Search

In this paper, we propose and analyze SPARQ-SGD, an event-triggered and compressed algorithm for decentralized training of large-scale machine learning models over a graph. Each node can locally compute a condition (event) which triggers a communication where quantized and sparsified local model parameters are sent. In SPARQ-SGD, each node first takes a fixed number of local gradient steps and then checks if the model parameters have significantly changed compared to its last update; it communicates further compressed model parameters only when there is a significant change, as specified by a (design) criterion. We prove that SPARQ-SGD converges as $O\left( {\frac{1}{{nT}}} \right)$ and $O\left( {\frac{1}{{\sqrt {nT} }}} \right)$ in the strongly-convex and non-convex settings, respectively, demonstrating that aggressive compression, including event-triggered communication, model sparsification and quantization does not affect the overall convergence rate compared to uncompressed decentralized training; thereby theoretically yielding communication efficiency for ‘free’. We evaluate SPARQ-SGD over real datasets to demonstrate significant savings in communication bits over the state-of-the-art.

Redheaded Agama lizards attack their prey in a well-organized manner. This work models the dynamic foraging behaviour of Agama lizards and their effective way of capturing prey into a mathematical model named as artificial lizard search optimization (ALSO) algorithm. The idea is based on a recent study in which the researchers reported that the lizards control the swing of their tails in a measured manner to redirect angular momentum from their bodies to their tails, stabilizing body attitude in the sagittal plane. A balanced lumping (between body and tail angles) plays a significant role in capturing the prey in a shot. In formulating the optimization problem, a swarm of lizard are considered that are hunting for the prey. To study the performance of the proposed ALSO, it has been simulated. A comparative study is done with some well-known nature-inspired optimization techniques on classical unimodal, multimodal and other benchmark functions. Further, the algorithm is also tested on an object detection application. The result proves the effectiveness of the proposed ALSO algorithm over other nature-inspired state of the art.

Purpose The purpose of this paper is to based upon the joining and characterization (mechanical and metallurgical) of ferritic stainless steel (SS)-430 using a microwave hybrid heating (MHH) phenomenon. Design/methodology/approach The preliminary experiments were conducted using nickel-based powder as interface material using a domestic microwave oven at a frequency of 2.45 GHz and 900 W power for 720 s. The processed joint was metallurgically characterized by means of X-ray diffraction, Energy-dispersive X-ray spectroscopy and Field emission scanning electron microscopy. Mechanical characterization was done by means of tensile and Vickers’ microhardness testing to check the hardness and strength of the joint. Findings The metallurgical study revealed that the microstructure and formation of numerous phases of Fe2Si3 accompanied by chromium and nickel carbides. The average hardness of 359 Hv at the center of the joint and 637 Hv around the boundaries of the joint was observed. The tensile strength of the joint was observed to 471 MPa with an elongation of 9.02%. The worn surface of the joint signifies the presence of plastic deformation and it was limited due to the presence of harder phases such as Ni3Si and Ni3C. Research limitations/implications The concept of microwave joining of metals is a very challenging task as the temperature can not be controlled in the inert atmosphere of the microwave. It has been also observed that at certain elevated temperatures, the material starts absorbing the microwaves; which is unknown. So, a more intensive study is required to overcome these kinds of limitations. Practical implications MHH technique can be used to process different materials such as ceramics, composites and polymers. SS-430 joined by microwave heating is highly corrosion resistive and has wide applications in refrigerators cabinet panels, dishwasher linings, chimney liners, lashing wires, etc. Originality/value As of the author’s best knowledge, no work has been reported on the joining of SS-430 which has huge potential in the industries. Also, no work has been reported on the characterization of microwave joined SS-430.

Salient object detection in a real-time environment demands high accuracy with less computation time. It is a changeling task to investigate a saliency model which improves saliency detection accuracy as well as reduces computation time. In this paper, we propose a hybrid model that improves detection accuracy with low computational time. The input image is simplified and clustered at multiple scales using SLIC and K-means clustering algorithms respectively. Gaussian Mixture Model (GMM) is developed on various color components of the digital image at multiple scales. The parameters of GMM are learnt using Expectation Maximization (EM) algorithm. The spatial variance of each color component is determined using GMM parameters and hence object position is estimated. Further, spatial variance of color components and object position is exploited to compute saliency map at a scale level. Afterwards, all the saliency maps generated across various scales are linearly combined to produce the final saliency map. The performance of the proposed model is compared in terms of Precision, Recall, F-Measure, Area under the Curve (AUC), Receiver Operating Characteristics (ROC) and Mean Absolute Error (MAE). Extensive experiments on six publicly available datasets viz. MSRA10K, DUT-OMRON, ECSSD, PASCAL-S, SED2, and THUR15K show that the proposed model outperforms or comparable against 11 state-of-the-art methods of the last decade. The key features of the proposed method are object completeness and efficiency in terms of computational time.

