prikaz prve stranice dokumenta Prepoznavanje zvijezda postupcima računalnog vida
Access restricted to authenticated users
master's thesis
Prepoznavanje zvijezda postupcima računalnog vida
2022. urn:nbn:hr:179:279808
Maslov, Marin
University of Split
Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture
Department of Electronics and Computing

Cite this document

Maslov, M. (2022). Prepoznavanje zvijezda postupcima računalnog vida (Master's thesis). Split: University of Split, Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture. Retrieved from https://urn.nsk.hr/urn:nbn:hr:179:279808

Maslov, Marin. "Prepoznavanje zvijezda postupcima računalnog vida." Master's thesis, University of Split, Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture, 2022. https://urn.nsk.hr/urn:nbn:hr:179:279808

Maslov, Marin. "Prepoznavanje zvijezda postupcima računalnog vida." Master's thesis, University of Split, Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture, 2022. https://urn.nsk.hr/urn:nbn:hr:179:279808

Maslov, M. (2022). 'Prepoznavanje zvijezda postupcima računalnog vida', Master's thesis, University of Split, Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture, accessed 08 May 2024, https://urn.nsk.hr/urn:nbn:hr:179:279808

Maslov M. Prepoznavanje zvijezda postupcima računalnog vida [Master's thesis]. Split: University of Split, Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture; 2022 [cited 2024 May 08] Available at: https://urn.nsk.hr/urn:nbn:hr:179:279808

M. Maslov, "Prepoznavanje zvijezda postupcima računalnog vida", Master's thesis, University of Split, Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture, Split, 2022. Available at: https://urn.nsk.hr/urn:nbn:hr:179:279808

Please login to the repository to save this object to your list.
Metadata
TitlePrepoznavanje zvijezda postupcima računalnog vida
Title (english)Star recognition using computer vision procedures
AuthorMarin Maslov
MentorDarko Stipaničev (mentor)
GranterUniversity of Split
Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture
(Department of Electronics and Computing)
Split
Defense date and country2022-07-07, Croatia
Scientific / art field,
discipline and subdiscipline		TECHNICAL SCIENCES
Computing
Abstract
Prije rješavanja problema ovog rada bilo je potrebno pripremiti ulazne podatke, tj. zvijezde koje su točkice pretvoriti u nešto prepoznatljivije. Klasifikatori objekata uglavnom imaju problema s prepoznavanjem točkastih uzoraka (pogotovo prorijeđenih kao što su zviježđa). Stoga su se na zvijezde postavljali koncentrične kružnice. U nastavku rada se tražio klasifikator, te sam koncept strojnog učenja koji bi najbolje riješio ovaj problem. Nakon treniranja HAAR kaskade (cjelinsko učenje), te MobileNet modela (prijenosno učenje), pokazalo se da je MobileNet jako netočan, dok je HAAR bio djelomično točan jer je bio jako osjetljiv na lažne pozitive. Kako je koncept prijenosnog učenja bio nekoristan, krenulo se u razmišljanje kako poboljšat HAAR detekciju, tj. kako odbaciti sve detekcije koju su zapravo lažni pozitivi. To se izvelo provjerom broja zvijezda u detektiranom području, te korištenjem kliznih parametara kojima se kaskada primjenjuje na slike s različitim veličinama maske i različitim brojem otkrivenih zvijezda. Ovo proširenje detektora je uzrokovalo gotovo 100% točnost prilikom detektiranja zviježđa, jer se gotovo u potpunosti izbjegavaju i odbacuju lažni pozitivi.
Abstract (english)
Before solving the problem described in this paper, it was necessary to prepare the input data, to turn the stars that are dots into something more recognizable. Object classifiers generally have trouble recognizing dot patterns (especially sparse ones such as constellations). Therefore, concentric circles were placed on the stars. After input dana preparation, it was time to choose a concept of machine learning that would be best to solve this paper's problem. Two concepts seemed best: ensemble learning and transfer learning. After training the HAAR cascade (ensemble learning) and the MobileNet model (transfer learning), MobileNet proved to be very inaccurate, while HAAR was partially accurate because it was very sensitive to false positives. As the concept of transfer learning was useless, we started thinking about how to improve the HAAR detection, or perhaps, how to reject all detections that are false positives. This was done by checking the number of stars in the detected area, and using sliding parameters that apply the cascade to images with different mask sizes and different numbers of detected stars. This expansion of the detector has caused almost 100% accuracy when detecting constellations, as false positives are almost completely avoided and rejected.
Keywords
Keywords (english)
Languagecroatian
URN:NBNurn:nbn:hr:179:279808
Study programmeTitle: Computing
Study programme type: university
Study level: graduate
Academic / professional title: magistar/magistra inženjer/inženjerka računarstva (magistar/magistra inženjer/inženjerka računarstva)
Type of resourceText
File originBorn digital
Access conditionsAccess restricted to authenticated users
Terms of useLicence In copyright icon
Created on2023-06-06 10:19:08