Search

Accurate wind speed forecasting is a fundamental requirement for advanced and economically viable large-scale wind power integration. The hybridization of the quaternion-valued neural networks and stationary wavelet transform has not been proposed before. In this paper, we propose a novel wind-speed forecasting model that combines the stationary wavelet transform with quaternion-valued neural networks. The proposed model represents wavelet subbands in quaternion vectors, which avoid separating the naturally correlated subbands. The model consists of three main steps. First, the wind speed signal is decomposed using the stationary wavelet transform into sublevels. Second, a quaternion-valued neural network is used to forecast wind speed components in the stationary wavelet domain. Finally, the inverse stationary wavelet transform is applied to estimate the predicted wind speed. In addition, a softplus quaternion variant of the RMSProp learning algorithm is developed and used to improve the performance and convergence speed of the proposed model. The proposed model is tested on wind speed data collected from different sites in China and the United States, and the results demonstrate that it consistently outperforms similar models. In the meteorological terminal aviation routine (METAR) dataset experiment, the proposed wind speed forecasting model reduces the mean absolute error, and root mean squared error of predicted wind speed values by 26.5% and 33%, respectively, in comparison to several existing approaches.

PurposeThis paper aims to identify and prioritize the dimensions that impact employee social sustainability in the airline industry in the United Arab Emirates (UAE).Design/methodology/approachThe five main criteria (employee well-being, communication, management support, reward and control system and training) and 18 sub-criteria were identified from the literature. The sample comprised four experts covering the HR, finance and training functions from a major UAE airline organization. Applying the analytical-hierarchy-process (AHP) methodology resulted in obtaining priority weights for the factors assigned to employee-social-sustainability implementation.FindingsManagement support was found to have the highest priority among the study dimensions impacting employee social sustainability. Surprisingly, reward system was found to be the least important dimension.Research limitations/implicationsThe study was carried out on a single airline organization, limiting the generalizability of the findings. Future studies should be extended to cater to different organizational contexts and varying operational conditions.Practical implicationsThe findings should be of value to human resource management and policymakers in developing countries, such as the UAE, where employee social sustainability should be sought as a means to develop an efficient and sustainable workforce in different industrial sectors.Originality/valueThis study is among the few pioneering studies that focus on employee social sustainability. The use of AHP to prioritize employee-social-sustainability dimensions is also considered pioneering within the field and is anticipated to support future studies, and a deeper understanding, of employee social sustainability.

Colorectal Cancer (CRC) is a leading cause of death around the globe, and therefore, the analysis of tumor micro environment in the CRC WSIs is important for the early detection of CRC. Conventional visual inspection is very time consuming and the process can undergo inaccuracies because of the subject-level assessment. Deep learning has shown promising results in medical image analysis. However, these approaches require a lot of labeling images from medical experts. In this paper, we propose a semi-supervised algorithm for CRC tissue classification. We propose to employ the hypergraph neural network to classify seven different biologically meaningful CRC tissue types. Firstly, image deep features are extracted from input patches using the pre-trained VGG19 model. The hypergraph is then constructed whereby patch-level deep features represent the vertices of hypergraph and hyperedges are assigned using pair-wise euclidean distance. The edges, vertices, and their corresponding patch-level features are passed through a feed-forward neural network to perform tissue classification in a transductive manner. Experiments are performed on an independent CRC tissue classification dataset and compared with existing state-of-the-art methods. Our results reveal that the proposed algorithm outperforms existing methods by achieving an overall accuracy of 95.46% and AvTP of 94.42%.