Sign language is the art of communicating using hand gestures. The communication between an impaired person and the one who doesn't understand sign language could be easy using the computer as a to...

With the advancement of time, the computer vision systeams are focusing on mimicking the human visual system. In this manuscript, we tried to develop a model which works at improving both the detection accuracy and computation time. First, two double opponent color based features and twelve directional edge features using Gabor filter are computed. Then the most dominant feature pertaining to the salient object is extracted using principal component analysis to form the saliency map. Further, a threshold is applied on the saliency map to detect the salient object present in the image. This threshold selection is a vital procedure. We mapped the Ising model of ferromagnetism to the salient object detection problem by employing an optimization problem for this threshold selection. Experimental results show that the proposed model outperforms the existing models in terms of detection accuracy and also takes less computation time in comparison to many methods.

Dense crowd counting and modeling at different gatherings has ignited a new flame in the visual surveillance research community. There is a high possibility of mishappenings in the form of stampede, mob fighting at these gatherings and the administration is helpless in these scenarios. There is a requirement of analyzing the crowd to prevent these dangerous situations. The proposed work is a case study of Kumbh Mela which models the crowd counting in densely populated images. In the proposed work, the orthographic projection of the crowd is captured using a camera attached to a drone, to reduce the effect of occlusion and scaling which, otherwise, may get introduce during image acquisition process. The captured data is fed to a Convolutional Neural Network for training the model to count head of persons present in the frame. The results obtained from the trained model are validated using geometry and imaging techniques. The proposed model has achieved a mean-absolute-error of 94.3 and a mean-squared-error of 104.6 which seems to outperform the existing state-of-the-art models with respect to the reported performance parameters. The proposed model can be used as a viable solution in applications related to modeling the crowd behavior.

Object detection is an important problem in computer vision attracting researchers from various domains. It deals with separating objects of interest from their background. Image-related low-level features are used to detect salient objects from images, however these features vary from image to image. There is no single feature that can clearly identify salient objects in all the images. Rather, a linear combination of these features may be useful for detection of salient objects. The computation of optimal weights of these features in a learning algorithm is an NP-Hard problem and requires an approximation of an optimization problem. So in the first phase, three low-level features capturing the edge and color information are extracted from the image. In the second phase, these features are integrated to form the saliency map. The linear weights are used for the combination which guides the dominance of feature(s). Thereafter, a threshold is applied over the saliency map to extract the salient objects. We use a linear weight vector and threshold that play a vital role in increasing the detection accuracy of the model. Identification of these parameters can be mapped to an optimization problem to identify the optimal linear weights best suited for the image. The problem is solved using an improved environmental adaptation method. There are three contributions of the paper. First being the identification of weight vector so that important feature(s) for an image are selected where the salient object is highlighted. The second is posing the salient object detection as an optimization problem with a relevant fitness function. Finally the third is utilization of Environment Adaptation Method (EAM) to solve the optimization problem pointed in the paper. The model is extensively validated over six complex datasets and achieves good results on standard performance measures used in comparison to twenty six related works. Four variants of the optimization algorithm are also presented. The EAM optimizer outperforms the GA, PSO and GWO optimizers.

The development of a small wrinkle defect in a buried pipeline can be a challenging issue for pipeline operators to approach. Although extensive research has been conducted to evaluate the ...

Wrinkle defects can be complex pipeline deformities to assess and can present the potential to initiate a pipeline release incident as a result of fatigue failure due to pressure cycling, if not dealt with accordingly. Specifically, the stress distribution arising due to applied loads such as internal pressure can vary rapidly due to the complex shape along the wrinkle profile, which may introduce complexities in subsequent assessments such as fatigue life analysis. This paper presents a methodology using numerical simulation for evaluating stress concentration factors of wrinkle defects of varying geometries. A nonlinear finite element model is developed to evaluate stress concentration factors induced by wrinkle defects within steel pipelines subjected to internal pressure. Afterward, data from full-scale laboratory tests for the wrinkled pipe specimens subjected to cyclic pressure fatigue loading are analyzed to evaluate stress concentration factors for comparable wrinkle profiles. Lastly, a comparison between the results of the stress concentration factors evaluated using the finite element method and test data is provided, followed by a brief discussion of potential sources of discrepancies between results obtained from these methods.

