Computer Sciences Commons^™

Cellphone-Acoustics Based Suas Detection And Tracking, Ryan D. Clendening

Theses and Dissertations

Small Unmanned Aerial Systems (sUAS) are an easily accessible technology that has become an increasingly large threat to US critical systems. This threatening technology demands using fault-tolerant, low-cost, replaceable, and accurate sensing resources, which counter the ubiquitous nature of sUAS [1]. Therefore, the methods developed in this thesis detect and track sUAS using easily accessible sensing resources, such as cellphones. First, we develop an acoustics sensor network-based sUAS detection methodology. In the latter effort, a deep learning model is trained using the acoustics data from the data collection to predict sUAS range from a cellphone. Combined, these two efforts demonstrate …

Entering Hyperspace: Conditional Hyperspectral Reflectance Image Generation Using Convolutional Neural Networks, Bret M. Wagner

Theses and Dissertations

The field of remote sensing continues to expand in both commercial and defense domains. Development of advanced space based EOIR sensors has driven corresponding demand for sensor data for algorithm development. The AFIT Sensor and Scene Emulation Tool (ASSET) produces realistic synthetic electro-optical and infrared (EO/IR) data with absolute truth for the purpose of clutter suppression, target detection, and tracking algorithm development. This thesis presents a novel model which transforms panchromatic images into realistic hyperspectral reflectance images. The direct application of this model is to allows users to generate hyperspectral background images as inputs to ASSET allowing users to benefit …

Automated Registration Of Titanium Metal Imaging Of Aircraft Components Using Deep Learning Techniques, Nathan A. Johnston

Theses and Dissertations

Studies have shown a connection between early catastrophic engine failures with microtexture regions (MTRs) of a specific size and orientation on the titanium metal engine components. The MTRs can be identified through the use of Electron Backscatter Diffraction (EBSD) however doing so is costly and requires destruction of the metal component being tested. A new methodology of characterizing MTRs is needed to properly evaluate the reliability of engine components on live aircraft. The Air Force Research Lab Materials Directorate (AFRL/RX) proposed a solution of supplementing EBSD with two non-destructive modalities, Eddy Current Testing (ECT) and Scanning Acoustic Microscopy (SAM). Doing …

Analysis And Optimization Of Contract Data Schema, Franklin Sun

Theses and Dissertations

agement, development, and growth of U.S Air Force assets demand extensive organizational communication and structuring. These interactions yield substantial amounts of contracting and administrative information. Over 4 million such contracts as a means towards obtaining valuable insights on Department of Defense resource usage. This set of contracting data is largely not optimized for backend service in an analytics environment. To this end, the following research evaluates the efficiency and performance of various data structuring methods. Evaluated designs include a baseline unstructured schema, a Data Mart schema, and a snowflake schema. Overall design success metrics include ease of use by end …

Developing And Assessing A Generalized Serious Game That Supports Customized Joint All-Domain Operations Related Learning Objectives, Jonathan D. Moore

Theses and Dissertations

As the threat of near-peer adversaries has increased, the DoD has increased its emphasis on Joint All-Domain Operations (JADO). This emphasis on JADO highlights the need for hands-on training that can engage military members at all levels. The serious game Battlespace Next (BSN) was designed to teach high-level JADO concepts by modeling real-world military assets in the context of a strategic card game. To keep pace with the evolving landscape of warfare as well as fit the needs of a variety of Department of Defense (DoD) communities, this research introduces the Battlespace Next Education Framework (BSNEF). The BSNEF allows JADO …

Improving Accessibility And Efficiency Of Analytic Provenance Tools For Reverse Engineering, Caleb W. Richardson

