Repository of FERIT Osijek
Not a member yet
4053 research outputs found
Sort by
Development of a mobile application for the assessment and monitoring of stroke risk using machine learning classification and generative AI
No full textU ovom diplomskom radu razvijena je aplikacija za procjenu i praćenje rizika klasifikacijskim
postupcima strojnog učenja i generativnom umjetnom inteligencijom. Mobilna aplikacija
ostvarena je pomoću integrirane razvojne okoline Android Studio zajedno s jezikom Kotlin. Za
izradu korisničkog sučelja korišten je okvir Jetpack Compose. Za izradu korisničkog računa
korišten je Firebase Auth, a za pohranu osobnih i zdravstvenih podataka Firebase Realtime
Database. Za klasifikaciju su korišteni postupci šume odluke, pojačanog stabla odluke i
neuronske mreže te je na temelju rezultata vrednovanja za implementaciju odabran algoritam
pojačanog stabla odluke. Procjena rizika odvija se putem modela izrađenog unutar Microsoft
Azure Machine Learning Studija, a generiranje preporuka slanjem upita OpenAi modelu.
Pohranjeni korisnički osobni podaci zajedno s unesenim zdravstvenim podacima koriste se pri
procjeni rizika od moždanog udara i generiranju preporuka od strane OpenAi-ja. Ispitivanje
mobilne aplikacije provedeno je kroz razne korisničke slučajeve prilikom čega je testirana i
utvrđena njena točnost korištenjem različitih korisničkih podataka.In this thesis, an application for risk assessment and monitoring using classification machine
learning techniques and generative artificial intelligence has been developed. The mobile
application was created using the Android Studio integrated development environment along
with the Kotlin programming language. The user interface was designed using the Jetpack
Compose framework. For user account management, Firebase Auth was used, while personal
and health data are stored in Firebase Realtime Database. Classification was performed using
decision tree, boosted decision tree, and neural network techniques, with the boosted decision
tree algorithm being selected based on the results. Risk assessment is conducted through a
model developed within Microsoft Azure's Machine Learning Studio, and recommendations
are generated by querying the OpenAI model. Stored user personal data, along with entered
health data, are used in the risk assessment for stroke and in generating recommendations by
OpenAI. The mobile application was tested through various user cases, evaluating its accuracy
using different user data
Optimization of segmentation of road scenes in real time
No full textCilj ovog diplomskog rada bio je implementirati U-Net, SegNet i ENet arhitekture te
njihov trening uz pomoć PyTorch biblioteke, a dobivene rezultate koristiti za usporedbu navedenih
arhitektura po pitanju točnosti, brzini izvođenja, učinkovitosti u stvarnom vremenu i kompleksnosti
implementacije. Trening je proveden na Cityscapes skupu podataka, na kojem je provedena pred-obrada
potrebna za korištenje pri treningu. Prije treninga, arhitekture su uspoređene po veličini i broju
parametara; tijekom treninga, prate se srednja vrijednost unije i presjeka te točnost piksela; a nakon
treninga uspoređuju se po trajanju i brzini treniranja, broju epoha te se provodi testiranje na testnom
dijelu skupa podataka. Pri testiranju se mjere vrijeme zaključivanja i mogućnost izvođenja u stvarnom
vremenu, a točnost se mjeri kvalitativno, pošto ne postoji izvor istine za segmentacije testnog skupa.
Navedenim usporedbama pružen je dubok uvid u primjenjivost navedenih arhitektura u kontekstu
automobilske percepcije te je svakoj arhitekturi pronađena prikladna primjena na temelju njezinih
rezultata i performansi – ENet je izabrana za zadatke u kojima su ključne brzina i učinkovitost resursa, a
U-Net i SegNet se, zbog svojih sličnih rezultata, predlažu za istu primjenu, a to su situacije u kojima je
točnost bitnija od iskorištenosti resursa.The aim of this thesis was to implement U-Net, SegNet and ENet architectures along with their
training with the help of the PyTorch library, and using the obtained results to compare the mentioned
architectures in terms of accuracy, execution speed, real-time efficiency and implementation
complexity. The training was carried out on the Cityscapes dataset, which was pre-processed for use in
training. Before the training, the architectures were compared according to their size and number of
parameters; during the training the mean Intersection over Union and the pixel accuracy were
monitored; after the training, they were compared according to the duration and speed of the training,
the number of epochs, and testing was carried out on the test part of the data set. During testing,
inference time and real-time-execution capabilities were measured quantitatively, and accuracy was
measured qualitatively, since there is no ground truth for the test set segmentations. The
aforementioned comparisons provided deep insight into the applicability of the architectures in the
context of automotive perception, and a suitable task was found for each architecture based on its
results and performance - ENet was chosen for tasks where speed and resource efficiency are key, while
U-Net and SegNet are suggested for the same tasks due to their similar results – tasks in which the
accuracy is more important than resource utilization
Intelligent management of traffic intersections based on computer vision
No full textGlavni cilj ovog diplomskog rada bio je razviti sustav za inteligentno upravljanje
semaforima koji smanjuje vrijeme čekanja vozila i povećava protočnost prometa na raskrižjima.