In this paper, we propose and analyze SQuARM-SGD, a communication-efficient algorithm for decentralized training of large-scale machine learning models over a network. In SQuARM-SGD, each node performs a fixed number of local SGD steps using Nesterov's momentum and then sends sparsified and quantized updates to its neighbors regulated by a locally computable triggering criterion. We provide convergence guarantees of our algorithm for general (non-convex) and convex smooth objectives, which, to the best of our knowledge, is the first theoretical analysis for compressed decentralized SGD with momentum updates. We show that the convergence rate of SQuARM-SGD matches that of vanilla SGD. We empirically show that including momentum updates in SQuARM-SGD can lead to better test performance than the current state-of-the-art which does not consider momentum updates., Comment: 58 pages, 8 figures

Segmenting tumor automatically in human brain Magnetic Resonance (MR) images is challenging because of uneven, irregular and unstructured size and shape of the tumor. This paper proposes an automated two stage brain tumor segmentation method. In the first stage, coarse estimation of the brain tumor is carried out using convex hull approach. The coarse estimate thus obtained is employed as the initialization for the active contour model applied in the second stage thereby eliminating the need of human intervention. Multiscale Harris energy is estimated at different levels to identify high-energy regions and thereafter constructing convex hull over the selected key-points in order to detect the abnormality in the input MR images. The proposed method is applied to 2-d axial images of fluid attenuated inversion recovery (FLAIR) and post-contrast T1-weighted (T1c) MRI images from the brain tumor segmentation benchmark challenge 2015 (BRATS2015) dataset. Different sub-compartments of the tumor such as enhanced tumor, edema, and necrosis are segmented and in addition combined in the form of tumor core, complete tumor, and enhanced tumor labels. The proposed method is evaluated in terms of Dice Similarity Coefficient (DSC), Sensitivity and Positive Predictive Value (PPV). Average DSC score of 81% for brain tumor core, 92% for complete brain tumor, and 83% for enhanced brain tumor are achieved which is better than several state-of-the-art methods.

The present work investigates the microstructural changes in an AISI 304L austenitic stainless steel during the early stages of tensile deformation (where austenite does not transform to strain induced martensite). In situ tensile experiments were conducted to record grain orientation changes and slip activation in the steel. The effect of grain size, neighboring grains, and annealing twins on orientation changes during deformation was investigated. Results showed that at a given strain level, grains lying in relatively softer regions and possessing higher Schmid factor values accommodated the plastic deformation initially and showed orientation changes toward the stable orientation. The relatively larger grains changed their orientations only at higher strain levels. Grain orientation changes were also influenced by size and crystallographic orientation of neighboring grains. For grains containing annealing twins, the orientation changes of twin and its grain were in different directions during deformation at a given strain level. Further, grains containing multiple twins showed delayed deformation. The study of tensile deformation behavior in this respect opens up new routes to alter and hence enhance the mechanical properties of materials by engineering their microstructure.

Amplification through hybrid optical amplifiers (HOAs) is a propitious and proficient technology for high speed and high capacity dense wavelength-division-multiplexing (DWDM) systems. HOAs are intended to improve system reach and to accomplish wide gain bandwidth with enhanced flatness of gain. In this work, an ultradense 16 channel WDM system is demonstrated and performance of diverse hybrid amplifiers is evaluated in terms of output power, Q-factor, gain flatness and BER. Spacing among the WDM channels is 25 GHz in order to make system bandwidth efficient and scrutinized its effect on four wave mixing (FWM) in case of EDFA-EDFA, Raman-EDFA and SOA-EDFA. It is observed that SOA-EDFA is more and Raman-EDFA is less prone to FWM. Moreover, for distance 20 Km–140 Km, Raman-EDFA is optimal configuration for amplification and from 150 Km–200 Km, SOA-EDFA shows better performance. For prolonged link lengths such as beyond 200 Km, EDFA-EDFA is a right hybrid amplifier. In order to achieve maximum gain flatness in proposed architecture, EDFA-EDFA is recommended to use.

Power production of a wind turbine is dependent upon its rotor size and at present wind turbines with large rotor diameter (>175 m) are available in the market. However major problems associated with such large size conventional turbines are their cost & noise pollution. Due to these reason researchers have diverted their attention towards lower sized equivalent multi-rotor wind turbines. These turbines are found to be cheaper and good performers. Keeping it in view, in this paper an effort has been made to compare the energy yield and economics of two types of wind turbines i.e. single rotor & multi rotor wind turbine. Power, energy and cost models as proposed are used to determine the annual energy yield and economics of multi-rotor turbines. Simulation results as presented in this paper justify the suitability of multi-rotor wind turbine in place of single rotor configuration. Such turbines deliver more energy yield with low installation cost in contrast to single rotor turbines.

