SPECIES IDENTIFICATION FOR AQUATIC BIOMONITORING USING DEEP RESIDUAL CNN AND TRANSFER LEARNING

Aquatic insects and other benthic macroinvertebrates are mostly used as bioindicators of the ecological status of freshwaters. However, an expensive and time-consuming process of species identification represents one of the key obstacles for reliable biomonitoring of aquatic ecosystems. In this paper, we proposed a deep learning (DL) based method for species identification that we evaluated on several available public datasets (FIN-Benthic, STONEFLY9, and EPT29) along with our Chironomidae dataset (CHIRO10). The proposed method relies on three DL techniques used to improve robustness when training is done on a relatively small dataset: transfer learning, data augmentation, and feature dropout. We applied transfer learning by employing ResNet-50 deep convolutional neural network (CNN) pretrained on ImageNet 2012 dataset. The results show significant improvement compared to original contributions and confirms that there is a considerable gain when there are multiple images per specimen.

APPLICATION OF WEB BASED TECHNOLOGIES FOR IMPLEMENTATION OF AUTOMATED SMART CITY SERVICES

The paper considers, proposes and describes possibilities and methods to solve problems in providing services in smart cities, where citizens have to appear in person in the city or municipality premises or at the teller/counter of the institution. In this way, by using information and telecommunication technologies, Web based solutions and Internet, citizens obtain services online, from their homes or working places, using all types of their PC equipment or smart mobile telephone, and do not waste their time in the city or municipality premises. Their contacts are also reduced, which is very important in the context of actual Corona virus pandemic. The services are provided and charged automatically and online. No cash is used, which is also a potential carrier of the Corona virus. The proposed method and proposed solution are based on application of the specially developed algorithm for service automation, developed and implemented adequate software application and designed hardware solution that fully supports the software solution and the process of service delivery automation. The proposed system decreases costs and increases availability, quality and speed of services realization in smart cities and municipalities. Also, the proposed solution uses reliable methods for identification and authentication of a person using a service. For identification are used pictures, taken by a Web camera or a smart mobile telephone, of an identity document and of the face of the user and appropriate software for text and face recognition.

INFLUENCE OF SKIN EFFECT ON THE CALCULATION OF BUSBAR SYSTEMS HAVING RECTANGULAR CROSS SECTION

One method for the calculation of current density distribution in a finite number of long parallel conductors, having rectangular cross section, is proposed in this paper. Numerical results aim to highlight the importance of the skin effect, which can be combined with the proximity effect. The method of superposition of these two effects was applied to the calculation of the electromagnetic field in electric power busbars systems. It has been shown that the skin effect has a much greater impact, especially when the conductors are thin and strong electric currents flow through them, so special attention is paid to its calculation. For numerical solution the integral equations are used. The function of current density is approximated by the finite functional series. This way leads to a very accurate solution with only two terms. Differential evolution method is applied for minimization of error function. To demonstrate the application of the proposed approach, numerical values for busbars are presented and compared with values obtained by using the finite elements method.

GENERATING KNOWLEDGE STRUCTURES FROM OPEN DATASETS' TAGS - AN APPROACH BASED ON FORMAL CONCEPT ANALYSIS

Under influence of data transparency initiatives, a variety of institutions have published a significant number of datasets. In most cases, data publishers take advantage of open data portals (ODPs) for making their datasets publicly available. To improve the datasets' discoverability, open data portals (ODPs) group open datasets into categories using various criteria like publishers, institutions, formats, and descriptions. For these purposes, portals take advantage of metadata accompanying datasets. However, a part of metadata may be missing, or may be incomplete or redundant. Each of these situations makes it difficult for users to find appropriate datasets and obtain the desired information. As the number of available datasets grows, this problem becomes easy to notice. This paper is focused on the first step towards decreasing this problem by implementing knowledge structures to be used in situations where a part of datasets' metadata is missing. In particular, we focus on developing knowledge structures capable of suggesting the best match for the category where an uncategorized dataset should belong to. Our approach relies on dataset descriptions provided by users within dataset tags. We take advantage of a formal concept analysis to reveal the shared conceptualization originating from the tags' usage by developing a concept lattice per each category of open datasets. Since tags represent free text metadata entered by users, in this paper we will present a method of optimizing their usage through means of semantic similarity measures based on natural language processing mechanisms. Finally, we will demonstrate the advantage of our proposal by comparing concept lattices generated using formal the concept analysis before and after the optimization process. The main experimental research results will show that our approach is capable of reducing the number of nodes within a lattice more than 40%.

