Digital Commons Network^™

Tomato Flower Detection And Three-Dimensional Mapping For Precision Pollination, Kaitlyn Mckensie Nelms

Masters Theses

It is estimated that nearly 75% of major crops have some level of reliance on pollination. Humans are reliant on fruit and vegetable crops for many vital nutrients. With the intensification of agricultural production in response to human demand, native pollinator species are not able to provide sufficient pollination services, and managed bee colonies are in decline due to colony collapse disorder, among other issues. Previous work addresses a few of these issues by designing pollination systems for greenhouse operations or other controlled production systems but fails to address the larger need for development in other agricultural settings with less …

Power System Stability Assessment With Supervised Machine Learning, Mirka Mandich

Masters Theses

Power system stability assessment has become an important area of research due to the increased penetration of photovoltaics (PV) in modern power systems. This work explores how supervised machine learning can be used to assess power system stability for the Western Electricity Coordinating Council (WECC) service region as part of the Data-driven Security Assessment for the Multi-Timescale Integrated Dynamics and Scheduling for Solar (MIDAS) project. Data-driven methods offer to improve power flow scheduling through machine learning prediction, enabling better energy resource management and reducing demand on real-time time-domain simulations. Frequency, transient, and small signal stability datasets were created using the …

Improving Convolutional Neural Network Robustness To Adversarial Images Through Image Filtering, Natalie E. Bogda

Masters Theses

The field of computer vision and deep learning is known for its ability to recognize images with extremely high accuracy. Convolutional neural networks exist that can correctly classify 96\% of 1.2 million images of complex scenes. However, with just a few carefully positioned imperceptible changes to the pixels of an input image, an otherwise accurate network will misclassify this almost identical image with high confidence. These perturbed images are known as \textit{adversarial examples} and expose that convolutional neural networks do not necessarily "see" the world in the way that humans do. This work focuses on increasing the robustness of classifiers …

Identifying Smokestacks In Remotely Sensed Imagery Via Deep Learning Algorithms, Kenneth Moss

Masters Theses

Locating smokestacks in remote sensing imagery is a crucial first step to calculating smokestack heights, which allows for the accurate modeling of dioxin pollution spread and the study of resulting health impacts. In the interest of automating this process, this thesis examines deep learning networks and how changes in input datasets and network architecture affect image detection accuracy. This initial image detection serves as the first step in automated object recognition and height calculation. While this is applicable to general land use classification, this study specifically addresses detecting smokestack images. Different dataset scenarios are generated from the massive Functional Map …

Classification Of Bacterial Motility Using Machine Learning, Yue Ma

Masters Theses

Cells can display a diverse set of motility behaviors, and these behaviors may reflect a cell’s functional state. Automated, and accurate cell motility analysis is essential to cell studies where the analysis of motility pattern is required. The results of such analysis can be used for diagnostic or curative decisions. Deep learning area has made astonishing progresses in the past several years. For computer vision tasks, different convolutional neural networks (CNN) and optimizers have been proposed to fix some problems. For time sequence data, recurrent neural networks (RNN) have been widely used.

This project leveraged on these recent advances to …

A Neuroscience-Inspired Approach To Training Spiking Neural Networks, James Michael Ghawaly Jr.

Masters Theses

Spiking neural networks (SNNs) have recently gained a lot of attention for use in low-power neuromorphic and edge computing. On their own, SNNs are difficult to train, owing to their lack of a differentiable activation function and their inherent tendency towards chaotic behavior. This work takes a strictly neuroscience-inspired approach to designing and training SNNs. We demonstrate that the use of neuromodulated synaptic time dependent plasticity (STDP) can be used to create a variety of different learning paradigms including unsupervised learning, semi-supervised learning, and reinforcement learning. In order to tackle the highly dynamic and potentially chaotic spiking behavior of SNNs …

