Contrasting to burning fossil fuels for power generation, zero emission H2 fuel cells are a very promising alternative. Toward reaching the optimal fuel cell performance, the conditioning of hydrogen fuel is essential. Hydrogen mixture that contains carbon monoxide will affect the performance of the fuel cell negatively, as CO acts as poison to the Pt electrode. CO cleanup processes are used to solve the issue with various catalysts such as platinum, gold and ceria. In the present work, the microwave prepared ceriumlanthanumcopper oxides catalyst was synthesized with a wide range of Cu percentages (320%). By altering the morphology of the catalyst, the properties (porosity, oxygen storage capacity and particle size) and the performance (selectivity and activity) can be changed as well. The synthesized samples are then characterized using different analytical tools, such as, XRD, thermal studies, SEM and catalytic performance.
The maskless lithography system is designed using UV Laser as a tool for rapid prototyping of microstructures. The GCode read by the system is generated using Inkscape graphics editing software. This software transforms the grayscale image for the intended shape to GCode by tracing the bitmap vectors. The system is optimized to obtain the line width which is provided in the input GCode. Negative photoresist and glass substrate are used to obtain the line feature sample. The system is studied with variation of exposure dosage and the operation path of laser beam by keeping other variables constant (thickness of spin coated layer, prebake time, focus spot, postbake time, and development time). It was observed that controlling exposure produced line width precise to targeted value.
Space adaptive processing (SAP) is a method of target detection that uses a onedimensional adaptive filter called the space filter in the spatial dimension. This method can be used with any desired array geometry. In the conventional method, the targetfree training data is used to build the covariance matrix that is used in determining the spatial filter weights. This conventional method is called the adaptive beamforming method and it suffers from the limited generalization of the angle of arrival of the target between the training and testing stages. Also, it has less immunity to jamming and noise signals compared to the learning based methods. This paper introduces two target detection approaches that exist in the literature and shows their limitations. To overcome these limitations, a learning based technique that involves pattern classification, dictionary learning, and sparse features representation is introduced.
This paper discusses a new algorithm to produce a colored version of gray scale natural still images. This algorithm employs an artificial neural network (ANN) to predict RGB channels using the Discrete Wavelet Transform (DWT). A group of natural color images is used to train three ANNs. The trained networks estimates low resolution RGB layers of the gray scale image which are the best match to the trained images. The colored version of the image is produced from the predicted RGB layers and information from grayscale image. The performances of the new algorithm are analyzed subjectively and objectively using the Peak Signal to Noise and Structural Similarity, as well as it is compared to similar algorithm based on the discrete cosine transform. Acceptable colorized images were obtained from different still images.
Wireless sensor networks emerge at the center of fast expanding Internet of Things revolution and hence research efforts are directed towards optimizing its composition and operational performance. The ability of these networks to sense and monitor a particular target space efficiently and effectively is closely related to the problem of space sensing and communication coverage. Composition of WSN in an open environment and managing a balance between the sensing and communication abilities of individual network nodes is not a trivial problem that needs to be solved to ensure efficient network communication and data transfer. Critical element of this problem is finding the optimal locations of sensor nodes that maximize network coverage subject to the complex communication constraints among the nodes. This paper aims to provide an overview of the related WSN coverage and connectivity problems, and summarize some of the techniques that can compose reliable WSNs at minimum time cost.
Artificial Neural Network (ANN) is a machine learning technique that can be used in classification. In this paper, Feedforward ANN with sigmoid activation function is used to classify Apnea and Hypopnea events from Electrocardiography (ECG) signals. Two ANN will be used, the first one will determine whether the behavior is normal or not. The second NN will work only in case of Apnea or Hypopnea events. According to literature, a lot of challenges have to be considered, especially when it comes to hardware implementation such as the number of neurons, the precision of weights, the required memory space and the processing speed. To detect the Apnea and Hypopnea events, we will use 45:25:1 neural network for the first stage and 27:27:1 for the second neural network with a sigmoid activation function. Additionally, the algorithm has been converted into RTL code and tested using Modelsim.
Early detection of polyps play an essential role for the prevention of colorectal cancer. Manual clinical inspection to detect polyps have many limitations such as the lack of experience of the medical examiner and thus could result to either false or missed polyps. Computer aided diagnosis system has been used as a compliment tool to help the medical expert in the process of polyp detection. Since the computer aided diagnosis systems have been introduced, variety of methods have been proposed in the literature utilizing different types of features and classifiers. Nowadays, there is an increasing evolution of using deep learning methods in image processing because they show improvements compared to the methods used in the literature. This paper utilizes the convolutional neural network (CNN) as a feature extractor and study the effect of varying the database size used for training in the context of automatic polyp detection.
One of the key challenges in estimating the accuracy of existing author identification methods in solving problems that exist in Emirati social media tweets is the lack of evaluation datasets. This paper presents the first Emirati tweets author attribution dataset that we have consructed, the evaluation of the state of the art in author attribution, and a novel text vectorization method that achieved highest classification accuracy relative to other representation methods.
Smallscale devices have been widely growing and developing to the point where they can be used as computers. In this paper, we discuss the file systems, acquisition methods and challenges of six of the main smallscale devices: iPhones, iPads, Android phones, Android tablets, Xbox One and PlayStation 4. In a high level, we critique the file systems and acquisition methods and highlight the advantages and disadvantages of each.
Finite state machines are used extensively in software modeling and testing. One of the important problems of FSMbased testing is the derivation of distinguishing sequences (DSs). Several different methods are given for deriving a DS, and these methods can take a lot of time and processing power as the complexity of the FSM increases. In this paper, we examine the use of SAT solvers for deriving a distinguishing sequence of a deterministic FSM more efficiently. We present the satisfiability problem in general and how to expand that to FSM and the performance comparison using two known SAT Solvers and an exact algorithm.