Theses and Dissertations

Reverse engineering is a vital technique for identifying and mitigating cyber threats. Yet, despite its importance, reverse engineering is a time-consuming process. Provenance tools help to improve the workflow of reverse engineers by providing an accessible method of viewing their flow through a binary. The current state-of-theart provenance tool for reverse engineering software called SensorRE, leverages an external server, web browser, and a large array of javascript libraries. This thesis presents Provenance Ninja, a software reverse engineering tool developed in Python that runs directly within Binary Ninja. Provenance Ninja captures reverse engineers’ provenance data and provides an interactive graph within …

Classifying Emotions And Toxicity On Audio To Text Signals From Videos On Youtube, Connice Trimmingham

Theses and Dissertations

Recent technological advancements have pushed humans past the boundaries of a computer screen. They have facilitated human-computer-interaction that were once inconceivable including emotional and audio engagement. Users are now able to utilize technology in more human-like ways and instantaneously transfer depth of opinions and emotions. This global prominence of modern technology, specifically with social media, has spawned a new norm, in which it is now a reasonable expectation of encountering high online toxicity on social media platforms. In addition, the complexities of human emotions make it challenging to determine whether toxic comments trigger certain emotions or vice versa. This study …

Real-Time Face Mask Detection And Recognition For Video Surveillance Using Deep Learning Approach, Ehsan Nasiri

Theses and Dissertations

Facial recognition refers to the determination of identity of an individual based on facial features. The application of facial recognition has been widely used for the purpose of security. The outbreak of coronavirus has made people wear masks as a preventive measure. However, this affects the proper identification of individuals resulting in several security threats. In this paper, the issues related to recognition of both masked and unmasked faces are mitigated and an approach for accurate detection of face masks and recognition of both masked and unmasked faces is Introduced. The proposed approach comprises several processes which are carried out …

Deep Learning Architectures For Visual Question Answering On Medical Images, Venkat Ramana Kodali Kodali

Theses and Dissertations

The purpose of this research is to apply both computer vision and natural language processing techniques for visual question answering (VQA) on a medical image dataset. Deep learning and machine learning libraries were used in the research. The research includes understanding key achievements in the field of visual question answering, identifying techniques applied in general images and applying them along with new techniques to medical images. There are many more articles explaining the application of visual question answering to general images than on applying VQA specifically to medical images. In this research, I initially developed a model of VQA that …

A Parameter Discovery Process For The Data Washing Machine Created For Unsupervised Data Curation, Kris E. Anderson

Theses and Dissertations

The Data Washing Machine (DWM) is a known and documented open-source Python Jupyter Notebook project that is the foundation for an Unsupervised Data Curation process. The DWM ingests reference data without a prior data cleansing activity and ultimately runs Entity Resolution (ER) on acceptable entity data to cluster duplicate references within the dataset. The DWM currently has 17 modifiable parameters that are used to help tokenize, cleanse, organize, link and cluster like references. With such a large number of parameters, some type of beginning settings as optimal as possible are needed for the DWM process for it to be useful …

Multimodal Image Retrieval Combining Image And Text, Md Imran Sarker

Theses and Dissertations

Retrieving visual or textual similarities from an image query and vice versa has drawn much interest in computer vision. With the growth of the E-commerce marketplace, image retrieval provides excellent competitive opportunities for vendors and customers through a robust recommendation system. Feature integration has always been an essential task for multimodal-based image retrieval approaches. However, different existing matching techniques have been used separately for visual and text similarity. Still, researchers are looking for new methods when it comes to multimodal Image Retrieval. In this paper, I study the image retrieval task, where the input query is an image plus text …

Engineering A Scalable Deep K-Means Algorithm On Spark And Tensorflow, Skyler Thompson

Theses and Dissertations

Due to modern data collection practices, datasets can be both high dimensional and very large in size. With the production of these large, high dimensional datasets comes the task of analyzing them. Deep neural networks are often applied to high dimensional datasets in a supervised manner where a network is trained using labeled datasets and then applied to other, similar datasets. However, labeled datasets are often expensive to produce and are not always available for training purposes in all scenarios. For unlabeled datasets, unsupervised learning using data clustering is often the learning method of choice. Deep clustering is a relatively …

