Carleton University Institutional Repository
Not a member yet
20816 research outputs found
Sort by
An Amodal Segmentation Pipeline for Critical Infrastructure Asset Imaging
No full textAs automated image-based critical infrastructure inspection sees wider adoption, there is a need to alleviate issues and occlusion affecting the quality of road-based infrastructure imagery. We have addressed this need by developing an image acquisition pipeline to provide artifact- and occlusion-free images of a specified infrastructure asset at a given location using a geotagged image dataset. We have developed deep learning models to reliably detect problematic imaging artifacts typical of non-specific image sources and compared their performance with promptable zero-shot models. We have developed a general method to detect the presence of obstructions occluding an asset using state-of-the-art amodal foundational models. Robust performance is observed, resulting in problem-free imaging in 68.5% of cases (up from 43.0%), while requiring an average of 1.72 images per asset. The proposed pipeline will be of interest to disaster response teams, utilities, and other critical infrastructure asset managers
Creating Care-Full True Crime: A Sandbox Methodology
Full text linkIn the last ten years, true crime podcasts have exploded into mainstream pop culture. Often, these stories explore issues of gender-based violence. While many listeners of true crime cite public education and safety as two of the main motivations for listening, the genre itself risks repeating the same issues of mainstream crime reporting when it comes to gendered violence. This thesis explores how true crime creators can avoid these issues and tell more “care-full” true crime stories about gender-based violence through the proposed “sandbox” framework. The Care-Full True Crime Sandbox was developed through a critical discourse analysis of seven serialized podcasts, and in consultation with eight subject matter experts on true crime storytelling. It is the first guide of its kind: A feminist sandbox format for true crime storytelling. The four proposed parameters shape a methodology for producers of true crime to care-fully produce content about gender-based violence
Climate Change Is Burning A Hole In Canadians’ Wallets: The Causal Effect Of A Canadian Wildfire On The Distribution Of Financial Debt
No full textCanada has been experiencing an increase in the occurrence of natural disasters in recent years, resulting in a growing interest on the effect that weather events have on the personal finances of its citizens. While monitoring point estimates of consumer debt can provide useful insights, they are limited in scope and may not capture the full picture of the variations over time. However, tracking individual accounts incurs a high privacy cost. This thesis proposes tracking the distribution of debt over time to balance the consideration of data sensitivity with strong insights in the behaviour of the population. Treating the evolution of densities across time as functional data objects, interest is in the impact of wildfires on debt distributions through a causal inference lens. A synthetic control model is used to account for the counterfactual evolution of debt distribution arising from the 2016 Fort McMurray wildfires
Invisible Media Labour: Participatory Moderation at Refugee Radio Stations in Germany
No full textThis dissertation explores the role of media structures in serving the public sphere. I argue for the concept of participatory moderation (PM) which refers to the labour of media professionals in engaging their audiences to prepare for deliberation on community issues. This includes activities such as organizing, coordinating, facilitating, educating, translating, interpreting federal and local policies, establishing connections within civil society organizations, and building networks that are not typically considered part of media jobs. Using a multi-sited ethnography approach, this study examines the presence of PM in the practices of three refugee radio programs established at community radio stations in Germany, operating in three federal states: Hamburg, Sachsen-Anhalt, and Baden-Württemberg. The study focuses on the use of these media structures by refugees during the post-refugee crisis period, from 2020 to 2023, to ensure their participation in the public sphere and larger conversations related to their problems. It reveals the challenges and inequalities that radio workers encounter in gaining access to the public sphere, leading to additional work to ensure community participation in the public sphere through these media projects. I argue that the efforts of staff media workers and volunteers, which the conceptual framework I develop in this dissertation interprets as activities related to participatory moderation, should be recognized as one of the ‘norms’ for media institutions serving the public sphere in the theories of the public sphere. Acknowledging and appreciating this labour should help challenge the conventional vision of the roles of media structures in deliberative processes. Key words: media and public sphere, deliberation, community radio, refugees, multi-sited ethnography, participatory moderation
Unequal Access : Categorising Refugees in European Resettlement and Humanitarian Admission Programmes