Tradicionalni sustavi upravljanja semaforima često stvaraju nepotrebne zastoje u prometu. Predloženo
je rješenje koje na temelju računalnog vida prilagodljivo upravlja semaforima. U radu je analizirano
simulacijsko okruženje CARLA simulatora te su pojašnjeni osnovni principi stvaranja prometa i RGB
kamera s kojima se snimala svaka traka prometnog raskrižja. U radu su detaljno prikazani razvojni koraci
implementacije, uključujući postavljanje prometnih scenarija unutar simulatora, odabir prikladnog
detektora objekata za detekciju vozila te integraciju detektora u simulacijsko okruženje. Podaci
prikupljeni putem računalnog vida se obrađuju kako bi se prilagodila trajanja semaforskih ciklusa, čime
se optimizira protok prometa. Sustav za inteligentno upravljanje semaforima na prometnom raskrižju
prilagođava semaforske cikluse u stvarnom vremenu na temelju trenutnog broja vozila u svakoj traci.
Evaluacija sustava obavljena je stvaranjem različitih prometnih scenarija na temelju kojih se usporedio
rad sustava za inteligentno upravljanje semaforima s tradicionalnim načinom upravljanja semaforima.
Rezultati evaluacije sustava pokazali su značajno smanjenje prosječnog vremena čekanja po vozilu te je
povećana ukupna protočnost prometa na prometnom raskrižju.The main goal of this thesis was to develop a system for intelligent traffic light management
that reduces vehicle waiting time and increases traffic flow at intersections. Traditional traffic light
management systems often cause unnecessary traffic jams. The proposed solution uses computer vision
to adaptively control the traffic lights. The thesis analyzes the simulation environment of the CARLA
simulator and explains the basic principles of traffic generation and the RGB cameras used to record
each lane of the traffic intersection. The paper provides a detailed description of the development steps,
including setting up traffic scenarios within the simulator, selecting an appropriate object detector for
vehicle detection, and integrating the detector into the simulation environment. Data collected through
computer vision is processed to adjust traffic light cycle durations, optimizing traffic flow. The intelligent
traffic light management algorithm adjusts the traffic light cycles in real-time based on the current
number of vehicles in each lane. The system evaluation was performed by creating different traffic
scenarios to compare the performance of the intelligent traffic light management algorithm with the
traditional traffic light management method. The evaluation results showed a significant reduction in
the average vehicle waiting time and an overall increase in traffic flow at the intersection
Use of blockchain technology in measurement systems
No full textAnaliza primjene blockchain tehnologije u mjernim sustavima, za distribuciju i naplatu energije.Analysis of usage of blockchain technology in measurement systems for distribution and billing
of energy
Application of LLMs in understanding and generating dietary advice
No full textOvaj diplomski rad bavi se primjenom velikih jezičnih modela (LLM) za razumijevanje i
generiranje savjeta o prehrani. Glavni problem obuhvaća prepoznavanje sastojaka i nutritivnih
vrijednosti s prehrambenih proizvoda te integraciju tih podataka u sustav za pružanje korisnih
preporuka. Problem je riješen razvojem mobilne aplikacije pomoću React Native i Python Flask
tehnologija. Implementacijom docTR modela za optičko prepoznavanje znakova omogućeno je čitanje
nutritivnih vrijednosti i sastojaka sa slike proizvoda. Sustav na temelju prepoznatih sastojaka i nutritivnih
vrijednosti te korisničkih podataka iz Firestore baze podataka generira savjet koristeći OpenAI-jev GPT-
3.5-turbo model. Ovaj rada svojim rezultatima potvrdio je mogućnost primjene LLM-a za pružanje
personaliziranih prehrambenih preporuka otvarajući nove mogućnosti u digitalizaciji prehrane.This thesis deals with the application of large language models (LLM) in understanding and
generating dietary advice. The main problem involves identifying ingredients and nutritional values from
food products and integrating this data into a system for providing useful recommendations. The
problem was solved by developing a mobile application using React Native and Python Flask
technologies. By implementing the docTR model for optical character recognition, it is possible to read
the nutritional values and ingredients from the product image. Based on recognized ingredients and
nutritional values and user data from the Firestore database, the system generates advice using
OpenAI's GPT-3.5-turbo model. With its result, this thesis confirmed the possibility of applying LLM to
provide personalized nutritional recommendations, opening up new possibilities in the digitization of
nutrition
Web application for boat rental
No full text"Ferit Boat Charter" je korisnički pristupačna web aplikacija dizajnirana za najam brodova, s ciljem pojednostavljenja procesa pregledavanja, iznajmljivanja i objavljivanja brodova za najam. Dostupnost određenih značajki ovisi o ulozi korisnika unutar aplikacije. Ima četiri vrste korisnika: posjetitelj (neprijavljeni korisnik), administrator, iznajmljivač i unajmitelj (vlasnik broda koji želi postaviti svoj brod na raspolaganje za najam). Sa svojim jednostavnim dizajnom, aplikacija osigurava lagan pristup svim svojim funkcionalnostima.