An Enhanced Cloud-Native Deep Learning Pipeline For The Classification Of Network Traffic, Ahmed Sobhy Elkenawy

Theses and Dissertations

In a rapidly changing world, the way of solving real-world problems has changed to leverage the power of the advancements in multiple fields. Cloud-native computing approaches can be utilized with deep learning techniques to provide solutions in several important areas. For instance, with the emergence of the pandemic, much dependence on modern technologies came out as a replacement for face-to-face interaction. Deep learning can reach a high level of accuracy, which makes it very effective in the support of modern services and technologies. However, there are some challenging issues because deep learning requires many large-scale experiments, which demand a lot …

Skeleton-Based Hand Gesture Recognition Using Data-Level Fusion, Oluwaleke Yusuf

Theses and Dissertations

Hand Gesture Recognition (HGR) is a form of perceptual computing that allows artificial systems to capture and interpret human gestures. HGR has applications in human-machine interaction, virtual reality, augmented reality, and human behavior analysis. The human hand can assume a near-infinite number of poses and orientations to form myriad gestures, thus increasing the difficulty of the HGR task.

The hand skeleton of connected joints effectively describes the hand’s geometric shape and thus contains richer semantic gesture information while eliminating noise from individual differences in physical hand characteristics. The efficacy and computational efficiency of skeleton-based HGR frameworks can be significantly enhanced …

The Application Of Graph Technology For Improving Entity Resolution Results In The Context Of Group Membership, Md Abdus Salam Siddique

Theses and Dissertations

The main objective of Entity resolution (ER) is to find duplicate records within the same data table from the same source or different data tables from various sources. A traditional pair-wise supervised entity resolution matching depends on pre-built rules for finding matched records. On the other hand, unsupervised or semisupervised also relies on pair-wise matching. In the maximum case, group membership is left behind for consideration. In this dissertation, I have discussed the design, implementation and evaluation of a graph-based entity resolution for group membership to enhance the pair-wise matching ER system. I have designed and implemented a pipeline for …

Machine Learning Models Interpretability For Malware Detection Using Model Agnostic Language For Exploration And Explanation, Ikuromor Mabel Ogiriki

Theses and Dissertations

The adoption of the internet as a global platform has birthed a significant rise in cyber-attacks of various forms ranging from Trojans, worms, spyware, ransomware, botnet malware, rootkit, etc. In order to tackle the issue of all these forms of malware, there is a need to understand and detect them. There are various methods of detecting malware which include signature, behavioral, and machine learning. Machine learning methods have proven to be the most efficient of all for malware detection. In this thesis, a system that utilizes both the signature and dynamic behavior-based detection techniques, with the added layer of the …

Towards An Unsupervised Bayesian Network Pipeline For Explainable Prediction, Decision Making And Discovery, Daniel Mallia

Theses and Dissertations

An unsupervised learning pipeline for discrete Bayesian networks is proposed to facilitate prediction, decision making, discovery of patterns, and transparency in challenging real-world AI applications, and contend with data limitations. We explore methods for discretizing data, and notably apply the pipeline to prediction and prevention of preterm birth.

Material Extrusion-Based Additive Manufacturing: G-Code And Firmware Attacks And Defense Frameworks, Haris Rais

Theses and Dissertations

Additive Manufacturing (AM) refers to a group of manufacturing processes that create physical objects by sequentially depositing thin layers. AM enables highly customized production with minimal material wastage, rapid and inexpensive prototyping, and the production of complex assemblies as single parts in smaller production facilities. These features make AM an essential component of Industry 4.0 or Smart Manufacturing. It is now used to print functional components for aircraft, rocket engines, automobiles, medical implants, and more. However, the increased popularity of AM also raises concerns about cybersecurity. Researchers have demonstrated strength degradation attacks on printed objects by injecting cavities in the …

