogeneous sensory data. Finally, we give an outlook on the future of intelligent transportation systems, with suggestion about a few prospective topics of research which have remained unexplored in the research community so far.

In visual question answering (VQA) task, we generally use a convolutional neural network (CNN) to extract image features, a recurrent neural network (RNN) for question representation. Instead of using a RNN for language representation, we use a CNN for question representation. The advantage of using a CNN for question representation is that CNN is more effective in capturing image and question words interactions which are not expressed by a RNN. Thus, in this paper, we have used three CNNs for VQA task. One CNN extracts visual features, second CNN for extracting question features, and third CNN for combining both extracted feature vectors. Further, we have employed a softmax layer to generate answer for a given question. The proposed VQA model is evaluated on DAQUAR, COCO-QA, and VQA2.0 datasets.

Neuromarketing is the field of neuroscience that potentially infers to gain knowledge of customers’ choice by analyzing the physiological signals. Here, the motive is to maintain the sustainability among traditional market research, which depends upon the preference of the consumer explicitly, on the other side neuromarketing, which shows the preference implicitly. Research work in this fastly emerging field is highly demanded due to its inherent potential. But, due to the paucity of sophisticated data-mining approaches for predicting and classifying consumers’ choices, it is not yet touched a gruntled level. Therefore, in this work, a deep convolutional neural network (DCNN) is proposed to understand consumer choice using electroencephalogram (EEG) signals which were recorded while consumers were browsing through various E-commerce products. The arbitrary time-domain signals are transformed into two-dimensional images, which are then used for the classification of the EEG signals using DCNN. The score-level fusion is used to examine the model performance and got 81% accuracy which surpasses the existing approaches.

The semantic segmentation task is the process of predicting each pixel in the image to some class. Classes could be road, person, car, tree, etc. Road scene segmentation has a vast application in the autonomous driving system. Researchers emphasize autonomous vehicles to reduce the human effort of driving and reduce the chances of road accidents caused by human error. Much research has been done to segment and classify objects in the road scene. We humans can easily detect and understand the objects around the road scene, but it is quite a challenging task for any machine. This article presents some state-of-the-art deep learning-based semantic segmentation techniques, namely FCN, SegNet, and UNet. And a new architecture named PNet for road scene semantic segmentation is proposed. The proposed network is trained end-to-end and comparatively evaluated with the state-of-the-art by using segmentation performance matrices mean Intersection over Union (mIoU) and Dice Coefficient measure (DCM). The publicly available CamVid dataset is used to train and validate the proposed PNet. We have achieved 70.7% mIoU and 80.9% DCM, which is comparatively better than the state-of-the-art.

The primary visual cortex consists of varied neuronal structures and connectivity links contributing to complex feature extraction such as edge detection, contour detection, depth and motion detection. But the functional basis of such complex computation is not yet realized. In the proposed work, an attempt has been made to model a single layer of retinal ganglion cell network incorporating morphologically detailed neurons with active dendritic integration to investigate the role of morphology, localized active membrane and connectome specificity. Two different connectome-specific instances of the layer are simulated with one network connected with excitatory connections with ON-bipolar cell behaved as scene segmenter, whereas another network connected with specific connections with ON- and OFF-bipolar cell with excitatory and inhibitory connections, respectively, behaves as an edge detector.

The popularity of the Human-Computer Interaction (HCI) technology exponentially grew with the advancement of artificial intelligences. This work with the help of computer vision will enable the user to interact with the computer at their fingertips. The user would essentially be able to control the cursor and do associative functionalities such as left-click, right-click, move, and almost everything a mouse can do. A cross-platform application is developed which is the underlining vision of replacing a mouse.

An innovative alternative method in the field of security that can enforce safety and security of information has been proposed here by performing audio encryption. Cryptography, a way to secure communications, allows only the sender and the intended recipient of a message to access its contents. Cryptography is derived from the Greek word “Kryptos,” meaning hidden, which is associated with content hiding by encryption. In this paper, we develop a novel time-stamp-based audio encryption method using the mind-cracking number place challenger Sudoku puzzle of higher dimension, i.e., 16 × 16 (popularly known as Super Sudoku) to add a new era in the field of security. In this work, we have developed a Super Sudoku cipher by applying time-stamp in it. Immediately after that, the extracted audio frames of the input audio files are XORed with our developed innovative Cipher to generate encrypted audio frames. Then these encoded audio frames are converted to an encrypted audio file to transmit with safety and security. At the receiver end, by applying the same cipher key, we can extract our original audio file.

Recent few years have witnessed a huge surge in the scientific documents on the Web, and consequently, the systems capable of retrieving mathematical information from such documents have also come into being. All the existing systems adopt a suitable strategy to handle a particular challenge of mathematical information retrieval (MIR). However, it is quite uncommon to find a single system which alone handles multiple unrelated challenges of MIR. This paper, therefore, discusses implementation of a mathematical information retrieval system (MathIRs) which comprises different modules to account for several different unrelated challenges and serves as one-stop for several mathematical information retrieval needs. Evaluation of MathIRs using standard scientific document set and query set and its comparison with other competent systems proves its competence in the field of MIR.

