Art Models Research Articles

Vehicle Detection in Challenging Scenes Using CenterNet Based Approach

Contemporarily numerous analysts labored in the field of Vehicle detection which improves Intelligent Transport System (ITS) and reduces road accidents. The major obstacles in automatic detection of tiny vehicles are due to occlusion, environmental conditions, illumination, view angles and variation in size of objects. This research centers on tiny and partially occluded vehicle detection and identification in challenging scene specifically in crowed area. In this paper we present comprehensive methodology of tiny vehicle detection using Deep Neural Networks (DNN) namely CenterNet. Substantially DNN disregards objects that are small in size 5 pixels and more false positives likely to happen in crowded area. Primarily there are two categories of deep learning models single-step and two-step. A single forward pass model is the one in which detection is performed directly to possible location over dense sampling, wherein two-step models incorporated by Region proposals followed by object detection. We in this research scrutinize one-step State of the art (SOTA) model CenteNet as proposed recently with three different feature extractor ResNet-50, HourGlass-104 and ResNet-101 one by one. We train our model on challenging KITTI dataset which outperforms in comparison with SOTA single-step technique MSSD300* which depicts performance improvement by 20.2% mAP and SMOKE by with 13.2% mAP respectively. Effectiveness of CenterNet can be justified through the huge improved performance. The performance of our model is evaluated on KITTI (Karlsruhe Institute of Technology and Toyota Technological Institute) benchmark dataset with different backbones such as ResNet-50 gives 62.3% mAP ResNet-101 82.5% mAP, last but not the least HourGlass-104 outperforms with 98.2% mAP CenterNet-HourGlass-104 achieved high mAP among above mentioned feature extractors. We also compare our model with other SOTA techniques.

From the Dialectical Perspective: Modeling and Exploiting of Hybrid Worm Propagation

The hierarchical network is the more effective platform, which provides multiple channels for various worm propagation. Thus, emerging worms can infect vulnerable hosts by scanning strategy and social media. However, the spread of scan-based worm is restrained due to uneven distribution of vulnerable hosts and NAT (Network Address Translation) technique. Meanwhile, topological dependency dictates to topology-based worm only infecting those hosts in social networks. To avoid their respective disadvantages, modern hybrid worm, which combines the above two propagation mechanisms, can implement efficient IP-address scanning by enhanced combination-scanning strategy, and spread more aggressively in social networks using enhanced reinfection mechanism. This paper presents a Hierarchical-Stochastic Propagation model to understand hybrid worm propagation. Inspired by hybrid worm, we design a new vaccine based on the Hierarchical-Measure Immunization strategy. For physical networking layer, we can estimate vulnerable-host distribution to find vulnerable hosts effectively through Maximum Likelihood estimation. For social networking layer, we use a novel propagation centrality measure to discover vital social nodes accurately. The experimental results show that our model can characterize the propagation mechanism of hybrid worms more comprehensively, and greatly outperforms state of the art models in terms of estimation accuracy. Meanwhile, our strategy is more effective to restrain the hybrid worm from spreading in networks.

Survey on Plant Disease Detection using Deep Learning based Frameworks

Early identification of plant diseases is crucial as they can hinder the growth of their respective species. Although many machine learning models have been utilised for detecting and classifying plant diseases. The advent of deep Learning, a subset of machine learning, has revolutionised this field by offering greater accuracy. Therefore, deep learning has the potential to greatly enhance the accuracy of plant disease detection and classification. Recent research progress on the use of deep learning technology in the identification of crop leaf diseases is reviewed in this article. The current trends and challenges in plant leaf disease detection using advanced imaging techniques and deep learning are presented. This survey aims to provide a valuable resource for the researchers investigating the detection of plant diseases and detection of those using state of the art models for ease of saving time and cost. Additionally, the article also addresses some of the current challenges and issues in the detection process that need to be resolved.