Wavelength division multiplexed passive optical is promising technique to achieve a high data rate and large number of user. The notable advantages of WDM PON is the combination of reliability, cheap in cost, accessible bandwidth, high security, large optical reach and it can support large number of ONU. There are multiple approaches to achieve high-speed WDN PON using different transmission techniques. In WDM, multiple lasers are required which increase the cost of the system. To reduce cost, an optical multicarrier generation system is proposed. An economical multiple carrier generation with the incorporation of sine generator and Mach–Zehndar modulator is demonstrated. Utmost work of sine generator and dual drive modulator was to attain low cost functioning of passive optical networks. Multicarrier generation was done and replacement of laser carriers with optical multicarrier generator. Carriers were generated with the frequency spacing of 20 GHz and these carriers were used in the passive optical networks with the tone-to-noise ratio of 40 dB, amplitude difference of 1.4 dB. For the transmission of downstream in the PON, differential phase shift keying was employed at 10 Gbps data speed. Transmission distance achieved was 30 km using single-mode fiber and this was a part of optical distribution network. Optical network unit was next part after ODN and signals were received with balanced receiver. Moreover, half signal was given to intensity modulator for the signal re-modulation. Bit error rate of 10–9 was achieved at all channels in the downstream. An upstream of 10 Gbps was accomplished in the passive optical network.

The objective of the present chapter is to highlight applications of machine learning and artificial intelligence (AI) in clinical diagnosis of neurodevelopmental disorders. The proposed approach aims at recognizing behavioral traits and other cognitive aspects. The availability of numerous data and high processing power, such as graphic processing units (GPUs) or cloud computing, enabled the study of micro-patterns hundreds of times faster compared to manual analysis. AI, being a new technological breakthrough, enables study of human behavior patterns, which are hidden in millions of micro-patterns originating from human actions, reactions, and gestures. The chapter will also focus on the challenges in existing machine learning techniques and the best possible solution addressing those problems. In the future, more AI-based expert systems can enhance the accuracy of the diagnosis and prognosis process.

Salient object detection in wavelet domain has recently begun to attract researchers’ effort due to its desired ability to provide multi-scale analysis of an image simultaneously in both frequency and spatial domains. The proposed algorithm exploits the inherent multi-scale structure of the double-density dual-tree complex-oriented wavelet transform (DDDTCWT) to decompose each input image into four approximate sub-band images and 32 high-pass detailed sub-band images at each scale. These 32 detailed high-pass sub-bands at each scale are adequate to represent singularities of any geometric object with high precision and to mimic zooming-in and zooming-out process of human vision system. In the proposed model, we first compute a rough segmented saliency map (RSSM) by fusing multi-scale edge-to-texture features generated from DDDTCWT with segmentation results obtained from bipartite graph partitioning-based segmentation approach. Then, each pixel in RSSM is categorized into either background region or salient region based on a threshold. Finally, the pixels of the two regions are considered as samples to be drawn from a multivariate kernel function whose parameters are estimated using expectation maximization algorithm, to generate a saliency map. The performance of the proposed model is evaluated in terms of precision, recall, F-measure, area under the ROC curve and computation time using six publicly available image datasets. Extensive experimental results on six benchmark datasets demonstrate that the proposed model outperformed the existing 29 state-of-the-art methods in terms of F-measure on all five datasets, recall on four datasets and area under ROC curve on two datasets. In terms of mean recall value, mean F-measure value and mean AUC value on all six datasets, the proposed method outperforms all state-of-the-art methods. The proposed method also takes comparatively less computation time in comparison with many existing spatial domain methods.

Integrating electric vehicles (EVs) into the smart grid can support various services for the power grid through the vehicle-to-grid (V2G) system. In this study, the effects of critical peaks (CPs) on EVs' charge/discharge process (CDP) while providing V2G services are critically investigated. A charge/discharge optimisation algorithm for EVs while considering varying charging costs and discharging incentives is proposed. Primarily, a probability model for the occurrence of CPs is formulated and incorporated in a time-of-use tariff plan. Considering the battery capacity loss in a CDP, an optimisation model is developed based on a non-linear programming model. An optimisation algorithm is proposed to enhance the EVs' CDP. The goal is to obtain the least possible charging cost per day while facilitating the V2G services, especially in case of CPs. The effects of CPs on the per day charging cost while considering real-life scenarios are investigated. Furthermore, the dependence of the energy discharged by the EV on the number of estimated battery cycle life and the per day charging cost considering battery replacement is analysed.

