Digital Commons Network^™

Geospatial Understanding Of Foundation Models: Self-Supervision, Robustness, And Resource-Constrained Learning, Gilbert Kibet Rotich

USF Tampa Graduate Theses and Dissertations

Research on large-scale geospatial understanding from satellite imagery remains relativelyunderexplored. Much of the field still relies on narrow, task-specific pipelines, and there is lim- ited work on developing true foundation models or on systematically evaluating how well these models generalize under domain shift, geographic variation, and real-world scene diversity. This dissertation advances foundation-model pretraining and evaluation for remote sensing along three complementary axes. First, we develop self-supervised multispectral pretrained models for satellite image classification by adapting BYOL, BarlowTwins, DINO, MoCo, and Simsiam to multispectral inputs, addressing the limited amount of curated satellite datasets relative to ImageNet. Pretraining on Eurosat and …

A Multi-Task Cnn-Lstm Model With Attention Mechanism For Climate Temperature Forecasting Over Iraq, Salim M. Mohammed, Omar M. Mustafa, Lailan M. Haji, Omar M. Ahmed

Al-Bahir

Accurate forecasting of climate temperature at the regional scale is crucial to agriculture planning, water resource management, and disaster preparedness. Still, due to the nonlinear and complicated temporal patterns in regions such as Iraq, conventional climate models fail to meet this imperative. This paper presents a novel deep learning model that not only predicts monthly average temperature values but also predicts the temperature trend, hence addressing the regional climate prediction problem holistically. The proposed model is designed as a multi-task deep learning framework integrating Convolutional Neural Networks, Long Short-Term Memory networks, and an attention mechanism to process historical monthly temperature …

Using Deep Learning To Increase Eye-Tracking Robustness, Accuracy, And Precision In Virtual Reality, Kevin Barkevich

Theses

Algorithms for the estimation of gaze direction from mobile and videobased eye trackers typically involve tracking a feature of the eye that moves through the eye camera image in a way that covaries with the shifting gaze direction, such as the center or boundaries of the pupil. Tracking these features using traditional computer vision techniques can be difficult due to partial occlusion and environmental reflections. Although recent efforts to use machine learning (ML) for pupil tracking have demonstrated superior results when evaluated using standard measures of segmentation performance, little is known of how these networks may affect the quality of …

Fusing Channel State Information And Computer Vision In An Agentic Ai Workflow For Accurate Human Activity Recognition And Crisis Prevention, Vishnu S. Pendyala, Milind Patil

Faculty Research, Scholarly, and Creative Activity

This paper presents a novel multimodal framework for Human Activity Recognition (HAR) in homes that combines Channel State Information (CSI) from WiFi access points with visual data from camera images. This approach introduces a dual-stream classification pipeline that leverages a Transformer-based deep learning model for processing CSI features and a computer vision model based on a fine-tuned Convolutional Neural Network (CNN) for image-based activity classification. To enable autonomous decision-making, an agentic AI module based on the Falcon Large Language Model (LLM) has been incorporated, which fuses predictions from both modalities through a reasoning-driven mechanism. This integrated system is capable of …

Vehicle Re-Identification System Using Residual Network With Instance-Batch Normalization, Wei Jie Low, Kah-Ong Michael Goh, Check-Yee Law, Connie Tee, Yong-Wee Sek, Md Ismail Hossen

Journal of Informatics and Web Engineering

Vehicle Re-identification (Re-ID) has become extremely important due to the increasing number of vehicles on the road and its potential to address traffic-related challenges. As a result, there is also a growing need for efficient methods to track and identify vehicles across multiple traffic cameras. One of the biggest challenges of this task is the variations in vehicle appearances across different camera angles. This is because vehicles can appear significantly different when captured from various camera angles and viewpoints. Furthermore, the current vehicle Re-ID solutions typically require extensive coding knowledge, making it inaccessible to many potential users. Therefore, we focus …

Radiology Report Generation Using Deep Learning And Web-Based Deployment For Chest X-Ray Analysis, David Agbolade, Peyman Heydarian, Shakeel Ahmad

Journal of Informatics and Web Engineering

The huge rise in the number of medical images has caused a major problem in radiology departments. Radiologists are now working harder than ever, which affects the quality of their diagnoses and patient care. It takes 15 to 30 minutes to write a manual radiological report for each case, and different people may see things differently. Modern departments process over 230 cases a week, which causes long delays in diagnosis. Automated report generation systems that are already in use have a lot of problems, such as not being able to be interpreted clinically, not having enough Digital Imaging and Communications …

Deep Learning-Based Automatic Detection And Diagnosis Of Tuberculosis From Chest X-Ray Images: A Comprehensive Analysis, Paschal C. Ahanotu, Deborah A. Adedigba, Raza Hasan, Sellappan Palaniappan

Journal of Informatics and Web Engineering

Tuberculosis (TB) continues to be one of the foremost public health issues in the world, and remains the second most salient communicable cause of death after COVID-19. In 2022, TB accounted for 10.6 million new infections and 1.3 million deaths globally. Conventional diagnostic approaches involving sputum smear microscopy, culture assays, and GeneXpert MTB/RIF are characterized by excessive turnaround times, elevated costs, and dependency on specialised infrastructure and trained personnel. These constraints are exacerbated in resource-poor countries, resulting in delayed diagnosis, delayed therapy initiation, and enhanced disease transmission. This work investigates the application of deep learning algorithms to automatically diagnose TB …

Xai -Driven Explainability For Cardiovascular Diseases Prediction, Jacqueline Dike, Jarutas Andritsch

