Učenje pod nadzorom korištenjem SSD i SVM metoda za nedestruktivna ispitivanja

master's thesis

• Title (english): Supervised Learning Using SSD and SVM Methods for Nondestructive Testing
• Author: Duje Medak
• Mentor: Sven Lončarić (mentor)
• Committee member: Sven Lončarić (predsjednik povjerenstva)
• Committee member: Marko Subašić (član povjerenstva)
• Committee member: Tomislav Petković (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2019-07-10, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Glavni cilj ovog rada je ispitati mogućnost korištenja dviju metoda za detekciju objekata u svrhu detekcije defekata na slikama dobivenim ultrazvučnim testiranjem. Prva predložena metoda bazirana je na histogramu orijentacija gradijenata (HOG) i stroju potpornih vektora (SVM). Druga metoda bazirana je na dubokoj konvolucijskoj mreži Single Shot Detector (SSD). Na početku rada dan je uvod u nedestruktivna testiranja i objašnjen je princip rada učestalo korištenih tehnika ultrazvučnog testiranja. Prikazani su i načini reprezentacije podataka i dan je pregled postojećih metoda za automatsku detekciju defekata. Nakon detaljnog opisa predloženih metode prikazana je i njihova primjena u detekciji defekata na B-prikazima. Metode su testirane na 98 slika iz testnog skupa i kombinacija HOG-a i SVM-a postigla je prosječnu preciznost 23.5% dok je SSD modelom postignuta prosječna preciznost od 84.5%.

Abstract (english)

The main goal of this work was to test the usage of two object detection methods for the task of defect detection from images acquired in the process of ultrasonic testing. The first introduced method is based on the histogram of oriented gradients (HOG) and support vector machines (SVM). The second method is based on deep convolutional network Single Shot Detector. At the beginning of the work, an introduction to nondestructive testing is given and the principle of two commonly used ultrasonic testing techniques is explained. Various types of data representation are shown and overview of the previous work in the area of automatic defect detection of ultrasonic data is given. After providing detail description of the proposed methods, their application in the task of defect detection from B-scans is shown. The methods were tested on 98 test images and the combination of HOG and SVM achieved an average precision of 23.5% while SSD achieved average precision of 84.5%.

• Keywords: obrada slika; analiza slika; ultrazvučne slike; nedestruktivno ispitivanje; automatska detekcija defekata; detekcija objekta; histogram orijentacija gradijenata; stroj potpornih vektora; duboke konvolucijske mreže; Single Shot Detector
• Keywords (english): image processing; image analysis; ultrasonic imaging; non-destructive testing; automated flaw detection; object detection; histogram of oriented gradijents; support vector machine; deep convolutional networks; Single Shot Detector
• Language: croatian
• URN:NBN: urn:nbn:hr:168:304113
• Study programme: Title: Computing; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka računarstva (magistar/magistra inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2020-01-07 20:16:45