Engineering Commons^™

Gabor Filter Initialization And Parameterization Strategies In Convolutional Neural Networks, Long Pham

Electronic Thesis and Dissertation Repository

Convolutional neural networks (CNN) have been widely known in literature to be extremely effective for classifying images. Some of the filters learned during training of the first layer of a CNN resemble the Gabor filter. Gabor filters are extremely good at extracting features within an image. We have taken this as an incentive by replacing the first layer of a CNN with the Gabor filter to increase speed and accuracy for classifying images. We created two simple 5-layer AlexNet-like CNNs comparing grid-search to random-search for initializing the Gabor filter bank. We trained on MNIST, CIFAR-10, and CIFAR-100 as well as …

Design, Implementation And Evaluation Of A Redundancy Management System For Fault-Tolerant Wireless Devices In Harsh Environments, Madison Mccarthy

Electronic Thesis and Dissertation Repository

Wireless sensor networks (WSNs), when deployed in harsh environments, can fail prematurely due to elevated rates of component failures. To counteract this problem, fault-tolerant techniques, such as redundancy, may be used. A redundant design requires a management system. Built-in tests (BITs) are one of the most commonly used approaches for managing redundancy, but it suffers from issues such as imperfect fault coverage and common-cause failures (CCFs). In this work, a BIT based redundancy management system has been designed that makes use of a supervisory unit and a modular architecture to address issues with imperfect fault coverage and CCFs. The design …

Development And Assessment Of Signal Processing Algorithms For Assistive Hearing Devices, Farid Moshgelani

Electronic Thesis and Dissertation Repository

Speech identification in the presence of background noise is difficult for children with auditory processing disorder and adults with sensorineural hearing loss. The listening difficulty arises from deficits in their temporal, spectral, binaural, and/ or cognitive processing. Given the lack of improvement with conventional assistive hearing devices, alternate speech processing methodologies, which exaggerate the temporal and spectral cues, need to be developed to improve speech intelligibility for individuals who have poor temporal and/ or spectral processing.

This thesis first, reports results from a series of experiments on subjective and objective assessments of two different schemes of envelope enhancement algorithms (dynamic …

Investigation Of Radiation-Hardened Design Of Electronic Systems With Applications To Post-Accident Monitoring For Nuclear Power Plants, Qiang Huang

Electronic Thesis and Dissertation Repository

This research aims at improving the robustness of electronic systems used-in high level radiation environments by combining with radiation-hardened (rad-hardened) design and fault-tolerant techniques based on commercial off-the-shelf (COTS) components. A specific of the research is to use such systems for wireless post-accident monitoring in nuclear power plants (NPPs). More specifically, the following methods and systems are developed and investigated to accomplish expected research objectives: analysis of radiation responses, design of a radiation-tolerant system, implementation of a wireless post-accident monitoring system for NPPs, performance evaluation without repeat physical tests, and experimental validation in a radiation environment.

A method is developed …

Autonomous And Real Time Rock Image Classification Using Convolutional Neural Networks, Alexis David Pascual

Electronic Thesis and Dissertation Repository

Autonomous image recognition has numerous potential applications in the field of planetary science and geology. For instance, having the ability to classify images of rocks would allow geologists to have immediate feedback without having to bring back samples to the laboratory. Also, planetary rovers could classify rocks in remote places and even in other planets without needing human intervention. In 2017, Shu et. al. used a Support Vector Machine (SVM) classification algorithm to classify 9 different types of rock images using a with the image features extracted autonomously. Through this method, they achieved a test accuracy of 96.71%. Within the …