Analysis of fine needle aspiration cytology FNAC slides of thyroid nodules is a very crucial test before the preoperative diagnosis of thyroid malignancy. Cytology slides may be composed of different types of cells. Differentiating between cancerous cells and healthy cells plays an important role in the treatment. However, the conventional visual inspection is very time consuming and the process might endure inaccuracies because of the subject-level assessment. To the best of our knowledge, no work has been done for the multi-class cell level classification of thyroid nodules. In this paper, we propose a method for classification of cytology images at the cell level by using fine-tuned VGG-19 and AlexNet models, exploiting the transfer learning approach to better fit the model for classification of our dataset. Model evaluations are done by calculating the precision, recall, F1-score, and accuracy. Although the data is highly imbalanced, but both model have shown very good performance by achieving an accuracy of 93.05% and 92.88% by VGG-19 and AlexNet respectively.

Compact signal and image representations are of crucial importance in a variety of application areas. Wavelet and wavelet-like transforms typically divide the frequency plane in a systematic non-adaptive approach. In this paper, we propose a learning-based method for adapting frequency domain tiling using the curvelet transform as the basis algorithm. The optimal tiling that better represents a single image or a given class of images is computed using denoising performance as the cost function. Simulated additive white Gaussian noise is removed from a given image using a thresholding algorithm. The curvelet tiling that generates maximal denoising performance as measured by PSNR or the logarithm of mean squared error (MSE) is considered optimal. The major adaptations considered are the number of scale decompositions, angular decompositions per scale/quadrant, and scale locations. A global optimization algorithm combining the three adaptations is proposed. Signal representations by adaptive curvelets are shown to outperform default curvelets in partial reconstruction error. Furthermore, adaptive curvelets are used in compressed sensing recovery of incomplete seismic datasets and face images. Visual and numerical improvements across a variety of images and different subsampling ratios are reported. Finally, adaptive curvelets denoising performance is tested on seismic datasets. Our results establish clear numerical and visual performance advantages over the default curvelet transform and the non-local means algorithm (NLM). HighlightsWavelet and wavelet-like transforms typically divide the frequency plane in a systematic non-adaptive approach.A learning-based method for adapting frequency domain tilings using curvelets as the basis algorithm is presented.Our results establish numerical and visual performance advantages over the default curvelet transform.

Image retrieval is an important problem in the area of multimedia processing. This paper presents two new curvelet-based algorithms for texture retrieval suitable, which are suitable for use in constrained-memory devices. The developed algorithms are tested on three publicly available texture datasets: CUReT, Mondial-Marmi, and STex-fabric. Our experiments confirm the effectiveness of the proposed system. Furthermore, a weighted version of the proposed retrieval algorithm is proposed, which is shown to achieve promising results in the classification of seismic activities.

In this paper, we propose a method for classifying melanoma images into benign and malignant using Convolutional Neural Networks (CNNs). Having an automated method for melanoma detection will assist dermatologists in the early diagnosis of this type of skin cancer. A regular convolutional network employing a modest number of parameters is used to detect melanoma images. The architecture is used to classify the dataset of the ISBI 2016 challenge in melanoma classification. The dataset was not segmented or cropped prior to classification. The proposed method was then evaluated for accuracy, sensitivity and specificity. Comparisons with the winning entry in the competition demonstrate that one can achieve a performance level comparable to state-of-the-art using standard convolutional neural network architectures that employ a lower number of parameters.

A newly emerged technique known as Digital Rock Physics demonstrates an ability to characterize properties of the porous media. This technique is based on the imaging of rock micro-structure using a micro-CT scanner. Images are segmented based on their grayscale values to extract pore network from the solid phase. Then, rock properties are estimated using extracted pore network and numerical simulations. Porosity and absolute permeability are two essential properties that can be derived from grayscale images. These properties represent storage and flow capacity of the rock. Some rock samples, particularly carbonate rocks have complex micro-structures at several length scales. Due to limited image resolution, 3D images of carbonate rock may not have top-bottom pore connectivity. In this case, one unable to simulate fluid flow throughout the images. Therefore, permeability is computed on small image sub-volumes, where pore connectivity is revealed locally. Such approach requires long simulation runs. In this paper, a new approach is proposed to estimate permeability from 3D carbonate rock images, where pore connectivity is not revealed from top to bottom. In this approach, first a number of texture classes that represents various textures in the 3D image are identified. For each identified texture class, several sub-volumes from the 3D image are extracted. These sub-volumes are representative of the identified textures and have local pore connectivity. To simulate fluid flow through the pore network of extracted sub-volumes Lattice Boltzmann method (LBM) is used. Permeability is then calculated using Darcy's equation. After determining permeability ranges of each textural class, the 3D image is divided into sub-volumes. Then, each sub-volume is classified into one of the specified texture classes using a modified version of the texture classification algorithm, Local Binary Pattern (LBP). In this study, traditional 2D LBP texture feature vector is adopted to handle 3D volume classification. After classification, each sub-volume is assigned a permeability value that is equivalent to its texture class. Finally, the permeability of the whole 3D image is computed by simulating Darcy's flow through the all sub-volumes. Preliminary outcomes of the proposed technique indicate that it is able to closely estimate experimental results from the laboratory.

