This paper presents GIF (Graphics Interchange Format) Steganography. This project is an extension of recently published research paper [26]. The algorithm uses an animated GIF file format to apply a secured variable image partition scheme for data embedding. The secret data could be character text, any Image, an audio file, or a video file converted in the form of bits. The algorithm will use a variable partition scheme with slight modification for data embedding. The proposed method pre-estimates the capacity of the cover GIF image frame. Our method built variable partition blocks in frame separately and incorporate them with randomly selected GIF frames. Then it embeds secret data on appropriate pixel of the GIF frame. Each selected partition block can store a variable number of data-bits that are decided based on block size. The performance of the proposed GIF algorithm experimented and evaluated based on different input parameters, like peak signal-to-noise ratio PSNR and mean square error MSE values.
Recommendation engines have been used nowadays in most electronic portals such as E-Commerce system, Video Streaming Portals, E-Learning system, and etc., recommendation approaches used in E-Commerce Systems are inadequate for E-Learning Systems, since recommendation should consider Learner’s knowledge level, Topic of Interest and Learning path to recommend Learning Objects. Further in E-Learning Systems like Programming Online Judging and Practice platform, complexity of recommended problem should match the Learner’s current ability. In this work, we propose a novel approach to implement practice problem recommendations based on High Utility Sequential Pattern Mining in Programming Online Judging platform. Our approach extracts utility also called as implicit rating based on the learner’s performance derived from the learner submission History to implement High Utility Sequential Pattern Mining and finally it computes recommendation score based on similarity between candidate problem and Learner’s current context to filter the recommendations. We investigate the proposed approach with the dataset taken from real world Programming Online Judging Platform, the experiment shows that our proposed approach outperforms well with good accuracy and coverage.
This research performs a comparison and assessment of different spatial and statistical models to predict the concentrations of the particulate matter (PM10), measured by the environmental stations of the Region of Madrid (Spain) Air Quality Network. To compare the models, the mean error and mean root error were used as variables, which led to discard the Inverse Distance Weighting method. In order to select between the most reliable model, Ordinary Kriging or Empirical Bayesian Kriging, cross-validation statistics were compared. Finally, T-Student analysis was applied to related samples, that delivered significant differences between measurements obtained from the two spatial interpolation methods.
Iris recognition and Diagnosis (IRD) System plays an important role in the diagnosis of iritis which leads to several eye damages and even to fatal end. Despite the significant advances are available for the iris recognition and diagnosis system, efficient and robust design of IRD in practical deployment situations exhibits many performance constraints and yet remains in the darker side of research. Recently, Deep Convolutional Neural (DCN) finds its applications in IRD systems which have shown promising advances towards the better accuracy. For integrating these networks on embedded systems, deep convolutional networks consumes the more computations which is again the significant challenge among the researchers. Hence the good trade-of between the performance and resource is mandatorily needed for an efficient IRD systems. This paper proposes resource efficient and high efficient REB-DeepIRISNets (Resource Efficient Bats- Deep Convolutional Networks) which consist of optimizing the input weights and hidden layers of DCN by the BAT evolutionary algorithms. To attain an accurate and high performance DCN models, three tier architectures has been proposed such as BAT optimization layer, Convolutional layers and Hardware-Software Codesign methodology. The article proposes a new dynamic fixed-point accelerator model and thoroughly illustrates the realization of convolutionary layers on an Zynq-7000 SoC platform co-designed by SW/HW. The proposed architecture has been evaluated using different IRIS datasets and compared with the other existing models such as OSIRIS, WAHET, CAHT, MASEK and FCN in which the proposed model has outperformed existing models in terms of diagnosis rate and resource utilization parameters.
Planning maintenance of facilities is an important role for production management. From preventive maintenance to predictive maintenance, the main purpose is cost down by reducing the chance of the unexpected shot down. Thus, this study intends to propose a novel genetic algorithm (GA)-based independent recurrent neural network (GA-based IndRNN), which is a kind of deep learning technique, and apply it to predict remaining useful life for the ball bearings using vibration signals. The GA is employed to select the important features from 50 extracted features. The result of the proposed method is compared with four methods by the winner of the competition, Sutrisno et al. (2012), Guo et al. (2017), RNN and IndRNN. The experimental results indicate that GA-based IndRNN is able to perform better than the other methods in terms of score.
For induction motor torque control, direct torque control is becoming the industry standard. This paper proposes a switching loss minimization technique for improved Direct Torque Control (DTC) of permanent magnet synchronous motors in order to increase the drive system\'s steady-state and dynamic results. Direct torque control (DTC) of a voltage source inverter-supplied PMSM is a simple scheme that requires little computation time, can be implemented without speed sensors, and is unaffected by parameter variations. In theory, the motor terminal voltages and currents are used to calculate the flux and torque of the motor. A voltage vector is chosen to restrict the flux and torque errors within their flux and torque hysteresis bands based on the instantaneous torque and stator flux magnitude errors, as well as estimates of the flux position. The electromagnetic torque, rotor speed, and stator current of DTC with PMSM and DTC with IM were successfully calculated using Total Harmonic Distortion (THD) in this article. As compared to DTC with IM, DTC with PMSM reduced THD by 12 percent in torque, speed, and stator current [21]. Switching Losses Minimization Technique by THD Minimization is used in this article. Since transistors are only switched when necessary to maintain torque and flux within their hysteresis limits, switching losses are minimised, resulting in increased efficiency and lower losses. Matlab SIMULINK has experimentally confirmed direct torque regulation with PMSM and IM.