The essential target of this research work is generating a three-dimensional (3D) design of cloud point laser scanner structure that performs computer vision system (CVS) processes intentionally fully capabilities of invest to estimate model structure resulting from laser scanner point clouds and recognizing main areas for further studies and development. This study is significant processes were displayed, begins from collecting the clouds of laser point in static mode, and then analyzing the existing data, an iterative closest point algorithm (ICP) used in point cloud registration and matching, noise reduction, synthesis scheme in laser scanner point clouds uses a random sample consensus algorithm (RANSAC), data generating 3D laser scanner, importing a structure model from laser scanner point clouds and fit in using CVS programs. The Poisson surface algorithm is utilized to pick up point clouds and mesh surfacing. The results displaced some of the structural troubles such as deformation and cracks over two places of the wall. The results of the proposed algorithm matched the original point clouds very well. This differential proved that the algorithm advanced in this research work is efficient and more workable for visualization, monitoring, and future data processing.

Features of the handwritten text play a vital role in the area of handwriting identification. It became more challenging when one has to identify gender, age, and handedness of the person through handwriting. In last two decade, the use of various feature extraction techniques immerged having advantages one on the other. Ample research is done on writer identification systems by implementing various feature extraction techniques. In this paper, we have shifted the concern toward various features and features extraction techniques implemented on gender identification through handwriting. The objective of this survey is to present the critical review of work done in area feature extraction in gender identification taking only handwriting into consideration. We have categorized all the feature extraction techniques used by the researchers for gender classification into four broad categories: statistical-, transform-, gradient-, and model-based techniques. From the survey, we have identified few techniques that deserve future attention of the researchers for optimal results.

Automatic identification of rice disease is much important in the agricultural sector as a large number of people in the world lives on rice. Symptoms in rice diseases occur in the leaves, and identifying the diseases by an experienced pathologist through physical observation of infected leaves is a slow process and expensive. Instead, this machine learning approach, especially deep learning-based technique that does not need any pre-processing and feature extraction, can classify the diseases efficiently. Many researchers worked on identifying rice diseases and achieved different results according to the implemented methods. In this paper, we study different deep learning models and transfer learning of the deep learning models in the identification of rice diseases. The rice dataset used in this study consists of 5932 images of four different classes. In addition, the performance of the different deep learning models is evaluated in terms of different parameters such as accuracy, loss, precision, recall, f1-score, and training time.

Vision-based hand gesture recognition is perhaps the most widely used technique for human–computer interaction technology with special application toward sign languages used by differently abled people. Perhaps the most important step in such a gesture recognition system is feature extraction. In this paper, we use a novel fingertip detection mechanism based on convexity defects and use nine geometrical features that are translation and rotation invariant. We use seven different classifiers on two different public hand digit datasets (NTU hand digit dataset and SP-EMD color-depth hand gesture dataset) and find that for both the datasets, the random forest classifier gives the best classification accuracy (94.2% and 90.1%, respectively).

The genetic algorithm (GA) was invented to mimic the realistic biological evaluation for optimization. But, in literature, we can observe that the genetic operators are purely based on either mathematical or random operations. Our proposed genetic operators are designed to mimic realistic behavior. For this purpose, we have introduced the concept of gene casements in crossover and Levy flight function in mutation. We have also incorporated solution combination operation to give it an elitist nature. Otsu’s, Tsallis, and Reyni’s entropic functions were validated as efficient thresholding techniques. Researchers weren’t able to conclude which method is superior among them. Their performance is highly dependent on the dataset. So, we have considered all of these methods to frame our methodology. We have considered the dataset from the WorldView-2 satellite sensor for the experimental study. The performance of the proposed method is validated with three different state-of-the-art methods.

Skeletonization provides a compact yet effective representation of an object using its medial loci defined by Blum’s grassfire transform process. It has been popularly used in many low- as well as high-level imaging applications. Quantitative characterization of object structures is a popular application of skeletonization, where a skeleton serves two purposes—generations of a compact representation of an object and location of representative points for defining local structure morphology. This paper reviews the roles of skeletonization and digital topological analysis in characterizing individual trabecular bone plate-rod microstructure assessing overall bone quality and fracture risk at in vivo imaging.

