Search

Student engagement in large lectures is challenging, and one of our strategies has been to introduce online chatting and polling during lectures. Although our initial experiences with a backchannel chat were positive across smaller STEM and larger Social Sciences courses, it became apparent that student usage might differ across disciplines. As such, this study aimed to directly compared the use of backchannel chat in second year Engineering (ENGG*2400; N=418) and Human Sexuality (FRHD*2100; N=263) courses at the University of Guelph. Kountu, a web-based online chat and polling application, was utilized during each Fall 2016 lecture for both courses. The majority of students (72% of FRHD*2100 and 64% of ENGG*2400) tried the chatroom at least once, with an ANOVA demonstrating that the FRHD*2100 students logged in significantly more often (p < .001; m = 9.53 lectures) than ENGG*2400 students (m = 6.55 lectures). ENGG*2400 students, however, posted (m = 14.34) and ‘liked’ (m = 15.25) significantly more messages per person during each lecture than FRHD*2100 students (posts m = 4.38; likes m = 5.64). Interestingly, an online survey of 79 participants indicated that ENGG*2400 students engaged in significantly more off-topic conversations and found their chatroom to be significantly more distracting during lecture than FRHD*2100 students. Additionally, FRHD*2100 students felt significantly more comfortable participating in their chatroom as compared to ENGG*2400. Using live examples and audience participation in Kountu, we will demonstrate the potential reasons for these differences as well as highlight discipline-specific advantages and limitations of using backchannel chat during lecture. more...

A practical and non-invasive method of calculating the efficiency of electric motors could help reduce anthropogenic green house gas emissions by up to 6%. Such a method has been developed using quantitative thermography, however currently, the time required for its implementation is prohibitive. In this thesis, registration and segmentation techniques are applied to the thermograms of the above method, particularly thermograms used in the lumped capacitance method (LCM) and those used to find the average temperature of motors, reducing the time required to process thermograms. The processing of LCM thermograms was completely automated (±5% difference when compared to results obtained manually) while processing of motor thermograms required the location of the motor be provided manually the first time a motor is examined, but was completely automated for subsequent thermo- grams of the same motor (±0.9°C and ±0.6°C difference for non-occluded and occluded motors respectively compared to manual results). NSERC Chair in Environmental Design Engineering more...

Medical image modalities, such as magnetic resonance (MR), positron emission tomography (PET), computed tomography (CT) enable the study of anatomy and function of animals and humans. The combination of images can often lead to additional clinical information not apparent in the separate images. This research seeks to address the issue of non-rigid multi-modality image registration by developing efficient and robust algorithms which are tested by both simulated and real image data sets. The research consists of two parts: rigid and non-rigid image registration, where rigid registration acts as the first step of non-rigid registration. We proposed a combination of Powell's direction set method (PDSM) with the differential evolution (DE) genetic algorithm for the rigid registration task, which is robust and effective for both synthetic and real image data. Mutual information (MI) allows the algorithm to be efficient for both single- and multi-modality registration task. In addition to the rigid transform, we also developed a fast fluid algorithm using the least mean square (LMS) inverse filtering technique. The test results show that the average CPU speed-up is about 2.5 and the registration accuracy is also improved to some extent (134% for intra-subject and 117% for inter-subject) for the non-rigid mono-modality image registration problem. MI is again chosen as the similarity measure for the non-rigid multi-modality registration in our proposed fast fluid algorithm. A canine brain image data set including both MRI and [mu]CT is built for the validation of the proposed fast fluid algorithm. Moreover, a segmentation-based validation procedure is introduced for the multi-modal registration problem. The results illustrate a speed-up of 2.2 in this case. more...

The development of next-generation speech enhancement systems depends on new forms of speech representation that can capture the non-stationarity of speech signal in a way amiable to element-based signal separation processing. Research in this thesis focuses on developing new structure-revealing signal representations that can meet this challenge. The idea is first investigated in developing a Gabor-atom based time-frequency component signal representation. Problems encountered in this stage of investigation lead to the main focal point of the research: the development of a new chirp-decomposition method suitable for speech-like multi-component signal analysis. Research starts from polynomial phase signal (PPS) analysis, followed by studying the complex ambiguity function (CAF) time-frequency distribution for chirp detection. Two new chirp detection algorithms are proposed. One is the principal-chirp Wigner-Hough (PC-WH) transform and another one is the ambiguity auto-term ridge (AAR) detection algorithm. The potential for using the resulting chirp-graph for speech separation is demonstrated in a new speech enhancement system. more...

Accurate segmentation of ultrasound images to improve clinical diagnosis is an important application studied in medical imaging. In this thesis, we propose a texture-based contour detection method using a hybrid neural network composed of both unsupervised and supervised classifiers. Our goal is to develop a computer aided system to facilitate the diagnosis of the clinical ultra sound tendon images and allow a repeatable calculation of the cross-sectional area of the tendon. A number of feature extraction techniques have been investigated. In terms of efficiency, only rotationally invariant features are adopted. The experimental results show that improvements can be further accomplished by a feature selection scheme. The experimental images were captured from equine legs by using broad band 5-10'MHz' scanning frequency on an 'ATL HDI3000 ' ultrasound machine. Gold standards were created by a clinical expert in sonography. Our results have shown an excellent correspondence between the generated segmentation and the gold standards. more...