PAACDA: Comprehensive Data Corruption Detection Algorithm

With the advent of technology, data and its analysis are no longer just values and attributes strewn across spreadsheets, they are now seen as a stepping stone to bring about revolution in any significant field. Data corruption can be brought about by a variety of unethical and illegal sources, making it crucial to develop a method that is highly effective to identify and appropriately highlight the various corrupted data existing in the dataset. Detection of corrupted data, as well as recovering data from a corrupted dataset, is a challenging problem. This requires utmost importance and if not addressed at earlier stages may pose problems in later stages of data processing with machine or deep learning algorithms. In the following work we begin by introducing the PAACDA: Proximity based Adamic Adar Corruption Detection Algorithm and consolidating the results whilst particularly accentuating the detection of corrupted data rather than outliers. Current state of the art models, such as Isolation forest, DBSCAN also called “Density-Based Spatial Clustering of Applications with Noise” and others, are reliant on fine-tuning parameters to provide high accuracy and recall, but they also have a significant level of uncertainty when factoring the corrupted data. In the present work, the authors look into the most niche performance issues of several unsupervised learning algorithms for linear and clustered corrupted datasets. Also, a novel PAACDA algorithm is proposed which outperforms other unsupervised learning benchmarks on 15 popular baselines including K-means clustering, Isolation forest and LOF (Local Outlier Factor) with an accuracy of 96.35% for clustered data and 99.04% for linear data. This article also conducts a thorough exploration of the relevant literature from the previously stated perspectives. In this research work, we pinpoint all the shortcomings of the present techniques and draw direction for future work in this field.

Application of Recurrent Graph Convolutional Networks to the Neural State Estimation Problem

Neural State Estimation (NSE) is a novel application of deep learning which is concerned with interpolating the state of a distribution power grid from a limited amount of sensor data and can be represented as a non-linear graph time-series nowcasting problem. Although several authors have proposed their solutions for NSE, there is neither a comparison of approaches nor an industry-standard state of the art model yet. The main purpose of this paper is to compare these solutions to a promising new approach: recurrent graph convolutional neural networks. There are theoretical reasons to assume that this class of models is suited for solving NSE. Our experiments verify that they achieve similar performance while also presenting many unique advantages compared to the previously proposed models.

Airstacknet: A Stacking Ensemble-Based Approach for Air Quality Prediction

The quality of the air we breathe during the courses of our daily lives has a significant impact on our health and well-being as individuals. Unfortunately, personal air quality measurement remains challenging. In this study, we investigate the use of first-person photos for the prediction of air quality. The main idea is to harness the power of a generalized stacking approach and the importance of haze features extracted from first-person images to create an efficient new stacking model called AirStackNet for air pollution prediction. AirStackNet consists of two layers and four regression models, where the first layer generates meta-data from Light Gradient Boosting Machine (LightGBM), Extreme Gradient Boosting Regression (XGBoost) and CatBoost Regression (CatBoost), whereas the second layer computes the final prediction from the meta-data of the first layer using Extra Tree Regression (ET). The performance of the proposed AirStackNet model is validated using public Personal Air Quality Dataset (PAQD). Our experiments are evaluated using Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Coefficient of Determination (R2), Mean Squared Error (MSE), Root Mean Squared Logarithmic Error (RMSLE), and Mean Absolute Percentage Error (MAPE). Experimental Results indicate that the proposed AirStackNet model not only can effectively improve air pollution prediction performance by overcoming the Bias-Variance tradeoff, but also outperforms baseline and state of the art models.

Student Gross Enrolment Ratio Forecasting: A Comparative Study Using Statistical Method and Machine Learning