There has been a huge surge in the production of sensor-based clinical devices for health monitoring systems over the last few years. This sudden spike has been due to many different factors such as development in sensor device technology and also the efforts to promote research that address the necessity for providing new ways for healthcare given the increasing challenges with an associated degree of an aging population. The processing and analysis of data is an important element of the research of such a system. The data generated from these healthcare devices are enormous and have the potential to ascertain well-being and to encourage effective management of health. In this work, a mechanism for the analysis of physiological sensor data from the healthcare devices, namely voting-based extreme learning machine, has been explored. The approach was also compared with the traditional extreme learning machine-based approach. Experimental results were very encouraging with respect to the performance accuracy as well as time taken by the voting-based extreme learning machine as compared to the traditional extreme learning machine to produce the output.

Segmenting the region of interest helps gastroenterologists for removing polyps during the surgery in the gastrointestinal tract. We propose a segmentation system to segment the area of the polyp from the informative frames. Informative frames are the frames that contain at least a polyp in colonoscopy still frames. Our proposed U-shaped convolution neural network model utilizes the concept of residual connection and Xception as a backbone structure. We consider colonoscopy still frames as input to the segmentation model and achieved a Dice and Jaccard score of 86.3 and 79, respectively. It outperformed the conventional U-net with a 3.7% performance gain with respect to Dice score. This proposed method can be used as a reliable alternative system to identify a region of polyps during colonoscopy analysis.

Fine motor skills and working memory skills are determinants of scholastic performances. Educational achievement and high skill performances offer high salaries with comfortable lifestyles in a competitive economy. Presently in the competitive environment, children are often diagnosed with disorders like development coordination disorder (DCD), attention deficit hyperactivity disorder (ADHD), obesity, amblyopia, Tourette syndrome, etc. Children who suffer from the above disorders also lack fine motor and working memory skills. These skills can be improved by providing proper training with various tools and techniques. An Android game is developed in this literature to measure and improve children’s working memory skills and fine motor skills at once. Android is the most widely used mobile operating system, and most handset manufacturers deliver Android operating systems with their smartphones. Due to the coverage of both skills and the use of Android as a gaming platform, this single game can be a cost-effective, time-efficient home-based training tool for improving both skills of children. This improvement may transfer a positive effect on children’s scholastic performances.

Iris recognition is one of the most consistent and unique biometric-based recognition systems. Iris-based authentication has been a challenging problem especially for noisy and non-ideal iris images, as the images include realistic noise factors. Presence of noise like reflection of light on eye images, i.e. specular reflection, causes problem in iris recognition. It is one of the major pre-processing tasks in iris recognition system. In this paper, two new and efficient approaches have been proposed and compared their robustness with a few existing specular reflection removal techniques, considering the challenging situation of non-ideal iris images. The first approach uses inward interpolation, and the second approach uses bilinear interpolation with extension to remove reflection. These suggested approaches are faster in execution as well as robust enough to remove reflection without disturbing the texture pattern of original iris images.

Cervical cancer is preventable if one can detect it at an early stage. Due to cervical cancers, the human body imparts various behaviors and seven different determinants of this behavior. Behaviors can be conscious and also subconscious. And also, all expressing behaviors might not be completely correlated. In this work, a strategy has been proposed to exclude non-correlated features to maintain or enhance the detection accuracy of cervical cancer at an early stage. Experimental results show a maximum of 94.4% accuracy, and thus, it proves the efficacy of the proposed work.

A COVID-19 patient suffers from blockage of breathing and chest pain at a critical condition due to the formation of fibrosis in the lungs and needs emergency lifesaving treatment. Before starting an adequate treatment, a confirmed diagnosis of COVID-19 is a mandatory criterion. For a patient with critical respiratory syndrome, rapid and precise diagnosis is a prime challenge. Different manual methods of clinical diagnosis are in practice. However, these manual techniques suffer from serious drawbacks such as poor sensitivity, false negative results, and high turn-around time. The diagnosis based on the radiographic image (X-ray or computed tomography) of infected lungs is another clinical method for rapid diagnosis of COVID-19. However, it requires an expert radiologist for precise diagnosis. Instead of a prolonged clinical process, an alternative way of rapid diagnosis is the only way of some lifesaving. As an elegant solution, some radiographic image-based automated diagnostic systems have been suggested using deep learning techniques. However, they suffer from some unavoidable limitations concerned with deep learning. This paper suggests a user-friendly system for instant diagnosis of COVID-19 using radiographic images of infected lungs of a critical patient. The model is designed based on classical image processing techniques and machine learning techniques that have provided low complexity but a very high accuracy of 98.51%. In this pandemic situation, such a simple and instantaneous diagnostic system can become a silver lining to compensate for the scarcity of expert radiologists.

