SN Computer Science

Digital elevation model (DEM) is a critical data source for variety of applications such as road extraction, hydrological modeling, flood mapping, and many geospatial studies. The usage of high-resolution DEMs as inputs in many application areas improves the overall reliability and accuracy of the raw dataset. The goal of this study is to develop a machine learning model that increases the spatial resolution of DEM without additional information. In this paper, a GAN based model (D-SRGAN), inspired by single image super-resolution methods, is developed and evaluated to increase the resolution of DEMs. The experiment results show that D-SRGAN produces promising results while constructing 3 feet high-resolution DEMs from 50 feet low-resolution DEMs. It outperforms common statistical interpolation methods and neural network algorithms.This study shows that it is possible to use the power of artificial neural networks to increase the resolution of the DEMs. The study also demonstrates that approaches from single image super-resolution can be applied for DEM super-resolution.

Covering face with accessories such as mask, scarf and sunglass is a common criminal activity in automated teller machine (ATM) robbery. Therefore, detection of covered face using ATM surveillance camera can be an effective solution to reduce robbery or crime. This paper presents a novel method to detect covered face from ATM surveillance camera images. Specifically, three facial features, i.e., skin color, elliptical face shape and facial width-to-height ratio (fWHR), incorporated with geometrical property of ellipse have been employed to estimate the covered region. In addition, three parameters, i.e., facial area, fWHR and covered area percentage, have been utilized for reliable classification. Experiment results demonstrate that the method can detect full covered, uncovered and partially covered faces at a correct detection rate of 98.3%, 93.3% and 97.78%, respectively. The overall correct detection rate is 96.48%, which is found to be better than previous studies. Also, the proposed method can handle faces covered with few new face hiding objects such as hijab, niqab and robber’s ski mask. Furthermore, processing time of the proposed algorithm is significantly improved while it is compared to the existing methods. The detection time varies between 31 and 67 ms which is equivalent to 15–32 frames per second.

Aspect level sentiment analysis has received much attention by researchers over the last few years. It aims first to determine the aspects in a given text (e.g., a comment, a sentence, a review, etc.) and second to perform the sentiment analysis (i.e., determine the polarity, such as positive, negative, or neutral) of the corresponding text with respect to each aspect. In this paper, we propose an original method of sentiment analysis for Tunisian social media. Our method is mainly based on domain ontologies for aspect extraction and deep learning models for aspect sentiment classification. Evaluation results are very encouraging, since we outperformed the baseline method with an enhancement of 17% for the task of aspect level sentiment classification.

Collecting large amounts of data is beneficial in machine learning to generate models that are less biased. There are many cases in which pieces of similar data are distributed among organizations, and it is difficult to integrate these data owing to issues involving privacy and cost. Integrating these distributed data without delivering the original data leads to the concept of data collaboration, which combines data held by different organizations in a secure manner. We propose a method in which a distance matrix of the original data obtained using common data among organizations is shared to learn neighbor information of the original data. Specifically, the proposed method robustly integrates distributed data, which is of as good quality as connected raw data, in cases where the amount of data in each organization is small and the data bias is large. In addition, the proposed method is applicable to data contaminated by noise. To demonstrate the effectiveness of the proposed method, we performed a classification task on open biological data divided into several pieces and found that the classification results for divided data were as precise as when all data were available. Finally, we show that the robustness of the method against noise improves the anonymity of the original data as a by-product.

This article aims to propose a model on the soil degradation risk along the Cameroonian shores of Lake Chad based on the statistical analysis of spectral indexes of Sentinel 2A satellite images. A total of four vegetation indexes such as the Greenness Index and Disease water stress index and nine soil indexes such as moisture, brightness, or organic matter content are computed and combined to characterize vegetation cover and bare soil state, respectively. All these indexes are aggregated to produce one image (independent variable) and then regressed by individual indexes (dependent variable) to retrieve correlation and determination coefficients. Principal Component Analysis and factorial analysis are applied to all spectral indexes to summarize information, obtain factorial coordinates, and detect positive/negative correlation. The first factor contains soil information, whereas the second factor focuses on vegetation information. The final equation of the model is obtained by weighting each index with both its coefficient of determination and factorial coordinates. This result generated figures’ cartography of five classes of soils potentially exposed to the risk of soil degradation. Five levels of exposition risk are obtained from the "Lower" level to the "Higher": the "Lower" and "Moderate to low" levels occupy, respectively, 25,214.35 hectares and 130,717.19 hectares; the "Moderate" level spreads ​​137,404.34 hectares; the "High to moderate" and "Higher" levels correspond, respectively, to 152,371.91 hectares and 29,175.73 hectares.

Electric arc furnace (EAF) serves as a major contributor for the global industrialization due to its wide versed application for manufacturing of high-grade steel. The dynamic operation of EAF is considered to be highly nonlinear and chaotic. Hence, to examine their operations and effects on the electrical network, it is necessary to create an accurate model of an EAF, several strategies including mathematical techniques and data-driven models have already been utilized to simulate the V–I behavior of electric arc furnaces. The paper focuses on examining the data-driven modelling techniques, especially least-square support vector machines (LS-SVM) for estimation of Electric Arc Furnace Parameters. The outcomes demonstrate that the suggested approach using a radial base function kernel offers a model to forecast both the arc current and arc voltage of EAFs.

The insurance sector is ingoing a new era of innovation because of revolutionary technologies like Blockchain, Internet of Things, Chatbots, Telematics, and Artificial Intelligence. Newness in insurance has a significant impact on both the nation’s economy and insurance performance. This research explores the effects of Insur-Tech on the efficiency of the insurance industry and provides multiple perspectives on the phenomenon. By examining the advantages that arise for the insurer and the consumer, this paper seeks to demonstrate the basics and significance of modern technologies in the insurance industry. Hypothesis is taken into account to explore the functioning of certain factors which play vital role in the insurance performance. Data have been collected from the secondary sources. IBM SPSS Statistics has been applied to analyze data from a plethora of dimensions. Regression analysis is applied to show the interplay between dependent and independent variables. Bottom line, this concept is proved worthy from the aspects of insurer or insurance firms. The findings of the study observed that technologies help to attract new clients and retention of old clients which help to increase the premium volume day by day. Moreover, insurance technologies help to distribution more data and to share more information regarding the present and potential clients and mitigate risk of insurance business.