Search

A review of studies involving machine learning-based modeling of nanofluid flows with a specific focus on their utilization in compact heat transfer devices for electronic device cooling is reported here. This paper includes separate discussions on studies involving mono-disperse and hybrid nanofluids whenever feasible. The emphasis is on studies published within the last three to five years, revealing a significant increase in research on nanofluid flow modeling using machine learning methods during this time frame. Various machine learning models have been extensively employed to model various nanofluid properties. A significant number of studies have also predicted flow and thermal characteristics for different flow geometries relevant to the cooling of compact electronic devices. It has been observed that a majority of these studies have utilized the data collected from previously published experimental studies, while a few have incorporated accurate theoretical models for data generation. Given the substantial variations in dataset sizes and input parameters considered across such studies, selecting the most accurate machine learning algorithm out of all is highly unlikely. A few studies have also utilized physics-informed machine learning methods for flow field prediction, and those are also covered in this review. It is observed that the investigations utilizing data-driven methods to accelerate or substitute the conventional CFD-based modeling methods for multiphase flows in general, and nanofluid flows in particular, are very scarce in the literature. Insights derived from the review are also presented, and potential future research directions in this field are outlined. [ABSTRACT FROM AUTHOR]

A novel and robust method for heart sound activity detection (HSAD) is presented. In this method, a new discriminative feature is developed based on the lag-1 autocorrelation coefficient and feature smoothing is introduced to identify the endpoints of low-level heart murmurs. Several experiments on a large scale phonocardiogram database show that the method significantly outperforms the other HSAD methods under varying levels of heart sounds and different types of noise. For a signal-to-noise ratio of 5 dB, the method achieves a sensitivity of 99.56% and a positive predictivity of 99.41% and has a smallest average time error value of 4.26 ms. [ABSTRACT FROM AUTHOR]

This document is a retraction note for an article titled "Effect of Legendre-Fenchel denoising and SVD-based dimensionality reduction algorithm on hyperspectral image classification." The article has been retracted by the Editor-in-Chief and the publisher due to concerns about compromised editorial handling and peer review process, inappropriate references, and not being in scope of the journal or guest-edited issue. The authors R. Reshma and V. Sowmya disagree with the retraction, while K. P. Soman has not responded to correspondence regarding the retraction. The publisher, Springer Nature, remains neutral in regards to jurisdictional claims and institutional affiliations. [Extracted from the article]

Around the world, lung disease is a prevalent cause of death and illness. In this article, we propose a lung disease detection system for the automated identification of critical lung diseases: 1. Tuberculosis 2. Regular Pneumonia 3. COVID Pneumonia and healthy individuals from chest X-ray (CXR) images. The proposed system incorporates a two-stage classification method using an ensemble of convolutional neural networks (CNNs). We evaluated the performance of the proposed system using various evaluation metrics with a test set containing 200 normal, 100 Tuberculosis, 200 Regular Pneumonia, and 100 COVID Pneumonia (a total of 600) CXR images. The proposed system achieves an average macro precision, recall, and F1-score of 0.99 and an accuracy of 0.99. The performance of the proposed system based on F1-Score (0.99) for three classes, namely Tuberculosis, Pneumonia, and COVID, is better than the following existing classification frameworks: ensemble of K-ResNets for the same three classes (F1-Score: 0.84), modified ResNet50 (F1-Score:0.89), KNN classifier with fractional multi-channel exponent moments (F1-Score: 0.89), VGG-16 (F1-Score: 0.89), and convolutional sparse support estimator network (F1-Score for COVID: 0.78). The proposed system can also determine the confidence associated with the challenging classification of Regular Vs. COVID Pneumonia. We also utilized visualization techniques like Grad-CAM and Saliency maps to analyze the features learned by the CNNs, to create transparency in the prediction. [ABSTRACT FROM AUTHOR]

Convolutional neural network (CNN)-based deep learning architectures are the state-of-the-art in image-based pattern recognition applications. The receptive filter fields in convolutional layers are learned from training data patterns automatically during classifier learning. There are number of well-defined, well-studied and proven filters in the literature that can extract informative content from the input patterns. This paper focuses on utilizing scattering transform-based wavelet filters as the first-layer convolutional filters in CNN architecture. The scattering networks are generated by a series of scattering transform operations. The scattering coefficients generated in first few layers are effective in capturing the dominant energy contained in the input data patterns. The present work aims at replacing the first-layer convolutional feature maps in CNN architecture with scattering feature maps. This architecture is equivalent to utilizing scattering wavelet filters as the first-layer receptive fields in CNN architecture. The proposed hybrid CNN architecture experiments the Malayalam handwritten character recognition which is one of the challenging multi-class classification problems. The initial studies confirm that the proposed hybrid CNN architecture based on scattering feature maps could perform better than the equivalent self-learning architecture of CNN on handwritten character recognition problems. [ABSTRACT FROM AUTHOR]

Scattering convolution network generates stable feature representation by applying a sequence wavelet decomposition operation on input signals. The feature representation in higher layers of the network builds a large-dimensional feature vector, which is often undesirable in most of the applications. Dimension reduction techniques can be applied on these higher-dimensional feature descriptors to produce an informative representation. In this paper, singular value decomposition is applied to the higher-layer scattering representation to generate informative feature descriptors. The effectiveness of the reduced scattering representation is evaluated on Malayalam printed and handwritten character recognition using support vector machine classifier. The reduced scattering representation improves the recognition performance when combining with lower-layer scattering network features. [ABSTRACT FROM AUTHOR]