Breast cancer is the most common cancer(s) among women worldwide. The survival rate decreases if the cancer is not detected at an early stage. Breast ultrasound (BUS) is emerging as a popular modality for breast cancer detection owing to its several advantages over other modalities. We proposed a novel deep learning framework named BUS-Net for automated lesion segmentation in BUS images in this work. However, every deep learning framework has disadvantages of its own; however, the drawbacks associated with individual models can be overcome when combined. Our proposed BUS-Net is an ensemble of three popular deep learning frameworks, namely attention U-net, U-Net and SegNet. The final segmentation map generated by BUS-Net is a pixel-level fusion on the outputs of each of the individual frameworks. The potentiality of BUS-Net was tested on a publicly available dataset named BUSI dataset. This dataset consists of 647 tumor images collected from 600 different female patients. To prevent biased results, the training and test set were separate. BUS-Net framework achieved an accuracy—93.19%, precision—93.18%, recall—88.75%, dice—90.77%, and volume similarity—95.55% for lesion segmentation in the test set. The degree of correlation between the lesion region segmented by the medical experts and that segmented by BUS-Net was high ( $$R^2 = 0.9131$$ R 2 = 0.9131 ). Further, the performance of BUS-Net was also compared with the state-of-the-art techniques. This comparison showed that BUS-Net maintains a tradeoff between precision and recall, proving the robustness, efficiency, and reliability of the framework.

Screening of breast cancer from histology images is a popular research problem in medical imaging. Most of the methods in recent days used deep learning models for predicting the same. But, at times, such methods dealt with not only higher-dimensional features but also may suffer from containing irrelevant and sometimes redundant features. To overcome this shortcoming, in the present work, we employ a popularly used particle swarm optimization (PSO) algorithm to obtain the near-optimal feature set. To extract the features from images we first preprocess the images to obtain stain normalized images and then pass them through a pre-trained MobileNet model for extracting the features. We have evaluated our model on a recent dataset, published through ICIAR BACH 2018 grand challenge. The experimental results show an improvement of 6.25% recognition accuracy with around 54% reduced features. We have also compared our result with two state-of-the-art CNN models: InceptionResNet and DenseNet, and we found that the use of MobileNet is better. The capability of the present model is comparable with some state-of-the-art methods on the BACH dataset.

In this paper, we propose a stochastic model based on cellular automata and graphs to explore the spread of infectious viruses (like SARS-CoV-2) in closed rooms. We also present a simulator implementing this model which allows studying how different policies affect the spread of viruses. As we show, the simulator can be used to explore scenarios in various points of interest (POIs) like shops, public trams or fitness centres. It could be useful for policymakers to check (by changing the parameters of the simulations) the effectiveness of different regulations like limiting the maximum occupancy of POIs and mandating the usage of face masks to decrease the spread of aerosols. The simulator can also be used to compare the hazard level that different kinds of POIs pose. Also, the simulations can be visualised and showed to the public to increase support for the introduced measures and obedience to restrictions.

A computer-aided diagnosis (CAD) system can be helpful for the detection of malignant tumors in the breast. Ultrasound imaging is a type modality with low cost and lower health risk. In this paper, we have classified benign and malignant breast tumors from ultrasound images. We have used the image quality assessment approach for this purpose. No-reference image quality metrics have been used as features for the classification task. We have used a public database of ultrasound images of breast tumors containing 780 images. The classification of breast ultrasound images using image quality assessment is a very novel approach, producing significant results.

A four-state switchable microstrip filter is presented in this article. A pair of 0° fed uniform impedance resonators coupled with a pair of uniform impedance resonators are used to produce dual passband response at 2.4 GHz and 3.5 GHz, respectively. Four pin diodes are employed to switch the dual passband response to two single passband responses and one all stop response. Four transmission zeroes are observed at 2.1 GHz, 2.68 GHz, 3.26 GHz, 3.61 GHz, respectively, that help to get better selectivity and adjacent passbands isolation. In all switchable states, the filter shows an upper stopband response up to 6.0 GHz. The design layout is very compact in size of 0.21 × 0.25 λg at 2.4 GHz.

Cloud radio access network (C-RAN) is popularly known as centralized RAN which is an architecture for 5G network for process and manage cloud computing in a real-time environment. Cloud RAN (C-RAN) is popularly known as centralized RAN for providing flexibility for capital expenditure as well as operational expenditure. The benefits of C-RAN minimize the total cost ownership (TCO) and also improve the network performance. It provides benefit for low-latency network in 5G network as ultra-reliable low-latency communications (uRLLC). The 5G-CRAN enhances the benefits of not requirement of rebuild the transport networks again. C-RAN architecture is an essential and dynamically mapping of remote radio heads (RRHs) with baseband units (BBUs). Otherwise, it will cause call blocking and less quality of network connections. The proposed paper optimize to reduce the blocking of calls and also balance the load of BBUs by applying Swarm Intelligence(SI)algorithms. In proposed work, the simulation results are proved that nature-inspired computing algorithm will reduce the blocked calls and maximize the balance of processing load of BBUs.