We have prepared nanocomposite films of reduced graphene oxide (rGO) and nickel cobalt oxide quantum dots (NCO-QDs) by chemical method, and these are represented by NCO-rGO. Different nanocomposite films have been formed by varying the ratio of NCO-QDs and rGO. The obtained value of the specific capacitance of the nanocomposite (NCO5-rGO1.5) is 265 F g−1 (260 C g−1) at 0.73 A g−1 which is much higher than rGO alone (96 F g−1 at 0.33 A g−1). As the asymmetric/hybrid configuration also enhances the potential window, so in this work, we have fabricated an asymmetric supercapacitor with NCO5-rGO1.5 as the positive electrode and rGO as a negative electrode using 1 M Na2SO4, neutral aqueous solution, as the electrolyte. The assembled supercapacitor delivered the specific capacitance value of 131 F g−1 (197 C g−1) at an energy density of 47 W h kg−1 and a power density of 631 W kg−1 within a wide potential window of 1.6 V. The potential window of the device can be extended up to 2.4 V, but in that case, its coulombic efficiency decreases. The device retains a specific capacitance of 69% after 1000 cycles of charge-discharge. Also, a solid-state device has been fabricated, which displays a potential of 3.4 V after charging 20 s. All these results may prove fruitful for the design of high energy density supercapacitors in the future.

The huge amount of data generated by devices, vehicles, buildings, the power grid, and many other connected things, coupled with increased rates of data transmission, constitute the big data challenge. Among many areas associated with the Internet of Things, smart grid and electric vehilces have their share of this challenge by being both producers and consumers (ie., prosumers) of big data. Electric vehicls can significantly help smart cities to become greener by reducing emissions of the transportation sector and play an important role in green smart cities. In this article, we first survey the data analytics techniques used for handling the big data of smart grid and electric vehicles. The data generated by electric vehicles come from sources that vary from sensors to trip logs. Once this vast amount of data are analyzed using big data techniques, they can be used to develop policies for siting charging stations, developing smart charging algorithms, solving energy efficiency issues, evaluating the capacity of power distribution systems to handle extra charging loads, and finally, determining the market value for the services provided by electric vehicles (i.e., vehicle-to-grid opportunities). This article provides a comprehensive overview of the data analytics landscape on the electric vehicle integration to green smart cities. It serves as a roadmap to the future data analytics needs and solutions for electric vehicle integration to smart cities.

The Biometrics field has gained much concern in the last few years as it is an effective alternative to ancient authentication systems like passwords etc. Authentication system is needed when it is necessary to verify if a user is who they claim to be. Due to increase in traits as a result of enhancement in number of users which in turn affect the performance of database and authentication systems. The biometric systems based on single trait are on demand due to its ease of use and expedited process to access the biometric features of individuals. However, due to advent of undeniable safety factors, biometric systems are more prone to these threats. Eventually, it is the need of the hour to develop multimodal security systems which can provide highest degree security along with full accuracy and efficiency. In this Research paper a prototype for the same has been proposed. The primary focus is to design a multimodal Biometric system based on teeth, speech and signature recognition. Moreover, the highest value achieved will be treated as perfect match.

Salient object detection is one of the outstanding capabilities of the human visual system (HVS). The researcher community aims at developing a salient object detection model that matches the detection accuracy as well as computation time taken by the humans. These models can be developed in either spatial domain or frequency domain. Spatial domain models provide good detection accuracy at the cost of high computational time while frequency domain models offer fast computational speed to meet real-time requirements at the cost of poor detection accuracy. In order to induce a trade-off between computational time and accuracy, we propose a model which provides high detection accuracy without taking much of computation time. To detect the salient object with an accurate shape, we first segment the given image by utilizing a bipartite graph partitioning approach which aggregates multi-layer superpixels in a principled and effective manner. Second, the saliency of each segmented region is computed based on a hypercomplex Fourier transform (HFT) saliency map reconstructed using amplitude spectrum, filtered at an appropriate scale chosen using statistical features extracted from grey- level co-occurrence matrix and original phase spectrum. Finally, a saliency map is generated by taking average of the HFT coefficients of each region in the segmented image and then using the average HFT intensity value of the entire image as a threshold to clearly separate salient object from the background. The performance of the proposed model is evaluated in terms of F---measure, area under curve (AUC), and computation time using six publicly available image datasets. Both qualitative and quantitative evaluations on six publicly available datasets demonstrate the robustness and efficiency of the proposed model against twenty popular state-of-the-art methods.

The capability of humans in distinguishing salient objects from background is at par excellence. The researchers are yet to develop a model that matches the detection accuracy as well as computation time taken by the humans. In this paper we attempted to improve the detection accuracy without capitalizing much of computation time. The model utilizes the fact that maximal amount of information is present at the corners and edges of an object in the image. Firstly the keypoints are extracted from the image by using multi-scale Harris and multi-scale Gabor functions. Then the image is roughly segmented into two regions: a salient region and a background region, by constructing a convex hull over these keypoints. Finally the pixels of the two regions are considered as samples to be drawn from a multivariate kernel function whose parameters are estimated using expectation maximization algorithm, to yield a saliency map. The performance of the proposed model is evaluated in terms of precision, recall, F-measure, area under curve and computation time using six publicly available image datasets. Experimental results demonstrate that the proposed model outperformed the existing state-of-the-art methods in terms of recall, F-measure and area under curve on all the six datasets, and precision on four datasets. The proposed method also takes comparatively less computation time in comparison to many existing methods.