Kako bi postigla svoje ciljeve, "Ferit Boat Charter" koristi modernu tehnologiju i trendove tržišta. Frontend aplikacije je razvijen koristeći HTML, CSS i JavaScript React biblioteku, dok backend koristi Javu sa Spring Boot okvirom. Prikupljanje i pohranjivanje podataka upravlja se putem MySQL baze podataka.
Korištenjem suvremene tehnologije i korisnički prijateljskog sučelja, "Ferit Boat Charter" pojednostavljuje proces najma brodova, pružajući zadovoljstvo kako iznajmljivačima, tako i vlasnicima brodova."Ferit Boat Charter" is a user-friendly web application designed for boat rentals, aiming to simplify the process of browsing, renting, and listing boats for rent. The accessibility of specific features depends on the user's role within the application, which includes four types: visitor (unregistered user), admin, renter, and boat owner listing their boat for rent. With its straightforward design, the application ensures easy access to all its functionalities.
To achieve its goals, "Ferit Boat Charter" leverages modern technology and market trends. The frontend of the application is developed using HTML, CSS, and the JavaScript React library, while the backend utilizes Java with the Spring Boot framework. Data retrieval and storage are managed through a MySQL database.
By employing current technology and a user-friendly interface, "Ferit Boat Charter" streamlines the boat rental process, catering to the needs of both renters and boat owners
Angular web application for physical therapy and rehabilitation
No full textCilj diplomskog rada je razvoj aplikacije za fizikalnu rehabilitaciju, koristeći .NET i Angular
tehnologije. Tema aplikacije je olakšati proces fizikalne rehabilitacije, osiguravajući time
fizijatarima i klijenatima platformu za korištenje. Fizijatri imaju mogućnost kreiranja i sastavljanja
rehabilitacijskog programa te mogućnost prilaganja informacija i uputa, te ga objaviti javno
korisnicima na raspolaganje. Korisnici imaju uvid u sve rehabilitacijske programe objavljeno
javno od strane fizijatara, te mogućnosti pretplate na program, čime omogućuju praćenja napretka
programa. Web aplikacije je razvijena u tehnologijama Angular i .NET-u.The goal of this thesis is to develop a physical rehabilitation web aplication using .NET and
Angular technologies. The application's purpose is to facilitate the physical rehabilitation process
by providing a platform for both physiotherapists and clients. Physioterapists have the ability to
create and deseign rehabilitation programs, attach information and instructions, and publish them
publicly for users. Users can view all publicly available rehabilitation programs created by
physiotherapists, subscribe to programs, and track their progress. The web application is developed
using Angular and .NET technologies
Web application for digital image processing
No full textCilj ove web aplikacije je omogućiti korisnicima jednostavnu obradu digitalnih slika
putem intuitivnog korisničkog sučelja. Aplikacija omogućuje korisnicima da prenesu sliku u jednom od
podržanih formata te odaberu različite opcije obrade putem padajućeg izbornika. Nakon što korisnik
potvrdi odabrane parametre, aplikacija šalje podatke na poslužitelj gdje se izvršava obrada slike koristeći
specijalizirane algoritme. Rezultirajuća slika se prikazuje korisniku te je dostupna za preuzimanje ili
daljnju obradu. Aplikacija je dizajnirana s naglaskom na jednostavnost i preglednost, omogućujući brz i
učinkovit rad. Implementacija same biblioteke koja sadrži algoritme za obardu slike provedena je u
programskom jeziku C koji pruža visoku učinkovitost izvedbe, a sama web aplikacija realizirana je u
JavaScript ekosustavu koji omogućava web komunikaciju te nudi veliki raspon biblioteka što pospješuje
jednostavnost implementacije.The aim of this web application is to provide users with simple digital image processing
through an intuitive user interface. The application allows users to upload an image in one of the
supported formats and select various processing options via a dropdown menu. Once the user confirms
the selected parameters, the application sends the data to the server where image processing takes
place using specialized algorithms. The resulting image is displayed to the user and is available for
download or further processing. The application is designed with emphasis on simplicity and clarity,
enabling fast and efficient execution. The implementation of the image processing algorithms is carried