The QoS of elastic and inelastic service is badly affected when the network is under a DDoS attack, and minimum bandwidth requirements of these flows are not met. The current work uses game theory to address this issue of optimum allocation of bandwidth to these flows while simultaneously thwarting the DDoS attack. It quantifies the efficiency and effective data rates achieved by these flows as payoff. Similarly, it quantifies the attack payoff using the bandwidth occupancy by attack flows and the relative dropping of the elastic and inelastic flows. Further, the attacker’s effort to compromise and capture a normal machine and turning it into a bot is quantified as the attack cost incurred. The scenario is modelled as a two players: nonzero sum and infinitely repeated game. The proposed mechanism enforces a data rate threshold on the flows using average available bandwidth during an interval of interest to lessen the network congestion at the bottleneck link in a dynamic manner. The corresponding payoffs are computed. Based on these payoffs, the Nash equilibrium at a particular instance of time is analysed using simulations. Subsequently, the Nash strategies are obtained which favour to enforce the optimum value of data rate threshold to maximize the combined payoff or QoS of elastic and inelastic services while making the attack costlier.

This paper analyzes the outage probability of a multihop transmission link with the maximal ratio combining (MRC) diversity technique associated at the intermediate relays and the destination. The exponential and equal correlation models are considered among the input receiving antennas of MRC diversity reception at the relays and end user. The source node is assumed to have a single antenna for transmitting the signal to the node at destination. The source node transmits the signal through decode and forward type of relay nodes to the destination node. The exact-form explanations have been acquired for the outage probability of the system considering the communication through Nakagami-m wireless fading channels with two correlation models. The response of correlation at the outage probability presentation is analyzed. It is noticed that outage performance degrades with an increase in correlation for both cases. The consequence of outage probability for separate stages of fading parameter, diversity order, and the different number of hops has been observed.

Cooperative communication has become more important because it extends the coverage areas of wireless communication, thus decreasing the transmission power of the base station. This paper analyzes the outage probability of decode-and-forward type relaying cooperative protocol over the Fisher–Snedecor F fading channels. A dual-hop transmission link is considered, where there are three nodes, i.e., a source, an intermediate relay, and a destination. The source–relay and relay–destination channels experience Fisher–Snedecor F fading. The outage probability performance in relation to fading parameter and shaping parameter is analyzed. Furthermore, the effects of Rayleigh fading and one-sided Gaussian fading on the outage performance of the communication system are investigated.

In this paper, we derive the outage probability (OP) and average bit error rate (ABER) expression in closed form, considering arbitrary number of branches for selection combining receiver over Nakagami-m channels. Here, we assume wireless information and power transfer (WIPT) system having power splitter (PS) at the receiver side. The power splitter separates the received signal into information transmission and energy harvesting (EH) receiver. The derived expressions of our system consider arbitrary channel parameters and diversity branches. Monte Carlo simulation for the OP and ABER curves with SC receiver is matched with the closed forms. The analytical results correlate with the simulation for different variables of the diversity order, fading parameters, and power splitter factor. This type of system will be beneficial for reliable data transmission in an energy-limited scenario.

The paper presents a compact dual-band reconfigurable filtering antenna. First and foremost, a novel reconfigurable filter with two switchable pass bands is designed. The proposed filter is accomplished by using two open-ended half-wavelength stepped-impedance resonators (SIRs) at the top and two open-ended half-wavelength uniform impedance resonators (UIRs) at the bottom to operate at 3.2 GHz and 5.5 GHz used for WiMax/WLAN applications, respectively. The PIN diodes are connected to the resonators and biased properly to switch the operation from dual-pass band to two single-pass bands and all stop operation. Second, a broadband antenna is proposed using a rectangular and a trapezoidal radiating patch to radiate from 2.5 GHz to 6 GHz. Finally, both the filter and the antenna structure is cascaded together to implement the proposed dual-band reconfigurable filtering antenna. The filtering antenna is simulated, and a well agreement is observed with the predicted response.

A gold nanoparticles (AuNPs)-based device for pregnancy detection form urine sample is described. Human chorionic gonadotropin (hCG) is released by placenta during pregnancy. The gold nanoparticle-based device detects the hormone by changing of its color. The rapid color changes detected by naked eye, and it also confirmed the pregnancy. However, in this method, reaction rate is very fast rather than any conventional method. In this work, we are introducing the application of a new device that mainly contains immobilized gold nanoparticles conjugated with primary antibody and the antigen to measure concentration of hCG from urine sample. The level of hCG for normal women 5mlU/ml, but during pregnancy, it increased up to 25–50 mlU/ml. The line color intensity of hCG at 10 pg/ml tested with device was almost same to the intensity that we measured at 30 pg/ml with the normal device. The same device can be used in future with proper marker for early detection of cancer.

