Automatic Facial Mask Rule Violation Detection using Keras & Tensorflow to Limit COVID-19

COVID-19 is a pandemic disease that spread by itself coming in the contact of people. It was initially started from China and now it has been spread all over the world and many casualties have been occurred. Social distancing commonly known as physical distancing is a non-pharmaceutical approach through which it can be reduced. But social distancing only works when people started wearing mask because it can spread by sneezing even having distance among people. So wearing mask is mandatory to stop spreading this virus at its possible extent. In this paper, it has been intended to identify the people who are wearing mask or not. By the help of CCTV camera it can be recognized at the entrance of various public places such as mall, airport, railway station, mart and many more. If facial mask can be recognized effectively with high level of accuracy then it can become mandatory for people who are violating the rules. The proposed system uses Keras and Tensorflow model for identifying whether people are following the rule or not. Tensorflow is a deep learning methodology through which facial mask can be detected with all kind of situations. Proposed system is able to classify whether a person wear a mask or not, it is also able to identify whether people incorrectly wearing mask i.e. partial wearing. It is mandatory to identify whether people are properly using the mask or not. System identify this kind of situation and classified them accordingly. System uses hybrid technique by combining two algorithms i.e. keras and tensorflow. By combining both the systems it can be identified more precisely to identify the rule violations. Keywords: COVID-19, Facial Mask, Convolutional Neural Network, Classifiers, Machine Learning, Image Processing, Pattern Recognition.

Multimodal transistors as ReLU activation functions in physical neural network classifiers

Artificial neural networks (ANNs) providing sophisticated, power-efficient classification are finding their way into thin-film electronics. Thin-film technologies require robust, layout-efficient devices with facile manufacturability. Here, we show how the multimodal transistor’s (MMT’s) transfer characteristic, with linear dependence in saturation, replicates the rectified linear unit (ReLU) activation function of convolutional ANNs (CNNs). Using MATLAB, we evaluate CNN performance using systematically distorted ReLU functions, then substitute measured and simulated MMT transfer characteristics as proxies for ReLU. High classification accuracy is maintained, despite large variations in geometrical and electrical parameters, as CNNs use the same activation functions for training and classification.

Characterising the Area Under the Curve Loss Function Landscape

One of the most common metrics to evaluate neural network classifiers is the area under the receiver operating characteristic curve (AUC). However, optimisation of the AUC as the loss function during network training is not a standard procedure. Here we compare minimising the cross-entropy (CE) loss and optimising the AUC directly. In particular, we analyse the loss function landscape (LFL) of approximate AUC (appAUC) loss functions to discover the organisation of this solution space. We discuss various surrogates for AUC approximation and show their differences. We find that the characteristics of the appAUC landscape are significantly different from the CE landscape. The approximate AUC loss function improves testing AUC, and the appAUC landscape has substantially more minima, but these minima are less robust, with larger average Hessian eigenvalues. We provide a theoretical foundation to explain these results. To generalise our results, we lastly provide an overview of how the LFL can help to guide loss function analysis and selection.

Boosted Weighted Optimized Convolutional Neural Network Ensemble Classification for Lung Cancer Prediction

One of the major causes of cancer-related mortality worldwide is lung tumors. An earlier prediction of lung tumors is crucial since it may severely increase the death rates. For this reason, genomic profiles have been considered in many advanced microarray technology schemes. Amongst, an Improved Dragonfly optimization Algorithm (IDA) with Boosted Weighted Optimized Neural Network Ensemble Classification (BWONNEC) has been developed which extracts most suitable features and fine-tunes the weights related to the ensemble neural network classifiers. But, its major limitations are the number of learning factors in neural network and computational difficulty. Therefore in this article, a Boosted Weighted Optimized Convolutional Neural Network Ensemble Classification (BWOCNNEC) algorithm is proposed to lessen the number of learning factors and computation cost of neural network. In this algorithm, the boosting weights are combined into the CNN depending on the least square fitness value. Then, the novel weight values are assigned to the features extracted by the IDA. Moreover, these weight values and the chosen features are processed in different CNN structures within the boosted classifier. Further, the best CNN structure in each iteration i.e., CNNs having the least weighted loss is selected and ensemble to predict and diagnose the lung tumors effectively. Finally, the investigational outcomes exhibit that the IDA-BWOCNNEC achieves better prediction efficiency than the existing algorithms.

Zernike Moment Based Classification of Cosmic Ray Candidate Hits from CMOS Sensors

Reliable tools for artefact rejection and signal classification are a must for cosmic ray detection experiments based on CMOS technology. In this paper, we analyse the fitness of several feature-based statistical classifiers for the classification of particle candidate hits in four categories: spots, tracks, worms and artefacts. We use Zernike moments of the image function as feature carriers and propose a preprocessing and denoising scheme to make the feature extraction more efficient. As opposed to convolution neural network classifiers, the feature-based classifiers allow for establishing a connection between features and geometrical properties of candidate hits. Apart from basic classifiers we also consider their ensemble extensions and find these extensions generally better performing than basic versions, with an average recognition accuracy of 88%.

Expert Knowledge Correlated Intrusion Detection System Evaluation Framework for Heterogeneous IoT

The dynamic contexts of heterogeneous Internet of Things (HetIoT) adversely affect the performance of learning-based network intrusion detection systems (NIDS) resulting in increased misclassification rates---necessitating an expert knowledge correlated evaluation framework. The proposed generalizable framework includes intrusion root cause analysis, correlation model, and correlated classification metrics that can be generalized over any NID dataset, corresponding expert knowledge, detection technique, and learning-based algorithm to facilitate context-awareness in reducing false alerts. To achieve this, we perform experimentations on the Bot-IoT dataset---with generalized traffic behaviors from multiple existing NID datasets---employing the Support Vector Machine (SVM) machine learning and Multilayer Perceptron (MLP) shallow neural network classifiers, demonstrating the generalizability, robustness, and improved performance of the propounded framework compared to the existing literature. Furthermore, the proposed framework offers minimal processing overhead on the classifier algorithms.<br>