out in the C programming language, chosen for its high performance capabilities, while the web
application itself is developed within the JavaScript ecosystem, facilitating web communication and
offering a wide range of libraries to enhance ease of implementation
Co-simulation model for determining the optimal settings of active power filters in a nonlinear low-voltage network
No full textU doktorskom rada razvijen je i prikazan kosimulacijski model za određivanje paramatara i optimalnih postavki paralelnog aktivnog filtera snage u nesimatričnoj niskonaponskoj mreži. Razvijeni model mreže sa svim kreiranim modelima trošila i modelima izvora u optimizacijskom procesu izvodi se 24 puta za 24 vremenska intervala. Izvršavanje optimizacijskog procesa izvršeno je pomoću algoritama za određivanje optimalnih postavki aktivnih filtera snage u niskonaponskoj mreži koji prati navedenu satnu promjenu opterećenja u jednom danu u svakom čvoru ispitne mreže. Algoritam kao optimizacijsku metodu upotrebljava metodu optimizacije roja čestica te s kreiranom varijabilnosti prolazi kroz različita stanja niskonaponske mreže. Tijekom svih vremenskih intervala algoritam s metodom optimizacije roja čestica određuje lokaciju te parametre i optimalne postavke paralelnih aktivnih filtera u niskonaponskoj mreži.
Određivanje i vrednovanjem razvijenog sustava promatra se u karakterističnim trenutcima kada nastupaju određena stanja ispitne mreže. Karakteristični promatrani trenutci definirani su iz dobivenih rezultata vrednovanja optimizacijske metode. Kreirana promjena opterećenja u mreži utjecalo je na odluku odabira lokacije te dovela do određivanja obuhvata djelovanja APF-a. Definiranje područja djelovanja kao i kriteriji za adekvatnost primjene aktivnog filtera naglašavaju važnosti integracije APF-ova u niskonaponsku distribucijsku mrežu.In the doctoral thesis, a co-simulation model was developed and presented for determining the parameters and optimal settings of shunt active power filters in a nonlinear low-voltage network. The developed network model with all created models of consumers and models of sources in the optimization process is performed 24 times, for 24 time intervals. The optimization process was carried out using algorithms for determining the optimal settings of active power filters in the low-voltage network, which monitors the specified hourly load change in one day in each node of the test network. The algorithm uses the particle swarm optimization method as an optimization method, and with the created variability it goes through different states of the low-voltage network. During all time intervals, the algorithm with the particle swarm optimization method determines the location, parameters and optimal settings of shunt active filters in the low-voltage network.
Determination and evaluation of the developed system is observed at characteristic moments when certain conditions of the test network occur. The characteristic observed moments are defined from the obtained evaluation results of the optimization method. The created loads change in the network influenced the decision to select the location and led to the determination of the scope of APF activities. Defining the area of operation as well as the criteria for the adequacy of the application of the active filter emphasize the importance of integrating APFs into the low-voltage distribution network
Integration of photovoltaic power plants into low voltage distribution grid
No full textU ovome radu teorijski je opisan zatim simuliran utjecaj integracije fotonaponske elektrane na
različite strukture niskonaponske mreže. Glavni fokus je bio prikazati što se događa s naponom na
sabirnicama, gubitcima i tokovima snaga u različitim scenarijima. Prvo su odrađene simulacije u
programu EasyPower, a potom se analiziraju i uspoređuju dobiveni rezultati. Na kraju je izveden
zaključak o utjecaju integracije fotonapona na obje strukture mreže i njihova pogonska
ograničenja.In this paper, the impact of integrating a photovoltaic power plant on various low-voltage network
structures is theoretically described and then simulated. The main focus was to demonstrate what
happens to the voltage at busbars, losses, and power flows in different scenarios. Simulations were
first conducted using the EasyPower software, followed by the analysis and comparison of the
obtained results. Finally, a conclusion was drawn about the impact of photovoltaic integration on
both network structures and their operational limitations