Designed and created by Toyota subsidiary Denso Wave, quick response (QR) code has now been commonly used in many suitable applications to deal with real-time data because it can be read quickly by smartphones or tablets. The covering of information is indeed the technique of keeping secret knowledge inside the source media. Two qualities would have a strong approach to data hiding: high consistency of the stego image as well as high potentiality for concealing. In this paper, a redundant discrete wavelet transform (RDWT) domain-based information hiding method using QR code decomposition is proposed. An appropriate cover is torn down by RDWT mechanism, and the private image is concealed within the sub-bands of the factorized QR code. For the purpose of demonstrating the effectiveness of the proposed approach with respect to imperceptibility, capability and robustness to different signal processing assaults, multiple tests and case analysis are conducted.

Web applications are constantly being developed to make life easier and more convenient for businesses and customers; it makes intruders involved in conducting malicious activities. Intruders use vulnerabilities to perform malicious attacks, and injection is the top-ranked vulnerability of web applications. SQL injection is a technique of code injection that places malicious code through web page input in SQL statements. Several numbers of case studies are found in previous research on vulnerability in the web application layer. Various models are introduced, built, and compared with many current SQL injection models and other vulnerabilities in the web application layer. However, there are few automation detections works on SQL injection that provide high precision and no finite state model-based works. This research aimed to propose a model and develop an automated SQL injection detection tool called ADT-SQLi based on the proposed model. In addition, this work was intended to simulate the proposed model with automata called a finite state machine. ADT-SQLi provides better efficacy on the identification of SQL injection and found ADT-SQLi as a finite model as it has exactly one of a finite number of states at any given time.

With time, humans have realized, the internet is the best alternative for exchange of messages through multimedia documents, especially images. Thus, its imperceptibility and security are big concerns. Recently, Ashraf et al. (Heliyon 6(5):e03771, [1]) evaluated similarity for image steganography using interval type-2 fuzzy logic, but the visual quality (PSNR) is not as good as current demand. To solve the problem, we have proposed a data hiding scheme by combining distinct and vague characteristics of an image through the techniques of superpixel and fuzzy logic respectively. Superpixel and TSK Fuzzy logic model with rule base is used to identify non-uniform regions of the image where secret bits are embedded in coefficients of quantized Discrete Cosine Transform (DCT). This technique can be used for secret data communications. With an average PSNR of 58 dB, the proposed approach provides excellent visual quality. Finally, in order to illustrate the efficacy of our technique, we compared the proposed system to existing methodologies.

Social media applications are an integral part of human life nowadays. Starting from sharing personal information like text and pictures, we now share the latest news and its related photos, question papers, assignments, online surveys, and so many more things. With so much sharing of our data, hackers have found very easy ways to steal our personal information through various social sites. Although different platforms keep coming up with new versions for better security and experience of their users, this breach of personal information demands advances in security protocols to safeguard our data. This has become the basis of this research. In this paper, we compare the security aspects of different social media platforms to see how to fit our favorite social media applications for our users and how many of them keep their promise of providing us better security.

The analysis of button press dynamics (BPD) may reveal more detailed information about decisions in human–computer interaction. One huge advantage of a BPD-specific analysis lies in its tangible nature. More precisely, each button press (BP) constitutes a two-dimensional signal consisting of time and intensity coordinates, which can be easily illustrated. Moreover, one can also interpret BP signals by extracting several intuitive features, such as the duration and the maximum intensity. In this study, we analyse the characteristics of such intuitive BPD features. To this end, we conduct cluster analysis experiments with the following evaluation protocol. First, for each person-specific set of BPs, we will define ground truth (GT) clusterings by evaluating the BPs as time series and applying dynamic time warping (DTW) for the calculation of distances between two instances. Subsequently, we will extract a small set of intuitive features. Finally, we will compute the similarity between the DTW- and feature-based clusterings, based on two popular similarity measures. The outcomes of our experiments lead to the following observation. Extending binary BP information by one additional feature, i.e. the maximum press intensity, can already significantly improve the analysis of BP evaluations.

Dynamic Time Warping (DTW) is a method generally used to align pairs of time series with different lengths, which is for instance applied in speech recognition. In this study, we use a category learning experiment (CLE) as use-case, in which the participants have to learn a specific target from a pool of predefined categories within a certain amount of time. From a companion system-based point of view, it is important to detect certain anomalies related to affective states, such as surprise or frustration, elicited during the course of learning. In this work, we analyse the button press dynamics (BPD) data from an auditory CLE with the goal of detecting anomalies of the aforementioned type. To this end, we first select a small set of participants, for which we have definite ground truth labels. Subsequently, we apply DTW in combination with hierarchical clustering to separate the anomaly-specific data from the remaining samples. We compare the outcomes to clustering results based on the extraction of intuitive task-specific features. Our results indicate that applying the DTW approach in combination with Single Linkage Clustering in order to detect CLE-related anomalies is preferable to its feature extraction-based alternative, in person-independent scenarios.