Sparse representations for signals and images have been used extensively in various image processing tasks. In this work, we use the curvelet transform as a sparsity inducing tool in neural networks. Nowadays, there is much interest in research and development of efficient algorithms that reduce the computational demands of training neural networks. We demonstrate that the compact directional representation generated by the curvelet transform has the potential of significantly reducing computational time and costs. In our proposed classification framework, curvelet coefficients (amplitude and angle) are mapped into the complex domain. Next, a single-layer Complex-Valued Neural Network (CVNN) is used to represent the complex input parameters. After training the network using a suitable training dataset, computed complex weight vectors are used to classify testing datasets. The developed algorithm was tested on the MNIST dataset. Our results indicate a performance that is comparable with state-of-the-art while offering dramatic computational savings.

This paper presents a new method for texture based image retrieval. The proposed algorithm uses a periodically extended variant of the curvelet transform. The sum of the absolute value of differences in the mean and standard deviation between curvelet wedges representing the query image and the test image is used as the distance index. Performance improvement is demonstrated using the CUReT database, where the proposed algorithm significantly outperforms previously proposed methods that were based on Curvelet, Gabor, LBP, and wavelet features. We also show that adapting curvelet tiles increases the performance of the proposed method.

In this paper, we propose a distance measure to evaluate visual similarity between two images. The algorithm searches for adaptive curvelet parameters that better represent a reference image. Next, the distance between the reference image and other images is computed as a weighted sum of distances between histograms of adaptive curvelet coefficients. The algorithm is tested on a data set of exemplary seismic activities. The developed measure is shown to be effective in extracting correct matches from the data set.

Curvelets were recently introduced as a popular extension of wavelets. In the curvelet domain the input image is represented by sets of coefficients representing signal energy in different scales and angular directions. In this paper an algorithm that searches for optimal tilings for use with the curvelet transform is introduced. We consider two adaptations: scale locations, and the number of angular divisions per scale. A search algorithm that searches for the optimal tiling with respect to denoising performance is introduced. Results show significant improvement over original curvelet tilings. Tiling results were also tested with a seismic compressed sensing recovery problem. A similar performance advantage is reported.

In this work we propose using the coefficient of variation as a cost function to improve seismic data representation in the curvelet domain. Performance improvement is demonstrated in denoising and compressed sensing data recovery. The demonstrated approach can be extended to other seismic applications and alternate transforms.

The curvelet transform is a recently introduced non-adaptive multi-scale transform that have gained popularity in the image processing field. In this paper, we study the effect of customized tiling of frequency content in the curvelet transform. Specifically, we investigate the effect of the size of the coarsest level and its relationship to denoising performance. Based on the observed behavior, we introduce an algorithm to automatically choose the optimal number of decompositions. Its performance shows a clear advantage, in denoising applications, when compared to default curvelet decomposition. We also examine how denoising is affected by varying the number of divisions per scale.