ARTIFICIAL NEURAL NETWORKS APPLICATION TO PREDICTION OF ELECTRICITY CONSUMPTION

In this paper, we will present an artificial neural network (ANN) model trained to forecast hourly electricity consumption of energy in industry for a day-ahead. We will start with a brief analysis of the global electricity market with a special reference to the Serbian market. Next, the daily electricity consumption amounts between August 1st and December 19th 2019 will be analyzed using statistical tools. According to the obtained results, we will give predictions of our models, based on different number of previous days.

IMPROVING INTERNET OF THINGS PARKING SYSTEMS

Edge data processing represents the new evolution of the Internet and Cloud computing. Its application to the Internet of Things (IoT) is a step towards faster processing of information from sensors for better performance. In automated systems, we have a large number of sensors, whose information needs to be processed in the shortest possible time and acted upon. The paper describes the possibility of applying Artificial Intelligence on Edge devices using the example of finding a parking space for a vehicle, and directing it based on the segment the vehicle belongs to. Algorithm of Machine Learning is used for vehicle classification, which is based on vehicle dimensions.

DETERMINING THE IMPACTS OF FREIGHT TRANSPORT MODE COMBINATIONS ON AIR POLLUTION USING ARTIFICIAL NEURAL NETWORKS

Transport is one of the largest emitters of harmful substances that affect air quality. Each combination of freight transport modes has a different volume and at the same time has a differentiated negative impact on air quality. That is why the European Union has been making special efforts for many years to create and implement strategies aimed at improving air quality. The main goal of this paper is to present a methodology that enables quantification and analysis of the impact of each freight transport mode combination on air quality using feed-forward neural networks. The developed model uses the parameters of the EU member states in the period from 2000 to 2014. In addition to the scientific and practical contribution, the development of the model provides a good basis for the universal platform formation in order to create and develop strategies, i.e. measures to improve air quality on a global level.

PSEUDO-MULTIVARIATE LSTM NEURAL NETWORK APPROACH FOR PURCHASE DAY PREDICTION IN B2B

This research focuses on trying to predict the moment of the next purchase for a customer in vendor-customer B2B scenario using an LSTM neural network and comparing prediction results from different input features. In a previous research we performed predictions for a specific customer product pair and used previous purchases for that pair as input data, but the number of such previous purchases was often very limited which resulted in low accuracy of predictions. By aggregating purchase data for all products a customer purchased, we were able to get more precise predictions of the next purchase. Additionally, expanding our input feature set yielded even better results. We performed an evaluation of LSTM networks trained with the most successful combination of input features for a six month period. Each of the networks was trained with purchase data up to the starting point of the selected period and the predictions were performed, after which additional input for the following seven days was added to the network. This process was then repeated for the entire six month period and a slight downward trend can be noticed for error metrics, leading to the conclusion that the network would perform even better over time with the addition of future purchases.

THE APPROACH TO MEASUREMENT OF REQUIREMENT QUALITY BY APPLICATION OF GENERALIZED PRIORITIZED FUZZY CONSTRAINT SATISFACTION PROBLEM

The requirement quality affects product development at all lifecycle stages, as well as the end product. Poorly defined requirements bring to extended deadlines, increased financial costs, even to project disruption. Current researches related to the good quality of requirements include characteristics of good requirements and the development of new elicitation techniques. Requirement quality evaluation should be tailored both to the professionals and users who defined requirements according to their needs. Therefore, the model is designed for requirement quality measurement based on the characteristics of good requirements by application of the Generalized Prioritized Fuzzy Constraint Satisfaction Problem. The model enables the participation of selected characteristics of good requirements in quality evaluation, according to priorities. The evaluator obtains information if the requirement satisfies the given quality satisfaction threshold based on the degree of fulfillment of selected characteristics of a good requirement. The model is applied to all types of requirements, as well as to the evaluation of requirements at all software development lifecycle stages.

STABILITY, FINITE-TIME STABILITY AND PASSIVITY CRITERIA FOR DISCRETE-TIME DELAYED NEURAL NETWORKS

In this paper, we present the problem of stability, finite-time stability and passivity for discrete-time neural networks (DNNs) with variable delays. For the purposes of stability analysis, an augmented Lyapunov-Krasovskii functional (LKF) with single and double summation terms and several augmented vectors is proposed by decomposing the time-delay interval into two non-equidistant subintervals. Then, by using the Wirtinger-based inequality, reciprocally and extended reciprocally convex combination lemmas, tight estimations for sum terms in the forward difference of LKF are given. In order to relax the existing results, several zero equalities are introduced and stability criteria are proposed in terms of linear matrix inequalities (LMIs). The main objective for the finite-time stability and passivity analysis is how to effectively evaluate the finite-time passivity conditions for DNNs. To achieve this, some weighted summation inequalities are proposed for application to a finite-sum term appearing in the forward difference of LKF, which helps to ensure that the considered delayed DNN is passive. The derived passivity criteria are presented in terms of linear matrix inequalities. Some numerical examples are presented to illustrate the proposed methodology.