In this work, VLSI circuit-based artificial neural network has been implemented. The artificial neuron consists of three components such as multiplier, adder and neuron active function circuit, which perform arithmetic operations for realizing neural network. The focus on this work is linearity investigation in nonlinear artificial neural network as well as learning efficiency. A multiplier design has been proposed for reducing nonlinearity in an artificial neuron. Sigmoid circuit has been used for activation function. New weight update technique with both retrieving and on chip learning function has been proposed for better accuracy and improved learning ability. In this neural network, pulse width modulation technique has been used to compute the output signals using both multiplication and summation operations. This improves the linearity of the neural network. The learning operation of the neural network has been verified through simulation results by adopting digital function like “NAND”. A high-speed and accurate error detection block has also been used for this purpose. Cadence Virtuoso has been used to perform circuit level simulation. Design has been done in TSMC 180 nm CMOS technology. In the proposed design, an error calculation time of 0–100 ps has been achieved, thereby making the overall operation fast and the learning efficiency 99%. Supply voltage is 1.8 V, and total power dissipation has been measured to be is 8 mW.

The following research introduced a novel co-planer full adder circuit, executed in quantum dot cellular automata (QCA). The proposed novel full adder circuit was then afterward used with a new circuit for executing a novel 4-digit carry swipe adder (CSA) in QCA innovation. A designer device involving QCA form 2.0.1 is used to carry out the functioning of planned full adder QCA circuits. The execution results demonstrate that a planned QCA FA circuits show improved performance contrasted with other similar circuits.

Through this paper, we have propounded and investigated a novel structure of a grooved trench MOS transistor with double gate architecture using TCAD simulations. To date, only a single metal trench MOSFET has been reported which having weaker control of the gate bias over the channel charge, unlike improved in our propounded structure. The propounded dual metal double gate grooved trench (DMDGGT) structure incorporated the advantages of enhanced gate controllability and subdued the drain-induced barrier lowering (DIBL) owing to the presence of a bi-metal gate with dissimilar work functions. In addition, due to the inherent advantage of a grooved trench gate device of having a longer effective channel length by the trench gate geometry, the device results in a significant reduction in short-channel effects (SCEs). The acquired results from the SILVACO ATLAS simulation exhibit a significant improvement of the propounded DMDGGT MOSFET as compared to its single metal counterpart for surface potential, electric field, threshold voltage, and drain characteristics, thereby substantiating the efficacy of the propounded device structure.

Zinc sulphide (ZnS) nanoparticles are known for their biological sensors. Conjugation of E. coli bacteria to these ZnS nanosamples are done as well as these bacteria are sensed on dried samples based on memristor-based property. We have varied the concentration of bacteria, and as a result, a variation of current occurs like a hysteresis (Figure eight pattern) loop. ZnS is prepared successfully by the chemical precipitation method. Characterization was done by various techniques to confirm its specifics. In this study, the conjugation of E. coli with the antibody-conjugated nanoparticle is confirmed by taking absorbance before and after adding of E. coli in UV–Vis spectroscopy that results in a significant shift in the wavelength of absorption. Electrical characterization leads to a change in voltage gap among different molar concentration of bacterial conjugated ZnS samples. Results show that process basically leads to a mem-mode observation (either memresistive, memcapacitive or meminductive) in nature. Almost orderly declining pattern of voltage gap with bacterial E. coli concentration is obtained for the as-fabricated devices using ZnS quantum dot in bacterial estimation after conjugating antibody E. coli.

Sorting implies the task of presenting a specific type of data in a specific order. These tasks are getting accomplished by different algorithms proposed by researchers. Researchers are trying to achieve this task in minimal space and time complexity with improved stability, correctness, finiteness and effectiveness. Sorting is used in wide range of fields namely, in Operating systems, Data Base Management systems, in searching and in various other data science related areas. In this paper, a divide and conquer approach-based algorithm is proposed to sort the data in a specific order using min–max searching. The time complexity of the proposed Swift Sort algorithm is O(nlogn) and O(n2) in the average and worst cases, respectively. Moreover, time complexity of the proposed algorithm is comparable to Quick Sort, Merge Sort, Heap Sort and TimSort but at the same time Randomised Quick Sort, Merge Sort and Heap Sort produces a better Time Complexity in their worst cases than Swift Sort. The experimental results prove the correctness of the proposed algorithm.

This paper aims to analyze the performance of a speaker independent spoken isolated digit recognition system in standard (Sohra) dialect of Khasi Language. Initially, we have prepared an ideal set of word-based digit pronunciations of 15 digits for the said dialect. The standard approach was used to collect speech data in an open room using Zoom H4N handy portable digital recorder from native language speakers of diverse age groups and gender. Each spoken isolated data has been processed and segmented using a WaveSurfer software tool. Sampling frequency was set at 16 kHz and bit resolution at 16 bits per sample. We extracted the relevant speech feature vectors using Mel-frequency cepstral coefficient (MFCC) and perceptual linear prediction (PLP) that are compatible with the hybrid acoustic models of conventional (Gaussian mixture model-hidden Markov model)-based models (monophone and triphone) and hidden Markov model-deep neural network (HMM-DNN). Using MFCC and PLP features, we found that the HMM-DNN model performs better than GMM-HMM-based models with a word error rate (WER) of 6.67% and 7.40%, respectively.