Deep Reinforcement Learning For Real-Time Residential Hvac Control, Evan Mckee

Masters Theses

The model-free Deep Reinforcement Learning (DRL) environment developed for this work attempts to minimize energy cost during residential heating, ventilation, and air conditioning (HVAC) operation. The HVAC load associated with heating and cooling is an ideal candidate for price optimization through automation for two reasons: Its power footprint in a typical home is sizeable, and the required level of participation from an inhabitant is passive. HVAC is difficult to accurately model and unique for every home, so online machine learning is used to allow for real-time readjustment in performance. Energy cost for the cooling unit shown in this work is …

Improving Manufacturing Data Quality With Data Fusion And Advanced Algorithms For Improved Total Data Quality Management, David Juriga

Masters Theses

Data mining and predictive analytics in the sustainable-biomaterials industries is currently not feasible given the lack of organization and management of the database structures. The advent of artificial intelligence, data mining, robotics, etc., has become a standard for successful business endeavors and is known as the ‘Fourth Industrial Revolution’ or ‘Industry 4.0’ in Europe. Data quality improvement through real-time multi-layer data fusion across interconnected networks and statistical quality assessment may improve the usefulness of databases maintained by these industries. Relational databases with a high degree of quality may be the gateway for predictive modeling and enhanced business analytics. Data quality …

Heat Flux Model Validation Utilizing Convolutional Neural Networks And Sub-Surface Thermocouples For Nstx-U, Thomas Patrick Looby

Masters Theses

A proof of concept convolutional neural network (CNN) has been developed to assist in operating tokamaks outside of existing empirical scalings for the heat flux width, λ_q [lambda-q]. NSTX-U has designed new plasma facing components (PFCs) to withstand increased halo current forces as well as elevated heat fluxes driven by increased poloidal field and neutral beam power compared to NSTX. Larger graphite tiles are castellated to 2.5 cm [centimenter] x 2.5 cm [centimeter] to reduce bending stresses. Maintaining PFCs below engineering limits will be an important consideration for operation of NSTX-U. Sub-surface thermocouples will be utilized to demonstrate validation …

Danna2: Dynamic Adaptive Neural Network Arrays, John Parker Mitchell

Masters Theses

Traditional Von Neumann architectures have been at the center of computing for decades thanks in part to Moore's Law and Dennard Scaling. However, MOSFET scaling is rapidly approaching its physical limits spelling the end of an era. This is causing researchers to examine alternative solutions. Neuromorphic computing is a paradigm shift which may offer increased capabilities and efficiency by borrowing concepts from biology and incorporating them into an alternative computing platform.The TENNLab group explores these architectures and the associated challenges. The group currently has a mature hardware platform referred to as Dynamic Adaptive Neural Network Arrays (DANNA). DANNA is a …

Computational Analysis Of Upper Extremity Movements For People Post-Stroke, Zachariah Edward Nelson

Masters Theses

Wearable sensors have been beneficial in assessing motor impairment after stroke. Individuals who have experienced stroke may benefit from the use of wearable sensors to quantify and assess quality of motions in unobserved environments. Seven individuals participated in a study wherein they performed various gestures from the Fugl-Meyer Assessment (FMA), a measure of post-stroke impairment. Participants performed these gestures while being monitored by wearable sensors placed on each wrist. A series of MATLAB functions were written to process recorded sensor data, extract meaningful features from the data, and prepare those features for further use with various machine learning techniques. A …

Primary User Emulation Attacks In Cognitive Radio - An Experimental Demonstration And Analysis, Benjamin James Ealey

Masters Theses

Cognitive radio networks rely on the ability to avoid primary users, owners of the frequency, and prevent collisions for effective communication to take place. Additional malicious secondary users, jammers, may use a primary user emulation attacks to take advantage of the secondary user's ability to avoid primary users and cause excessive and unexpected disruptions to communications. Two jamming/anti-jamming methods are investigated on Ettus Labs USRP 2 radios. First, pseudo-random channel hopping schemes are implemented for jammers to seek-and-disrupt secondary users while secondary users apply similar schemes to avoid all primary user signatures. In the second method the jammer uses adversarial …