Educational data mining has advanced substantially within the past decade. These mining strategies lay out a plan for increasing overall academic enrollment. An increase in student enrolment, in general, would enhance academic performance. Therefore, the student enrollment pattern demands great attention, as it is a vital performance indicator of academic sustainability. In this paper, student enrolment data is pre-processed to obtain the gross enrolment ratio (GER). GER analysis and forecasting were performed using the state of art models Autoregressive Integrated Moving Average (ARIMA) and Long Short-Term Memory (LSTM). The purpose of this study is to analyze and compare student GER (time series data) using ARIMA (statistical methods) and LSTM (machine learning approach), forecast GER using a better method, and propose corrective measures for increasing student enrolment. The comparison results confirmed that LSTM out-performs ARIMA by an average of 0.1322% and 5.6% in both Root Mean Square Error (RMSE) and Accuracy. The predicted GER using LSTM for the academic year 2035 is 34.23% which is far lower than 50% which is targeted by Govt. of India. An in-depth analysis of student enrolment and GER in higher education in Mizoram was done, and corrective measures were proposed for enhancing GER.

A Multi Task Allocation Based Time Optimization Framework Using Social Networks in Mobile Crowd Sensing

The quality of data and its sensing cost is the important concern for task allocation in crowd sensing. The sensing capabilities of a device to send the collection of sensor data to a cloud requires crowd sensing in order to receive reliable data. Crowd sensing is used in many areas such as traffic monitoring, smart cities, health care, transportations, environmental monitoring and many more. Most of existing works are only based on assumptions in task scheduling about the number of candidate users and are mainly performed optimization of single task allocation. If the candidate users are few, then the completion of task with in the schedule can be difficult for many sensor applications. In this work, we proposed a social network-based task allocation scheme for the optimization of multi task allocation. The main idea of the proposed work is to maximize the task completion within the allocated schedule. It is evident that the task scheduling algorithms are NP-hard and we introduced a decreasing threshold task allocation (DTT) and fast greedy selections (FGS) algorithms along with Crow COOT Foraging Optimization (CCFO) algorithm to allocate the tasks parallelly with maximum efficiency. The proposed algorithms such as C-DTT (CCFO-DTT) and C-FGS (CCFO-FGS_ are used for the efficient allocation of tasks. The combination of these algorithms can be helpful in selecting the candidate users who will perform the completion of maximum tasks. Due to the selection of proper users in each round, the time consumption of the tasks to be completed is greatly reduced. The experimental results also indicates that the proposed work performs well in the optimization of multi task allocation than the other state of the art models.

Pedestrian Simulation with Reinforcement Learning: A Curriculum-Based Approach

Pedestrian simulation is a consolidated but still lively area of research. State of the art models mostly take an agent-based perspective, in which pedestrian decisions are made according to a manually defined model. Reinforcement learning (RL), on the other hand, is used to train an agent situated in an environment how to act so as to maximize an accumulated numerical reward signal (a feedback provided by the environment to every chosen action). We explored the possibility of applying RL to pedestrian simulation. We carefully defined a reward function combining elements related to goal orientation, basic proxemics, and basic way-finding considerations. The proposed approach employs a particular training curriculum, a set of scenarios growing in difficulty supporting an incremental acquisition of general movement competences such as orientation, walking, and pedestrian interaction. The learned pedestrian behavioral model is applicable to situations not presented to the agents in the training phase, and seems therefore reasonably general. This paper describes the basic elements of the approach, the training procedure, and an experimentation within a software framework employing Unity and ML-Agents.

Threshold Based Image Enhancement Method for Low Contrast X-Ray Images Using CLAHE

While a large dataset in deep learning may seem like a positive factor, it may not always produce good results. Image quality is one of the factors that directly affects model performance, which in turn affects the quality of training. In this study, the effect of low contrast X-ray images on the detection of Covid-19 and pneumonia was investigated. Because the details are extremely important in the detection of these diseases. If the images are low contrast, it can cause some details to be missed in the detection of the disease. This problem can be solved using adaptive histogram methods such as CLAHE. The CLAHE method can apply various filters to low contrast images to bring them to the desired levels. The data set contains 8849 human lung X-ray images. The Vgg16 model was used for training. Vgg16 is a state of the art model architecture in deep learning. The image dimensions are 150x150. Classification performed before low-contrast images were filtered achieved 95.22% accuracy. After filtering based on the threshold value, accuracy increased to 97.35%. In the next stage, by searching for the best values for the parameters, accuracy was increased to 97.86%.