Salient object detection is one of the challenging problems in the field of computer vision. Most of the models use a center prior to detect salient objects. They give more weightage to the objects which are present near the center of the image and less weightage to the ones near the corners of the image. But there may be images in which object is placed near the image corner. In order to handle such situation, we propose a position prior based on the combined effect of the rule of thirds and the image center. In this paper, we first segment the image into an optimal number of clusters using Davies-Bouldin index. Then the pixels in these clusters are used as samples to build the Gaussian mixture model whose parameters are refined using Expectation-Maximization algorithm. Thereafter the spatial saliency of the clusters is computed based on the proposed position prior and then combined into a saliency map. The performance is evaluated both qualitatively and quantitatively on six publicly available datasets. Experimental results demonstrate that the proposed model outperforms the seventeen existing state-of-the-art methods in terms of F ---measure and area under curve on all the six datasets.

The synthesis of reduced graphene oxide (rGO) involves the reduction of graphene oxide (GO). GO is usually synthesized by Hummer’s method that includes extensive use of concentrated acid and specific reaction conditions. To omit this, we developed a greener and safer alternative for rGO synthesis. Electrochemical etching of carbon rods was done in the lemon juice solution. We prepared four different samples to see the effect of different operating voltages on the quality of rGO. Raman analysis and elemental analysis were carried out to investigate the variation in the synthesized rGO samples because of different applied voltages. X-ray diffraction spectra were analyzed by the Debye-Scherrer equation to perceive crystallite size variations and d-spacing. Electron microscopy displayed rGO sheets with folded edges. UV–Visible absorption spectra confirmed the rGO formation and upsurge in the number of layers with an increase in voltage. Comparative studies confirmed the best quality rGO produced at 10 ​V bias voltage. This process may replace other rGO production techniques for its chemical-free and hassle less synthesis approach with the possibility of further improvements.

The suitability of an energy source for automotive application is determined by its gravimetric and volumetric energy density, portability, and safety characteristics. Hydrocarbon fuels predominate the energy resource in transportation sector mainly because of the favorable energy density and easiness for fuel handling, storage and transportation. The alternative energy sources have started to play an important role because they have a lower carbon footprint and the potential to decrease the reliance on the fossil fuels. Among the leading contenders, batteries and fuel cells draw the most attention in recent years because of their potential towards renewable mobility. In this paper, the energy density of various energy sources is reviewed, and its effect on the payload and driving range of the vehicles is discussed. The specific energy of hydrocarbon fuels (from fossil fuels to renewable fuels), fuel cells and batteries (from theoretical value to packaged value) is compared. The evolution of the energy density of the batteries is also investigated and the challenges towards the further enhancement of the battery energy density are presented. Present application status of the three energy sources is summarized, and the suitable energy source for various transportation sectors is suggested. Finally, the present status of hybrid vehicles and renewable hydrocarbon fuels with lower carbon footprint is discussed, which shows promise to overcome the environmental impact without sacrificing overall vehicle performances.

In this multimedia era, the education system is going to adopt the video technologies, i.e., video lectures, e-class room, virtual classroom, etc. In order to manage the content of the audiovisual lectures, we require a huge storage space and more time to access. Such content may not be accessed in real time. In this work, we propose a novel key frame extraction technique to summarize the video lectures so that a reader can get the critical information in real time. The qualitative, as well as quantitative measurement, is done for comparing the performances of our proposed model and state-of-the-art models. Experimental results on two benchmark datasets with various duration of videos indicate that our key-lecture technique outperforms the existing previous models with the best F-measure and Recall.

With brisk growth in video data the demand for both effective and powerful methods for video summarization is also elevated so that the users can browse the apace and comprehend a large amount of video content. Our paper highlights a novel keyframe extraction technique based on the clustering to attain video summarization (VS). Clustering is an unsupervised procedure and these algorithms rely on some prior assumptions to define subgroups in the given dataset. To extract the keyframes in a video, a procedure called k-medoids clustering is used. To find the number of optimal clusters, is a challenge here. This task can be achieved by using the cluster validation procedure, Calinski–Harabasz index (CH index). This procedure is based on the well-defined criteria for clustering that enable the selection of an optimal parameter value to get the best partition results for the dataset. Thus, CH index allows users a parameter independent VS approach to select the keyframes in video without bringing down the further computational cost. The quantitative and qualitative evaluation and the computational complexity are drained to compare the achievements of our proposed model with the state-of-the-art models. The experimental results on two standard datasets having various categories of videos indicate PICS model outperforms other existing models with best F-measure.