No full textThis book is available in open access to all readers. Please download the book and consult the copyright page and acknowledgements to determine the organizations responsible for financially supporting that access, in addition to McGill-Queen's University Press (https://www.mqup.ca) and Local Engagement Refugee Research Network (https://carleton.ca/lerrn) with the assistance of the Carleton University Library. The series editors would really appreciate hearing about readers' experiences and uses of this edition of the ebook. To share, please write to [email protected] European states tighten their borders, refugees are regularly forced to take costly and highly dangerous routes to seek protection, sometimes with fatal consequences. The resettlement and humanitarian admission programmes that remain allow only a small number of migrants to enter directly from first countries of refuge. With less than 1 per cent of the world’s refugees resettled, such programs are extremely limited, forcing admission states and other actors to prioritize some groups and individuals over others. Unequal Access analyzes these dynamics and the complex boundaries of inclusion and exclusion they produce. Focusing on Europe and programs admitting people to Germany from Lebanon and Turkey, Natalie Welfens explores multilevel policy developments, from the national to the global. She follows the admission chain – from policy formulation, via refugee selection and pre-departure preparations, to refugee reception – and illustrates how policy categories transform based on intersecting social markers such as nationality, gender, and age. Unequal Access reveals the inequalities embedded in the categorization practices of resettlement and humanitarian admission programs, demonstrating how these practices profoundly shape access to protection for refugees
Root Cause Analysis of Frequency Oscillations Observed in Ontario’s Distribution System
Full text linkRecurring frequency oscillations in Ontario’s distribution system have led to multiple false trips of a synchronous generator-based biomass power plant, disrupting operations and raising concerns about system stability. Over an 18-month period, these oscillations, ranging from 5.6 to 6.45 Hz, occurred across various load conditions without any coinciding faults or sudden disturbances. The complexity of the system, coupled with significant uncertainties in system parameters, made traditional modeling approaches impractical. To address this challenge, this thesis introduces a two-phase root cause analysis methodology that integrates data-driven techniques with system modeling. The first phase treats the system as a "black box," utilizing signal processing and feature extraction to identify subtle characteristics of the oscillations without requiring detailed system data. This step enables the narrowing down of potential scenarios responsible for the oscillations. The second phase involves system modeling and sensitivity analysis, leveraging small-signal analysis, the shooting method combined with the Newton-Raphson algorithm, and bifurcation analysis to pinpoint the dominant factors driving instability. The study identifies Hopf bifurcation as a key contributor to the observed oscillations, with systems operating in unity power factor (UPF) mode being particularly susceptible due to reduced load margins compared to those in voltage control (PV) mode. To mitigate the oscillations, control parameters were adjusted based on load margin sensitivity analysis, improving stability across different operating conditions. The proposed solution was validated through Real-Time Digital Simulation (RTDS) and Hardware-in-the-Loop (HIL) testing, demonstrating its effectiveness in eliminating the oscillations and preventing false tripping events. This research contributes a new approach to root cause analysis in power systems with high uncertainty and large-scale complexity. By combining data-driven techniques with targeted modeling, it provides a practical framework for identifying, analyzing, and mitigating oscillations without relying on extensive system data. The findings not only resolve a critical issue in Ontario’s distribution system but also offer insights applicable to broader power system stability challenges
Reconstructing Massive-MIMO Channels From Limited CSI Logs Using Machine Learning
Full text linkThe increasing demand for high-performance wireless communication is fueling the advancement of cutting-edge technologies, including massive Multiple-Input Multiple-Output (m-MIMO) systems. While promising, these systems introduce complex propagation environments that challenge conventional modelling and testing techniques. Accurate simulation of wireless channels is essential for developing and validating new algorithms before deployment. However, existing simulation frameworks often rely on hand-crafted scenarios or a limited set of generalized channel configurations. This thesis presents a machine learning-based framework to reconstruct realistic Clustered Delay Line (CDL) channel models from limited Channel State Information (CSI) logs. Unlike many research approaches that depend on extensive channel data that must be collected with expensive equipment, this practical method relies solely on CSI routinely collected by commercial base stations. It estimates essential channel characteristics, including the line of sight conditions and both large-scale and small-scale parameters, by leveraging synthetic training data and neural networks designed for denoising, classification, and regression tasks. The resulting CDL profiles accurately reflect the true propagation conditions encountered in real environments where the CSI was collected. The reconstructed channels support rapid, simulation-based evaluation of wireless systems and reduce dependence on costly over-the-air (OTA) testing. In addition to diversifying the environments available in CDL-based testing, this framework enables the replication of deployment-specific scenarios to investigate observed failures, assess robustness under realistic channel conditions, and validate algorithmic improvements. Experimental results demonstrate that meaningful channel characteristics can be recovered from the limited information available in practical CSI, with validated performance on synthetic datasets. The proposed framework reduces the dependence on expensive OTA experiments while enhancing diagnostic capabilities by facilitating data-driven channel modelling