Hybrid twins based on optimal transport

Nowadays data is acquiring an indisputable importance in every field including engineering. In the past, experimental data was used to calibrate state-of-the art models. Once the model was optimally calibrated, numerical simulations were run. However, data can offer much more, playing a more important role than calibration or statistical analysis in the modeling/simulation process. Indeed, today data is gathered and used to train models able to replace complex engineering systems. The more and better the training data, the more accurate the model is. However, in engineering experimental data use to be the best data but also the most expensive in time and computing effort. Therefore, numerical simulations, cheaper and faster, are used instead but, even if they are closed to reality, they always present an error related to the ignorance of the engineer over the complex real system. It seems thus coherent to take advantage of each approach. This leads to the “hybrid twin” rationale. On the one hand, numerical simulations are computed as primary data source, assuming their inherent error. On the other hand, some experimental data is gathered to train a machine learning correction model which fills the prediction-measurement gap. However, learning this ignorance gap becomes difficult in some fields such as fluids dynamics, where a regression over the localized solutions can lead to non physical interpolated solutions. Therefore, the “hybrid twin” methodology proposed in this article relies on Optimal Transport theory, which provides a mathematical framework to measure distances between general objects and a completely different interpolation approach between functions.

Detailed Eye Region Capture and Animation

AbstractEven if the appearance and geometry of the human eye have been extensively studied during the last decade, the geometrical correlation between gaze direction, eyelids aperture and eyelids shape has not been empirically modeled. In this paper, we propose a data‐driven approach for capturing and modeling the subtle features of the human eye region, such as the inner eye corner and the skin bulging effect due to globe orientation. Our approach consists of an original experimental setup to capture the eye region geometry variations combined with a 3D reconstruction method. Regarding the eye region capture, we scanned 55 participants doing 36 eyes poses. To animate a participant's eye region, we register the different poses to a vertex wise correspondence before blending them in a trilinear fashion. We show that our 3D animation results are visually pleasant and realistic while bringing novel eye features compared to state of the art models.

Rapid diagnosis of Covid-19 infections by a progressively growing GAN and CNN optimisation

Background and objectiveCovid-19 infections are spreading around the globe since December 2019. Several diagnostic methods were developed based on biological investigations and the success of each method depends on the accuracy of identifying Covid infections. However, access to diagnostic tools can be limited, depending on geographic region and the diagnosis duration plays an important role in treating Covid-19. Since the virus causes pneumonia, its presence can also be detected using medical imaging by Radiologists. Hospitals with X-ray capabilities are widely distributed all over the world, so a method for diagnosing Covid-19 from chest X-rays would present itself. Studies have shown promising results in automatically detecting Covid-19 from medical images using supervised Artificial neural network (ANN) algorithms. The major drawback of supervised learning algorithms is that they require huge amounts of data to train. Also, the radiology equipment is not computationally efficient for deep neural networks. Therefore, we aim to develop a Generative Adversarial Network (GAN) based image augmentation to optimize the performance of custom, light, Convolutional networks used for the classification of Chest X-rays (CXR). MethodsA Progressively Growing Generative Adversarial Network (PGGAN) is used to generate synthetic and augmented data to supplement the dataset. We propose two novel CNN architectures to perform the Multi-class classification of Covid-19, healthy and pneumonia affected Chest X-rays. Comparisons have been drawn to the state of the art models and transfer learning methods to evaluate the superiority of the networks. All the models are trained using enhanced and augmented X-ray images and are compared based on classification metrics. ResultsThe proposed models had extremely high classification metrics with proposed Architectures having test accuracy of 98.78% and 99.2% respectively while having 40% lesser training parameters than their state of the art counterpart. ConclusionIn the present study, a method based on artificial intelligence is proposed, leading to a rapid diagnostic tool for Covid infections based on Generative Adversarial Network (GAN) and Convolutional Neural Networks (CNN). The benefit will be a high accuracy of detection with up to 99% hit rate, a rapid diagnosis, and an accessible Covid identification method by chest X-ray images.

