Public defence in Bioelectronics and Instrumentation, M.Tech. Ajinkya Gorad

The title of the thesis: Machine learning and state space methods for healthcare, speech, and maritime awareness 

Thesis defender: Ajinkya Gorad
Opponent: Prof. Heidi Kuusniemi, University of Vaasa, Finland
Custos: Prof. Simo Särkkä, Aalto University School of Electrical Engineering, Department of Electrical Engineering and Automation

This thesis explores the use of modern machine learning and probabilistic state-space methods to extract useful structured information from noisy, time-varying sensory data in diverse real-world applications. The work spans three key areas: non-contact respiratory monitoring using thermal imaging, isolated speech recognition with spiking neural networks, and multi-modal ship and sea ice detection for maritime awareness. 

In healthcare, the thesis presents a novel method for analyzing breathing patterns using nasal temperature changes captured via a thermal camera, enabling non-invasive respiratory monitoring. In speech recognition, spiking neural networks and cochlear-inspired preprocessing are used to classify spoken commands, along with the introduction of a new memory performance metric for such networks. 

For maritime applications, the thesis demonstrates bearing estimation of ships using acoustic foghorn signals, and ship and sea ice tracking through a fusion of visible and thermal imagery. Deep learning and Kalman filter-based methods are applied for semantic segmentation and object tracking under challenging Arctic conditions. 

Key methodological contributions include automatic differentiation-based parameter estimation in extended Kalman filters, and the development of a deep Rao-Blackwellized Monte Carlo data association particle filter that integrates visual appearance, position, and motion cues. 

The research is supported by extensive data collection from real-world maritime environments (e.g., MS Megastar, icebreaker Sampo), lab-based respiratory studies, and publicly available datasets. The results advance sensor fusion and tracking techniques in health, speech, and environmental monitoring, demonstrating how unified sensory modeling can be deployed in practical, multi-domain systems.

Key words: liquid state machines, respiratory monitoring, deep object tracking, maritime awareness, thermal imaging

Thesis available for public display 10 days prior to the defence at Aaltodoc. 

Contact information: 

email: ajinkya.gorad@aalto.fi
webpage: https://users.aalto.fi/~gorada2/