Fine-Grained Text-to-Shape Generation via CLIP Latent Space Adaptation
Full text linkGenerating 3D shapes from text is challenging due to the limited availability of text-to-3D datasets. Existing methods train generative models by conditioning on multi-view CLIP [38] image embeddings of shapes during training and use CLIP text embeddings for inference. However, due to the nature of the CLIP’s training dataset, the supported text descriptions lack fine details. In addition, the CLIP text encoder has a limited token length, which restricts its ability to support rich captions. In this paper, we explore supervised fine-tuning of CLIP-based shape generation with richer captions that enhance control over the generation. We introduce a fine-tuning approach that maps the text embedding space to the CLIP image embedding space using diffusion models on the target dataset. The fine-tuned model is tailored to the target dataset while maintaining its effectiveness on open-set captions. We demonstrate the improvements provided by our approach with various evaluations and analyses
Overburdened and Overwhelmed: Employee Wellbeing in High-Demands Organizations During the COVID-19 Pandemic
Full text linkFrontline workers experienced some of the most difficult working conditions of any workers during the COVID-19 pandemic. The objective of this thesis is to examine how the COVID-19 pandemic impacted the wellbeing of workers in high demands organizations. The theoretical framework guiding our research is based on four established theories from the research literature: Lazarus and Folkman’s (1984) transactional model of stress and coping, Hobfoll’s (1989) conservation of resources, Kahn et al.’s (1964) theory of role dynamics, burnout (Maslach et al., 2001) and theories of excessive (Falkenberg et al., 2005) and disruptive change. Our research involves a longitudinal study of the wellbeing of frontline workers who were required to meet the high demands imposed on them by their work and families. Data were collected at two time-points using web-based surveys. The Time 1 survey was completed six months prior to the pandemic. The Time 2 survey was completed one year later. The surveys used well-established measures from the academic literature to quantify the constructs of interest for this thesis (workplace stressors, work and family role overload, coping strategies, burnout, and perceived stress). In this thesis, we put forward three papers to study the mental wellbeing of frontline workers before and during the pandemic. Each paper uses PLS-SEM to examine how the COVID-19 pandemic impacted wellbeing for individuals who were forced to balance high demands from work and family. Paper 1 studies the impact of the pandemic on the relationship between workplace stressors, work role overload and perceived stress. Paper 2 examines how pre-pandemic demands and pre-pandemic coping impacted stress during the pandemic. Paper 3 studies the pandemic’s impact on the relationships of job stress and work role overload to stress and burnout. This thesis contributes to our understanding of (1) the relationships of workplace stressors to work role overload and stress, (2) how workers’ pre-pandemic demands and coping contributed to stress during the pandemic, and (3) the wellbeing of workers during a period of disruptive change, many of whom were stressed and overworked prior to the pandemic due to excessive change
Making Safe Spaces for Inuit Youth: The Role of Inuit-led Cultural Programming in Supporting Inuit Youth Mental Wellness
Full text linkWhile there is extensive research on ongoing mental health crises in Inuit Nunangat and how it can be addressed, there is very little academic literature that explores Inuit-led approaches to supporting mental wellness in Inuit communities. This dissertation, in partnership with the Arctic Rose Foundation’s (ARF) Messy Book Program, explores an Inuit-led, arts-based afterschool program that aims to support the well-being of Inuit/Northern Indigenous youth. In doing so, his dissertation seeks to answer: Do Inuit-led youth programs, like the Messy Book Program, contribute to an increased sense of belonging and cultural connection? In turn, do these feelings improve mental health in Inuit youth, and if so, how? This thesis, involving fieldwork in Rankin Inlet, engaged 13 individuals who work(ed) or partner(ed) with the Messy Book Program through in-person and online interviews, as well as nine community members in a focus group, between 2022 and 2024. In building an understanding of the ARF’s programming through these discussions, as well as through ongoing engagement with the ARF, this dissertation outlines the significance, impacts and limitations of four core components of the Messy Book Program: Cultural Cognizance, Safe Space, Youth Mentorship, and Inuit-Led Expressive Arts. These concepts are analyzed through the lens of Inuit Qaujimajatuqangit and Inuit-specific wellness frameworks with the objective to contextualize the Messy Book Program within broader community wellness goals. Central to this thesis, the research partnership with the ARF allowed me unique insight into the growth and adaptation of the Messy Book Program from the start of our research partnership in 2021 through to the conclusion of fieldwork. This thesis therefore captures snapshots of the Messy Book Program at different periods of time – as the program navigated the COVID-19 pandemic, opened (and closed) program sites, and expanded to new regions and Indigenous cultural contexts. Given the fast growth of the program to respond to mental health crises in Inuit and Northern Indigenous communities, the goal of this dissertation is to present these snapshots as a means of positioning Inuit-led programs, like the Messy Book Program, as an integral aspect of community mental health initiatives