Partial Discharge (PD) signal interpretation has become an essential tool for reliable quality assessment of insulation systems in HV electric machines. Automated classification and recognition of the different types of faults occurring within the insulation system is an important open research problem in this field. In this work, we propose a new approach for the interpretation of PD data. A novel segmentation algorithm is developed to segment PD patterns into their main connected components, with the goal of using these segments later on, to extract distinctive features necessary for fault identification and classification. Several segmentation results are reported along with examples illustrating possible usage in fault identification.

Partial discharge (PD) techniques are very powerful engineering diagnosis tools to assess the health of electric cables and insulation of high voltage electric rotating machines. Unfortunately, PD patterns are often corrupted by noise and interference rendering their interpretation difficult and even erroneous in some cases. For that, it is important to remove these interferences from PD patterns. In this paper, we propose a new technique, based on a correlation procedure, to achieve the above purpose. The algorithm was successfully applied on a large number of corrupted PD patterns generated from real-life signals acquired from some high voltage electric machines used by some Abu Dhabi companies.

Purpose of the paper is to present an application of a newly developed intensity based clustering algorithm in separating pulsating signals due to multiple PD sources and noise. It is applied to signals projected into a time-frequency plane which preserve information on the signal shapes thus, signals having different shape are grouped differently in the map. By using the proposed approach, the number and the optimum border between different groups can be obtained and the original Phase Resolved PD pattern can be split into the same number of sub-patterns. The application of the proposed algorithm to data obtained testing stator bars are reported and discussed to show its validity.

In this paper we propose a new density based clustering algorithm. As with other density based clustering algorithms our approach does not require the number of clusters as input. A modification of the Kuwahara filter, used in image processing, is used to generate a special density map in which the brightness of pixels is indicative of the density of the data points. A framework for clustering is derived and its performance is demonstrated on a number of different data sets.

Results of a comparative investigation having the purpose to select the most appropriate Partial Discharge (PD) pattern for robust identification of defect typologies in insulation systems of ac rotating machines, are reported in this paper. PD measurements were performed on some Roebel bars affected by specific defects. The Different PD-patterns were evaluated to verify their capability of identifying defects occurring in different setup conditions, different locations, or by varying voltage levels (robust identification). It was found that among the different PD patterns considered in this investigation, the Phase Resolved PD pattern exhibit features allowing the robust identification of a large number of defects types. The other patterns endured a high degree of ambiguity.

Interference detection and elimination are keys to partial discharge analysis. In this paper we propose a new algorithm for suppression of periodic pulse shaped interference from partial discharge data. After a brief review of interference suppression techniques used in this field, this paper presents the developed algorithm. The visual characteristics that PD measurements exhibit when plotted with respect to the test signal phase angle are exploited to eliminate the pulse shaped interference patterns. In order to distinguish between PD data and interference pulses two features were used. The intensity of charge columns accumulated at every phase angle and the relative scarcity of partial discharges in the neighboring angles. The developed algorithm was tested on sets of PD measurements obtained from various generators used in different oil fields in Abu Dhabi, UAE and the results showed a very good performance in eliminating interference patterns.

The procedure currently used for cochlear implementation requires wide surgical exposure to identify anatomic landmarks. At our institution a minimally invasive technique is being developed that will permit to perform the procedure with a small burr hole. This technique does, however, require identifying pre-operatively a surgical path that reaches the cochlea without touching sensitive structures. This path can be found interactively by localizing a point in the facial recess and another point in the basal turn of the cochlea in the pre-operative CT images. Unfortunately, this is a difficult task because these structures are small and difficult to visualize. As an alternative, outlines of those two structures can be drawn first by a qualified surgeon in one image volume selected as an atlas. This atlas can then be registered to other image volumes to permit automatic localization. In this work a 12 parameter affine registration is performed first using mutual information as a similarity measure. After that a non rigid registration algorithm is applied to register the ear in the atlas to the patient's ear. The structures outlined in the atlas are deformed using the computed transformations and the resulting intensity centroids are used to draw the required safe path. The developed algorithm has been tested on eleven ears. In every instance, the path was deemed acceptable.