This paper describes the importance of performing preprocessing techniques namely, denoising and dimensionality reduction to the hyperspectral data before classification. The two main problems faced in hyperspectral image processing are noise and higher dimension. Legendre-Fenchel transformation denoises each band in the data while preserving the edge information. To overcome the issue of high data volume, inter-band block correlation coefficient technique followed by singular value decomposition and QR decomposition is utilized to reduce the dimension of hyperspectral image without affecting the critical information. The preprocessed data are classified using kernel-based libraries, namely GURLS and LibSVM. Performance of these techniques is evaluated with accuracy assessment measures. The experiment was performed on five datasets. Experimental analysis shows that the proposed denoising technique increases the classification accuracy. In the case of Indian Pines data, with 10% of the training data, the classification accuracy is improved from 83.5 to 97.3%. And also, dimensionality reduction technique gives good classification accuracy even with 50% reduction in the number of bands. The classification accuracy of the Salinas-A and Pavia University data is 99.4 and 94.6% with the 50% dimensionally reduced (100 and 50 bands, respectively) number of bands. The bands extracted by the dimensionality reduction technique using the denoised hyperspectral data differ from that of the hyperspectral data without denoising. This emphasizes the importance of denoising the dataset before applying dimensionality reduction technique. In case of Pavia University, the band numbers above 50 (out of 100 bands) which were not informative bands before denoising are selected as informative bands by dimension reduction technique after denoising. [ABSTRACT FROM AUTHOR]

Remote sensing satellites are proficient in taking earth images across various regions in visible part of electromagnetic spectrum. The images can be panchromatic image of a single band, multispectral image of three to seven different bands, and hyperspectral image taken from about 220 contiguous spectral bands. These images are used together or on its own, depending on the significance and usage of the preferred application. Pan-sharpening is one method which is used to improve the quality of a low resolution multispectral image by fusion with a high resolution panchromatic image. This paper proposes a method based on M-band wavelets for the pan-sharpening of a low resolution multispectral image. The method tries to improve the spatial characteristics while preserving the spectral quality of the data. The proposed technique uses weighted fusion rule and average fusion rule. The data used for the experiment were acquired by high resolution optical imagers onboard QuickBird, WorldView-3, WorldView-2 and GeoEye-1. A comparison with existing fusion techniques is done based on image quality metrics and visual interpretation. The experimental results and analysis suggests that the proposed pan-sharpening technique outperforms other compared pre-existing pan-sharpening methods. [ABSTRACT FROM AUTHOR]

Pneumonia is an acute respiratory infection caused by bacteria, viruses, or fungi and has become very common in children ranging from 1 to 5 years of age. Common symptoms of pneumonia include difficulty breathing due to inflamed or pus and fluid-filled alveoli. The United Nations Children's Fund reports nearly 800,000 deaths in children due to pneumonia. Delayed diagnosis and overpriced tests are the prime reason for the high mortality rate, especially in underdeveloped countries. A time and cost-efficient diagnosis tool: Chest X-rays, was thus accepted as the standard diagnostic test for pediatric pneumonia. However, the lower radiation levels for diagnosis in children make the task much more onerous and time-consuming. The mentioned challenges initiate the need for a computer-aided detection model that is instantaneous and accurate. Our work proposes a stacked ensemble learning of deep learning-based features for pediatric pneumonia classification. The extracted features from the global average pooling layer of the fine-tuned Xception model pretrained on ImageNet weights are sent to the Kernel Principal Component Analysis for dimensionality reduction. The dimensionally reduced features are further trained and validated on the stacking classifier. The stacking classifier consists of two stages; the first stage uses the Random-Forest classifier, K-Nearest Neighbors, Logistic Regression, XGB classifier, Support Vector Classifier (SVC), Nu-SVC, and MLP classifier. The second stage operates on Logistic Regression using the first stage predictions for the final classification with Stratified K-fold cross-validation to prevent overfitting. The model was tested on the publicly available pediatric pneumonia dataset, achieving an accuracy of 98.3%, precision of 99.29%, recall of 98.36%, F1-score of 98.83%, and an AUC score of 98.24%. The performance shows its reliability for real-time deployment in assisting radiologists and physicians. [ABSTRACT FROM AUTHOR]

Cybercriminals use domain generation algorithms (DGAs) to prevent their servers from being potentially blacklisted or shut down. Existing reverse engineering techniques for DGA detection is labor intensive, extremely time-consuming, prone to human errors, and have significant limitations. Hence, an automated real-time technique with a high detection rate is warranted in such applications. In this article, we present a novel technique to detect randomly generated domain names and domain name system (DNS) homograph attacks without the need for any reverse engineering or using nonexistent domain (NXDomain) inspection using deep learning. We provide an extensive evaluation of our model over four large, real-world, publicly available datasets. We further investigate the robustness of our model against three different adversarial attacks: DeepDGA, CharBot, and MaskDGA. Our evaluation demonstrates that our method is effectively able to identify DNS homograph attacks and DGAs and also is resilient to common evading cyberattacks. Promising results show that our approach provides a more effective detection rate with an accuracy of 0.99. Additionally, the performance of our model is compared against the most popular deep learning architectures. Our findings highlight the essential need for more robust detection models to counter adversarial learning. [ABSTRACT FROM AUTHOR]