Coronavirus disease (COVID-19) has affected all walks of human life most adversely, from entertainment to education. The whole world is confronting this deadly virus, and no country in this world remains untouched during this pandemic. From the early days of reporting this virus from many parts of the world, many news videos on the same got uploaded in various online platforms such as YouTube, Dailymotion, and Vimeo. Even though the content of many of those videos was unauthentic, people watched them and expressed their views and opinions as comments. Analysing these comments can unearth the patterns hidden in them to study people’s responses to videos on COVID-19. This paper proposes a sentiment analysis approach on people’s response towards such videos, using text mining and machine learning. This work implements different machine learning algorithms to classify people’s sentiments and also uses text mining principles for finding out several latent themes, from the comments collected from YouTube.

Part-of-speech tagging (POS) is an important and very fundamental process in natural language processing (NLP). POS tagging is required as a preprocessing task in many types of linguistic research such as named entity recognition (NER), word sense disambiguation, information extraction, natural language translation, and sentiment analysis. In this paper, we propose a practical Bengali POS tagger, which takes as input a text written in Bengali and gives a POS tagged output. In recent times, Bi-LSTM networks have been proven effective in sequential data processing but not very much tested on resource-poor and inflectional languages like Bengali. This paper addresses the issues of the POS tagging task for the Bengali language using Bi-LSTM with transfer learning by applying pre-trained word embedding information. The POS tagged output from our proposed model can be used directly for other applications of Bengali language processing as our proposed tagger can also handle out-of-vocabulary (OOV) words. Our experiment reveals that Bi-LSTM with transfer learning is effective for tagging Bengali documents.

In this paper, the temporal phase influence on speech signal is demonstrated through different experimental models, notably for speaker verification. Feature extraction is a fundamental block in a speaker recognition system responsible for obtaining speaker characteristics from speech signal. The commonly used short-term spectral features accentuate the magnitude spectrum while totally removing the phase spectrum. In this paper, the phase spectrum knowledge is extensively extracted and studied along with the magnitude information for speaker verification. The Linear Prediction Cepstral Coefficients (LPCC) are extracted from speech signal temporal phase and its scores are fused with Mel-Frequency Cepstral Coefficients (MFCC) scores. The trained data are modeled using the state-of-art speaker specific Gaussian mixture model (GMM) and GMM-Universal Background Model (GMM-UBM) for both LPCC and MFCC features. The scores are matched using dynamic time warping (DTW). The proposed method is tested on a fixed-pass phrase with a duration of <5 s in a speech signal. The score level fusion technique helps in the reduction of equal error rate (EER) and improves recognition rate.

The spoken form is the most predominant form of communication laden with rich information than its textual counterpart. Development of summarization techniques for spoken data in a resource-poor language is a challenging task, especially due to the unavailability of appropriate datasets. In this paper, we focus on creating a dataset for speech summarization in Assamese, a low-resource language of Northeast India, the genre of speech being broadcast news. The contribution of this paper is twofolds: Firstly, a dataset of the audio samples of the broadcast news comprising of acoustic/prosodic features is developed, where each tuple is the manually segmented utterance representing a sentence. Each utterance is labeled as either summary-worthy or not summary-worthy, in accordance to the handcrafted summaries. Secondly, the relation of the prosodic features in indicating the informativeness of an utterance is evaluated. Binary classification is performed with the help of various classifiers, and a statistical analysis of the performance of the classifiers is presented. This approach for speech summarization is highly relevant for resource-poor language which does not have adequate resources to generate the textual representation of the audio. To the best of our knowledge, this is the first effort to develop a dataset in Assamese language for speech summarization.

In any online social media platform, it is necessary to reduce the effect of rumor data from original information as it may cause harm to society. Influential users can be detected through different centrality measures. When the rumor is generated through some influential users, they have more impact on society. Here, we have proposed a prognostic method to distinguish those influential users of online social media, based on network analysis. The susceptible-infectious-recovered (SIR) model has been used for simulation of the propagation of information. For the particular seed nodes which have been chosen by practicing different centrality measures, detailed relative learning in terms of infected nodes is also exhibited. Selection of seed nodes through centrality measure is computationally exhaustive; therefore, we form a composite model, where the original social network is decomposed using the k-core, and centrality nodes are obtained from that decomposed network. Centrality measurements thus derived from the generated network are used as the seed of information propagation. Another important result derived from the empirical study is that not only the influential nodes, but neighbors of the influential nodes also have a greater impact on maximizing the effect of rumors.