Raspoznavanje visokorizičnih mentalnih stanja pilota na temelju obrade fizioloških signala metodama dubokog učenja

master's thesis

• Title (english): Recognition of pilots' high-risk mental states based on physiological signal processing by deep learning methods
• Author: Sven Nekić
• Mentor: Siniša Popović (mentor)
• Committee member: Siniša Popović (predsjednik povjerenstva)
• Committee member: Alan Jović (član povjerenstva)
• Committee member: Nikolina Frid (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2024-07-10, Croatia

Abstract

U ovom radu primijenjena su tri modela dubokog učenja za klasificiranje visokorizičnih
mentalnih stanja pilota. Ispitane arhitekture su ResNet autoenkoder i klasifikator, ResNet
klasifikator i U-NET 1D. Za procjenu su korištene metrike evaluacije točnosti, preciznosti,
odziva, F1 i Cohen Kappa. ResNet Autoenkoder i klasifikator te U-NET 1D pokazuju vrlo
ograničenu sposobnost klasificiranja korištenjem LFSO (engl. Leave Five Subject Out) unakrsne validacije, dok ResNet klasifikator daje bolje... More rezultate. ResNet klasifikator LFSO
validacijom postigao je najbolju točnost od 68.91% na skupu za validaciju, ali i dalje postoji
velik broj lažno pozitivnih klasificiranja. U-NET 1D model pokazao je loše rezultate korištenjem LFSO validacije, ali znatno bolje rezultate korištenjem K-Fold validacije, posebno
u binarnoj klasifikaciji, gdje postiže točnost od 84.15%. Less

Abstract (english)

In this paper, three deep learning models were applied to classify high-risk mental states
of pilots. The tested architectures are ResNet autoencoder and classifier, ResNet classifier
and U-NET 1D. The evaluation metrics of accuracy, precision, recall, F1 and Cohen
Kappa were used for evaluation. ResNet Autoencoder and classifier and U-NET 1D show
very limited accuracy using LFSO (Leave Five Subject Out) cross-validation, while ResNet
classifier gives better results. The ResNet classifier... More with LFSO validation achieved the best
accuracy of 68.91% on the validation set, but there are still a large number of false positive
classifications. The U-NET 1D model showed poor results using LFSO validation, but significantly better results using K-Fold validation, especially in binary classification, where it
achieves an accuracy of 84.15%. Less

• Keywords: Duboko učenje; visokorizična mentalna stanja; obrada fizioloških signala; komparativna analiza
• Keywords (english): Deep learning; high-risk mental states; physiological signal processing; comparative analysis
• Language: croatian
• URN:NBN: urn:nbn:hr:168:262339
• 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: Open access
• Created on: 2025-02-25 13:37:12