HYBRID FIREFLY META OPTIMIZATION FOR BIO MEDICAL IMAGE PROCESSING USING DEEP LEARNING

Signal and image processing is a part of biomedical science. In that, Biomedical image processing have a great value such as recognition, segmentation and classification for disease diagnosis. In one part of biomedical science, brain tumor classification is considered with Magnetic Resonance Images (MRI) images using state of art models. Initially, the Convolutional Neural Network (CNN), Fast Convolutional Neural Network (FCNN), U-Net and M-Net model was considered for classification. Further, the Hybrid Firefly Meta Optimization (HFMO) is proposed for the better prediction purpose. The proposed work has three phases like normalization with augmentation, deep attention segmentation and classification. In the first phase, data augmentation is applied to increase the scope of the dataset. In the second phase, a deep attention technique is applied to concentrate on hotspot in the image during segmentation. Finally, a hybrid firefly optimization is applied to enhance the performance of the model in convolution neural network by backtracking the process. The measuring parameters like Dice coefficient, Jaccard index, Positive Projected Value (PPV), True Positive Rate and False Positive Rate were evaluated. The comparative analysis of various state of art models with proposed classifier were demonstrated. Thus the proposed technique produces the training accuracy as 98.62%, testing accuracy as 95.31 % and 1 % of loss.

Comparative Analysis of Background Subtraction Models Applied on a Local Dataset Using a New Approach for Ground-truth Generation

Abstract— Background subtraction is the dominant approach in the domain of moving object detection. Lots of research have been done to design or improve background subtraction models. However, there is a few well known and state of the art models which applied as a benchmark. Generally, these models are applied on different dataset benchmarks. Most of the time Choosing appropriate dataset is challenging due to the lack of datasets availability and the tedious process of creating the ground-truth frames for the sake of quantitative evaluation.
 Therefore, in this article we collected local video scenes for street and river taken by stationary camera focusing on dynamic background challenge. We presented a new technique for creating ground-truth frames using modelling, composing, tracking, and rendering each frame. Eventually we applied nine promising benchmark algorithms used in this domain on our local dataset. Results obtained by quantitative evaluations exposed the effectiveness of our new technique for generating the ground-truth scenes to be benchmarked with the original scenes using number of statistical metrics. Furthermore, results shows the outperformance of SuBSENSE model against other tested models.

SYMBOLS OF HEAVEN OBJECTS IN MASTERPIECE BY N.S. GUMILEV: MOON’s SYMBOLS

In this article there are the main symbolical meanings of the moon in N. Gumilev's poetry as parts of ‘cosmism’ model. The main semantic groups are represented, similar and opposite meanings of the moon are compared, influence of the biographic facts on interpretation of national folklore is defined, individual and author's concept of lunar "Pale maiden" is proved, the especial importance of travelling as way of world’s researching is showed. The static data defining quantity of mentions of the moon in the poetry are interpreted. The hypothesis of further observations of heavenly bodies and other space object’s interpretation in N. Gumilev's masterpiece as an aspect of individual art model is represented.

Deep Learning Analysis for Reviews in Arabic E-Commerce Sites to Detect Consumer Behavior towards Sustainability

Recently, online e-commerce has developed a major method for customers to buy various merchandise. Deep learning analysis of online customer reviews can detect consumer behavior towards sustainability. Artificial intelligence can obtain insights from product reviews to design sustainable products. A key challenge is that many sustainable products do not seem to fulfill consumers’ expectations due to the gap between consumers’ expectations and their knowledge of sustainable products. This article proposes a new deep learning model using dataset analysis and a neural computing dual attention model (DL-DA). The DL-DA model employs lexical analysis and deep learning methodology. The lexical analysis can detect lexical features in the customer reviews that emphasize sustainability. Then, the deep learning model extracts the main lexical and context features from the customer reviews. The deep learning model can predict customers’ repurchase habits concerning products that favor sustainability. This research collected data by crawling Arabic e-commerce websites for training and testing. The size of the collected dataset is about 323,150 customer reviews. The experimental results depict that the proposed model can efficiently enhance the accuracy of text lexical analysis. The proposed model achieves accuracy of 96.5% with an F1-score of 96.1%. We also compared the proposed model with state of the art models, where our model enhances both accuracy and sensitivity metrics by more than 5%.