In this current age of digital photography, the digital information is expanding exponentially. The use of such information in fields like research, automation, etc. has experienced a rise over the last decade. Also, employing machines to automate any task has been performed since forever. This leads to extensive use of the machine in solving the task of understanding the digital information called computer vision. Machine learning has always played an eminent role in various computer vision challenges. But, with the emergence of deep learning, machines are now outperforming humans. This has led to exaggerate the use of such deep learning techniques like convolutional neural network (CNN) in almost every machine vision task. In this paper, a new technique is proposed that could be used in place of CNN for solving elementary computer vision problems. The work uses the ability of the spatial transformer networks (STN) to effectively extract the spatial information from an input. Such an information is invariant and could be used as input to more plain neural networks like artificial neural network (ANN) without performance being compromised.

Automation in detecting and segmenting brain tumor is the need of the era in order to diagnose human brain magnetic resonance images (MRIs) and required for better treatment planning as compared to the manual process. Manual diagnosis of brain tumor MRI is a time-consuming process and often depends on the expertise of the clinician or radiologist which may lead to a chance of human error. However, automatic brain tumor detection has been a complex task in medical image analysis due to unknown, unstructured nature of abnormality and a huge variability in shape, location, and characteristics of different sub-compartments of the tumor. In this paper, we propose a fully automatic model for brain tumor detection based on parameter-free clustering algorithm and morphological dilation and hole-filling operations. The method is applied to an axial slice of the T1c modality of BRATS 2015 training dataset. In our experiments, we segmented the tumor from the contrast-enhanced T1-weighted image and compared the results with the available ground truth. Results of tumor segmentation achieved of 75% of the Dice similarity coefficient (DSC) for a tumor core region when compared to the ground truth.

In this work, we propose an unsupervised learning network model based on Self-Organizing Map (SOM) to detect and summarize the events in multi-view surveillance videos. Action recognition in multi-view surveillance videos can be a primary element of a multimedia surveillance systems. The basic elements: feature extraction, structuring, mapping, and visualization are counted by our proposed model while video skimming as action recognition of the activities are subjected as the event detection and summarization. The model can work better for real-time applications, for e.g., surveillance, security systems, etc. The qualitative as well as quantitative assessment is done in order to compare the performances of our proposed SOM-based summarization technique and state-of-the-art models. Experimental results on three benchmark data-sets with various types of videos indicate that the proposed method outperforms the state-of-the-art models with the best F-measure.

The recognition of player in the soccer game is a challenging task due to the continuous variation in the position, view, orientation and distance of the players as well as the camera. In order to resolve this problem, the player's identification number (which is printed on their jersey) is recognized and based on the number, players are identified. This paper proposes a noble Optical Character Recognition (OCR) system that automatically extracts and recognizes the player's jersey number in the image. A hierarchical computational framework is introduced which uses high and low-level vision features of the image to identify the player. While following the hierarchical approach a parsing tree is constructed by segmenting the image into snippets based on different algorithms. The entire parsing tree is divided into 4 levels: Level 1 deals with the playfield detection and shot classification, Level 2 tackles the player localization and extraction, Level 3 covers the player's number extraction and shape restoration, and Level 4 recognizes the player's number. The proposed OCR system has consistently shown outstanding results in terms of efficiency, effectiveness, and robustness over a large soccer dataset.

In this paper, we introduce a fast and novel biologically plausible frequency domain approach to detect salient object which incorporates both local and global salient features. The proposed approach involves three phases. In the first phase, locally salient features are obtained as suggested in the research work of Bian and Zhang. Globally salient features are computed in the second phase using fast Walsh-Hadamard transform since it is computationally more efficient and faster than fast Fourier transform. Finally the saliency map is generated in terms of linear weighted combination of local and global saliency maps where the weights are determined using entropy measure. The performance is evaluated both qualitatively and quantitatively on two publicly available datasets and one new dataset derived from a publicly available dataset. Experiments show that the proposed model significantly outperforms other relevant existing state-of-the-art methods in both spatial and frequency domain. The proposed method is also computationally less expensive to detect salient object accurately.

