Računalno raspoznavanje hrane na fotografijama

master's thesis

• Title (english): Computer Food Image Recognition
• Author: Domagoj Polančec
• Mentor: Igor Mekterović (mentor)
• Committee member: Igor Mekterović (predsjednik povjerenstva)
• Committee member: Ljiljana Brkić (član povjerenstva)
• Committee member: Boris Vrdoljak (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2018-07-10, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

U sklopu ovog rada implementiran je sustav za prepoznavanje i raspoznavanje hrane. Sustav se sastoji od web stranice, web poslužitelja, programa za klasifikaciju i programa za treniranje. Za prepoznavanje i raspoznavanje modela predložen je pristup baziran na dubokom učenju. Za model koji se koristi i za problem prepoznavanja i raspoznavanja hrane korištena je Xception arhitektura s manjim modifikacijama. Sam proces učenja uključivao je ručno podešavanje hiperparaemtara optimizacije tijekom učenja. Ovakav pristup je značajno poboljšao točnost modela. Pokušano je dalje poboljšati model uvođenjem hijerarhijski ulančanih modela, ali taj pristup nije davao zadovoljavajuće rezultate. Pokazano je da se minimalnom modifikacijom postojeće arhitekture, korištenjem augmentacije skupa podataka i ručno upravljanog treniranja mogu dobiti vrlo dobri rezultati čak i s ograničenim računalnim resursima.

Abstract (english)

As a part of this thesis a system for food detection and recognition in images was implemented. The system consists of a web page, web server, classification program and training program. The thesis suggests a deep learning based approach for food image detection and recognition. The model that was used is a slightly modified Xception architecture. The training process itself included manually adjusting optimizer hyperparameters during training based on model performance. This approach proved to be very effective in improving the accuracy of the model. An attempt was made to improve the model further by introducing hierarchical models, however that approach did not yield satisfactory results. It was shown that by slightly modifying the existing architecture, using various data augmentation techniques and manually controlling training hyperparameters very good results can be achieved with limited computer resources.

• Keywords: spraoznavanje hrane; duboko učenje; konvolucijske neuronske mreže
• Keywords (english): Food Recognition; Deep Learning; Convolutional Neural Networks
• Language: croatian
• URN:NBN: urn:nbn:hr:168:688282
• 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: 2019-03-14 00:04:32