A novel evolutionary optimization algorithm based solution approach for portfolio selection problem

<span>The portfolio selection problem is one of the most common problems which drawn the attention of experts of the field in recent decades. The mean variance portfolio optimization aims to minimize variance (risk) and maximize the expected return. In case of linear constraints, the problem can be solved by variants of Markowitz. But many constraints such as cardinality, and transaction cost, make the problem so vital that conventional techniques are not good enough in giving efficient solutions. Stochastic fractal search (SFS) is a strong population based meta-heuristic approach that has derived from evolutionary computation (EC). In this paper, a novel portfolio selection model using SFS based optimization approach has been proposed to maximize Sharpe ratio. SFS is an evolutionary approach. This algorithm models the natural growth process using fractal theory. Performance evaluation has <span>been conducted to determine the effectiveness of the model by making comparison with other state of art models such as <a name="_Hlk101168095"></a>genetic algorithm (GA) and simulated</span> annealing (SA) on same objective and environment. The real datasets of the Bombay stock exchange (BSE) Sensex of Indian stock exchange have been taken in the study. Study reveals the superior performance of the SFS than GA and SA.</span>

An approach for combining multimodal fusion and neural architecture search applied to knowledge tracing

Knowledge Tracing is the process of tracking mastery level of different skills of students for a given learning domain. It is one of the key components for building adaptive learning systems and has been investigated for decades. The empirical success of deep neural networks in the past few years has encouraged researchers in the learning science community to take similar approaches. However, most existing deep learning based knowledge tracing models have the following limitations: (1) Only use the correct/incorrect response, ignoring useful information from other modalities. (2) For works do consider multimodality use simple concatenation, which might not be the best way for modality fusion. (3) Design their network architectures manually via trial and error. To solve these problems, we propose Multimodal Fusion and Neural Architecture Search (MFNAS) in this paper. The commonly used neural architecture search technique could be considered as a special case of our proposed approach when there is only one modality involved. We further propose to use a new metric called time-weighted Area Under the Curve (weighted AUC) to measure how a sequence model performs with time. Our proposed approach MFNAS allows more efficient design of Knowledge Tracing models. Besides, evaluated on two public real datasets, the discovered model is able to achieve around 12% improvement in coefficient of determination, 2% improvement in AUC and weighted AUC compared with state of the art models. What’s more, unlike most existing works, we conduct McNemar’s test on the model predictions and the results are statistically significant.

A Deep Learning based Air Quality Prediction Technique Using Influencing Pollutants of Neighboring Locations in Smart City

The level of air pollution in smart cities plays a critical role in the community’s health and quality of life. Thus, air pollution forecasting would be beneficial and would guide citizens in avoiding exposure to dangerous emissions. The air health of a place can be diagnosed by close observation of the AQI (Air Quality Index) of that place. Moreover, the AQI of a place may have some influence on the pollutant concentration of the neighboring places. To address this issue, this work introduces a hybrid deep learning framework that is able to predict the values of a corresponding metric: AQI of smart cities. As a part of this work, two algorithms are proposed. The first one replaces the missing values in the dataset and the second one formulates the influence of the nearby places’ pollutant concentrations on the air quality of a particular place. A deep learning-based forecasting model is also proposed by combining 1D-CNN and Bi-GRU. To test the applicability of the framework, a large-scale experiment is carried out with the real-world dataset collected from New South Wales, Australia. Experimental results validate that the proposed framework provides a stable forecasting result, it confirms that the AQI of a place gets affected by the pollutant concentration of the nearby places and the comparison of forecasting result with the existing state of the art models shows that the proposed model outperforms the other models.