The interminable need of number of channels in communication systems escorts more number of wavelengths to be used. This broad spectrum of wavelengths when expands in both C and L band, the need of hybrid optical amplifiers (HOA) comes into play. In this work, we proposed an automatic gain-controlled HOA using Residual pumping for C/L band. The desired gain is fixed at 27 dB at pump power swinging between 300 and 400 mW. The input powers are automatically adjusted so as to provide invariable and flat gain eliminating the wavelength dependency of optical amplifier to great extents. Analysis is carried out in a 32 channel WDM system and power transients are also measured in terms of powers of surviving channels and excursion amplitudes by adding and dropping up to 28 channels. The output is adjudicated in terms of maximum gain achieved, OSNR, BER, gain flatness, noise figure and power transient excursion amplitudes.

In this computer era, the most interesting thing is to determine the human emotions with machines. Human relies on emotion for conveying their response to others. With advancement in the field of artificial intelligence, machines are also learning to understand these human emotions. With the evolution of deep learning in the field of computer vision, computers are used for solving problems like object detection, motion recognition, anomaly detection, and video surveillance. With our work, we tried to employ same deep learning methods to tackle the problem of emotion detection. Some of the previous works explored the capabilities of deep learning and achieved the efficiency like never before. Therefore, to make the interaction between human and computer more realistic we trained the machine to understand non-verbal communication in form of emotions. In this paper, we used a novel approach to detect the emotion based on the lips structure over the period of time. The use of the recurrent neural network to analyze the pattern with time provided the classification of emotions into 6 different classes. The qualitative, as well as quantitative evaluation, is done in order to compare the performances of our proposed model and state-of-the-art models. D-FES can perform in the real-time environment for accurate tracking and classification of emotions.

Human visual system has the ability to detect objects of interest in real time with a very high accuracy. It becomes an arduous task for a machine to mimic the same. There is a tradeoff between detection accuracy and computation time, while detecting salient objects. The authors made an attempt to develop a model which in least amount of time could produce better detection accuracy. In the proposed, SLIC algorithm is used to decompose an image into superpixels which are used to generate a color spatial feature, which is further combined with the center based position prior to form the saliency map. The proposed model is matched with seventeen related models on six publicly available datasets. The experimental section shows that the proposed model generates the best results when compared in terms of area under curve, recall, precision and F-measure on all the six datasets. It is also better among many models in terms of computation time.

The increasing use of ad blocking software poses a major threat for publishers in loss of online ad revenue, and for advertisers in the loss of audience. Major publishers have adopted various anti-ad blocking strategies such as denial of access to website content and asking users to subscribe to paid ad-free versions. However, publishers are unsure about the true impact of these strategies [2, 3]. We posit that the real problem lies in the measurement of effectiveness because the existing methods compare metrics after implementation of such strategies with that of metrics just before implementation, making them error prone due to sampling bias. The errors arise due to differences in group compositions across before and after periods, as well as differences in time-period selection for the before measurement. We propose a novel algorithmic method which modifies the difference-in-differences approach to address the sampling bias due to differences in time-period selection. Unlike difference-in-differences, we choose the time-period for comparison in an endogenous manner, as well as, exploit differences in ad blocking tendencies among visitors' arriving on the publisher's site to allow cluster specific choice of the control time-period. We evaluate the method on both synthetic data (which we make available) and proprietary real data from an online publisher and find good support.

The rapid growth of video data demands both effective and efficient video summarization methods so that users are allowed to speedily browse and comprehend a large amount of video content. Hence, it is very challenging to store and access such audiovisual information in real time where an immense amount of recorded video content is rising within one second. In this paper we proposed an equal partition based clustering technique for summarizing the events in videos which can work better for real time applications (for e.g., surveillance video in various security systems). In clustering, the difficulty is to obtain the optimal set of clusters, which is gained by implementing Davies-Bouldin Index, a cluster validation technique which permits the users with free parameter based video summarization method for selecting the numbers of key–frames without incurring additional computational cost. The qualitative as well as quantitative evaluation is done in order to compare the performances of our proposed model and state-of-theart models. Experimental results on two benchmark datasets with various types of videos expose that the proposed method outperforms the state-of-the-art models with the best Precision and F–measure.

We have fabricated protein polymer-gold nanoparticle (P-GNP) nanocomposites that exhibit enhanced binding and delivery properties of the small hydrophobic molecule drug, curcumin, to the model breast cancer cell line, MCF-7. These hybrid biomaterials are constructed via in situ GNP templated-synthesis with genetically engineered histidine tags. The P-GNP nanocomposites exhibit enhanced small molecule loading, sustained release and increased uptake by MCF-7 cells. When compared to the proteins polymers alone, the P-GNPs demonstrate a greater than 7-fold increase in curcumin binding, a nearly 50% slower release profile and more than 2-fold increase in cellular uptake of curcumin. These results suggest that P-GNP nanocomposites serve as promising candidates for drug delivery vehicles.