Enhancing Neuromorphic Computing With Advanced Spiking Neural Network Architectures, Paolo Gabriel Alejandro Cachi Delgado

Theses and Dissertations

This dissertation proposes ways to address current limitations of neuromorphic computing to create energy-efficient and adaptable systems for AI applications. It does so by designing novel spiking neural networks architectures that improve their performance. Specifically, the two proposed architectures address the issues of training complexity, hyperparameter selection, computational flexibility, and scarcity of neuromorphic training data. The first architecture uses auxiliary learning to improve training performance and data usage, while the second architecture leverages neuromodulation capability of spiking neurons to improve multitasking classification performance. The proposed architectures are tested on Intel's Loihi2 neuromorphic chip using several neuromorphic datasets, such as NMIST, …

Development Of Tangible Code Blocks For The Blind And Visually Impaired, Hyun Woo Kim

Theses and Dissertations

The fields of Science, Technology, Engineering, and Mathematics (STEM) have been growing at an accelerating rate in recent times. Knowing how to program has become one key skill for entering all of these STEM fields. However, many students find programming difficult. The block based programming language, Scratch, was specifically designed to lower hurdles to learning how to program for sighted students. Unfortunately, although very effective and widely used in K12 classrooms, Scratch, similar to other block based languages, is inaccessible to students who are blind and visually impaired (BVI). This thesis is part of a larger project to make the …

Innovations In Drop Shape Analysis Using Deep Learning And Solving The Young-Laplace Equation For An Axisymmetric Pendant Drop, Andres P. Hyer

Theses and Dissertations

Axisymmetric Drop Shape Analysis (ADSA) is a technique commonly used to determine surface or interfacial tension. Applications of traditional ASDA methods to process analytical technologies are limited by computational speed and image quality. Here, we address these limitations using a novel machine learning approach to analysis. With a convolutional neural network (CNN), we were able to achieve an experimental fit precision of (+/-) 0.122 mN/m in predicting the surface tension of drop images at a rate of 1.5 ms^-1 versus 7.7 s^-1, which is more than 5,000 times faster than the traditional method. The results are validated on real images …

Face Anti-Spoofing And Deep Learning Based Unsupervised Image Recognition Systems, Enoch Solomon

Theses and Dissertations

One of the main problems of a supervised deep learning approach is that it requires large amounts of labeled training data, which are not always easily available. This PhD dissertation addresses the above-mentioned problem by using a novel unsupervised deep learning face verification system called UFace, that does not require labeled training data as it automatically, in an unsupervised way, generates training data from even a relatively small size of data. The method starts by selecting, in unsupervised way, k-most similar and k-most dissimilar images for a given face image. Moreover, this PhD dissertation proposes a new loss function to …

Methods For Drone Trajectory Analysis Of Bottlenose Dolphins (Tursiops Truncatus), Jillian D. Bliss

Theses and Dissertations

With the increase in the use of UAS (Unmanned Aerial Systems) for marine mammal research, there is a need for the development of methods of analysis to transform UAS high resolution video into quantitative data. This study sought to develop a preliminary method of analysis that would quantify and present a way to visualize the dynamics and relative spatial distribution and changes in distribution of bottlenose dolphins (Tursiops truncatus) in the waters of Turneffe Atoll, Belize. This approach employs a previously developed video tracking program ‘Keypoint Tracking’ that enables manual tracking of individual dolphins and the creation of …

Assessing Wood Failure In Plywood By Deep Learning/Semantic Segmentation, Ramon Ferreira Oliveira

Theses and Dissertations

The current method for estimating wood failure is highly subjective. Various techniques have been proposed to improve the current protocol, but none have succeeded. This research aims to use deep learning/semantic segmentation using SegNet architecture to estimate wood failure in four types of three-ply plywood from mechanical shear strength specimens. We trained and tested our approach on custom and commercial plywood with bio-based and phenol-formaldehyde adhesives. Shear specimens were prepared and tested. Photographs of 255 shear bonded areas were taken. Forty photographs were used to solicit visual estimates from five human evaluators, and the remaining photographs were used to train …

Augmented Reality Fonts With Enhanced Out-Of-Focus Text Legibility, Mohammed Safayet Arefin

Theses and Dissertations

In augmented reality, information is often distributed between real and virtual contexts, and often appears at different distances from the viewer. This raises the issues of (1) context switching, when attention is switched between real and virtual contexts, (2) focal distance switching, when the eye accommodates to see information in sharp focus at a new distance, and (3) transient focal blur, when information is seen out of focus, during the time interval of focal distance switching. This dissertation research has quantified the impact of context switching, focal distance switching, and transient focal blur on human performance and eye fatigue in …

Representation Learning For Open Set Recognition And Novel Category Discovery, Jingyun Jia

Theses and Dissertations

As machine learning models have achieved great success in various research and industry fields, the success of these models heavily relies on the massive amount of data collection and human annotations. While the real world is an open set, the daily emerged categories and the lacking of annotations have become new challenges for machine learning models. The absence of newly emerged categories in training samples can be captured by Open Set Recognition (OSR). Then, given the newly emerged samples, the process of automatically identifying the novel categories is called Novel Category Discovery (NCD). In this dissertation, we focused on learning …

Atomlbs: An Atom Based Convolutional Neural Network For Druggable Ligand Binding Site Prediction, Md Ashraful Islam

Theses and Dissertations

Despite advances in drug research and development, there are few and ineffective treatments for a variety of diseases. Virtual screening can drastically reduce costs and accelerate the drug discovery process. Binding site identification is one of the initial and most important steps in structure-based virtual screening. Identifying and defining protein cavities that are likely to bind to a small compound is the objective of this task. In this research, we propose four different convolutional neural networks for predicting ligand-binding sites in proteins. A parallel optimized data pipeline is created to enable faster training of these neural network models on minimal …

Evaluating Deep Learning Explanations On Risc-V Assembly As A Reverse Engineering Aid, Daniel F. Koranek

Theses and Dissertations

This dissertation addresses several problems surrounding the detection of malware using deep learning models trained on assembly language examples. First, it examines the feasibility of detecting examples of malice using deep learning models trained on RISC-V instruction traces. Next, it examines whether models for detecting trace features and code features in RISC-V assembly can be made explainable (providing rationale for a model’s decision based upon the model’s internal workings) or interpretable (providing additional rationale as model output to support a human’s agreement with the model output). Third, this work examines ways in which it is possible to give additional contextual …

Design Of Ethical Autonomous Agents For Unmanned Aerial Vehicles Using Fuzzy Logic, Gavin Giovanni Smith

Theses and Dissertations

Autonomous systems have, over the years become part of our everyday lives. These systems have been deployed to executed a diverse range of applications in different industries; finance, healthcare, military, and in particular, the flight industry. With the rise of UAVs, new opportunities arose, but with those opportunities came new pitfalls within any industry. For UAVs, one of the pitfalls came in the form of ethical decisionmaking, which led to a variety of questions. Can the Autonomous systems within UAVs be designed with ethics in mind? Which ethical guidelines would we use to implement such a system? How would we …

Non-Negative Matrix Factorization In The Identification Of Co-Mutations, Michael Robert Kolar

Theses and Dissertations

One of the difficulties of genetic research is the asymmetrical relationship between data collection techniques and data analysis techniques. The goal of this research was to test a novel application of non-negative matrix factorization, which would allow researchers to more easily identify co-mutations. Those co-mutations then can then be further verified by frequency analysis. This pruning process allows researchers to identify more fruitful research opportunities, saving time, energy, and funding. Past research has utilized non-negative matrix factorization to extract factors which meaningfully express underlying data features. This study extends the depth of non-negative matrix factorization knowledge in various ways. First, …