Journal of Informatics and Web Engineering

The adoption of artificial intelligence (AI) in cardiovascular disease prediction has significantly improved risk stratification, offering new avenues for early diagnosis and preventive care. With the growing availability of electronic health records and structured clinical datasets, machine learning (ML) and deep learning (DL) models have demonstrated strong predictive capabilities. However, despite their performance, its adoption in healthcare is often constrained by the lack of transparency and interpretability in many ML and DL models. This lack of explainability undermines clinical trust and raises ethical concerns. In high-stakes domains such as CVD prediction, clinicians require not only accurate outputs but also clear …

Evaluating Accuracy Latency And Robustness Of Face Recognition Models For Real-Time Web Applications, Alani Fatini Sharzizi, Nur Afiqah Sahadun, Abdulkadir Hassan Disina, Harinda Fernando

Journal of Informatics and Web Engineering

Face-recognition technology is one of the most important advancements in the field of computer vision. They play a crucial role in many applications, including biometric authentication, surveillance, online security, and interactive web systems. The use of web-based solutions is increasing continuously. Therefore, accurate and fast recognition models employing few resources are required in real-world applications. However, because of the challenges related to such environments, including the lighting, occlusion, pose, and computing power of client devices, it is difficult to ascertain which model will be most successful in a real-life scenario. The purpose of this research is to compare four deep …

Developing Deep Neural Network Based Brain Computational Models From Psychophysics Data Of Some Simple Perceptual Phenomena: Visual As Well As Auditory, Chandran Keerthi S

Doctoral Theses

The thesis, consisting of nine chapters, explores the methodology of building testable brain computational models using Deep Neural Networks (DNN), which are trained by psychophysics data. Psychophysics is the quantitative study of perception of physical stimuli. In psychophysics experiments, one or more parameters associated with the stimuli are changed, and the human subject’s responses to the stimuli are recorded. This thesis encompasses both experimental psychophysics works, as well as computational models of the phenomena involved. The contributory chapters of the thesis start in Chapter 2 with the perspective building of the novel methodology followed throughout this research involving psychophysics on …

Artificial Intelligence Methods For Circadian Rhythm Recovery And Disease Identification From High-Dimensional Omics Data, Aram Ansary Ogholbake

Theses and Dissertations--Computer Science

High-throughput transcriptomic and proteomic technologies have enabled opportunities for studying biological processes and disease mechanisms. However, extracting meaningful biological information from these high-dimensional datasets remains challenging due to limited sample sizes, biological heterogeneity, measurement noise, and the absence of biological annotations. In particular, many molecular datasets lack temporal information required for circadian analysis, making the study of circadian rhythms difficult. Moreover, disease diagnosis and biomarker discovery from transcriptomic data often rely on complex machine learning models whose predictions are difficult to interpret and may not generalize well across independent datasets and experimental platforms. These limitations motivate the development of artificial …

Security-Oriented Voice Authentication Using Machine Learning, Ifeoluwa Stella Elegbe

College of Graduate Studies: Theses & Dissertations

This study develops and evaluates a machine learning and deep learning-based voice authentication system for secure identity verification. As traditional authentication methods such as passwords, PINs, and security tokens continue to face challenges, including identity theft, forgetting, and unauthorized access, voice biometrics offers a more secure, convenient, and user-friendly alternative, especially for remote, hands-free, and accessibility-focused applications. The study adopts a closed-set speaker identification framework, where the system determines the most likely speaker from a predefined group of enrolled users. A structured methodology is implemented, beginning with audio preprocessing and feature extraction. Key acoustic features, including Mel-Frequency Cepstral Coefficients (MFCCs), …

A Hybrid Object Detection And Temporal Attention Framework For Intelligent Video Surveillance, Sudheer Reddy Bandi, Eswari Vanaparthi, Yakshini Edapalli, Roshan Kavuri

Mansoura Engineering Journal

The rapid growth of urban areas has greatly heightened the need for smart video surveillance systems that can automatically process extensive amounts of CCTV footage. Traditional surveillance methods largely depend on human monitoring, which is not only inefficient but also susceptible to human mistakes, especially in intricate and crowded environments. To tackle these issues, this paper introduces a combined object detection and temporal attention for intelligent video surveillance that concurrently analyzes spatial and temporal data from video streams. The proposed system analyzes real-time CCTV footage utilising a multi-pathway frame extraction technique that includes slow, fast, and full-frame sampling to capture …

Real Time Waste Classification Using Deep Learning: Comparing Mobilenetv2 And Resnet 18 With Transfer Learning And Fine Tuning, Abdul Moaiz

ICT

Waste contamination is a major problem all over the world. The Environmental Protection Agency reports that over two thirds of waste found in general household and commercial bins could have been placed in the recycling or organic waste bins instead in Ireland, with food waste and plastics being the most common misplaced items. This project presents a deep learning solution to classify nine categories of waste from camera images in real time with the goal of helping users sort waste correctly at the source. Following the CRISP DM framework, two convolutional neural network architectures were trained and compared, MobilenetV2 and …

Effective Deep Learning Architectures For Structured Data Analysis And Generation, Md Atik Ahamed

Theses and Dissertations--Computer Science

The effective utilization of structured data is fundamental to modern machine learning, yet it presents distinct challenges in both predictive analysis and generative modeling. Traditional deep learning architectures, particularly Transformers, often suffer from quadratic computational complexity when processing long sequences. This dissertation addresses these limitations by introducing novel architectures based on State-Space Models (SSMs) and Diffusion Models. In the area of predictive analysis, we focus on overcoming the computational bottlenecks of attention mechanisms for tabular and time-series data. First, we introduce MambaTab, a selective state-space architecture designed for efficient tabular classification. By leveraging the linear complexity of SSMs, MambaTab significantly …

Deep Learning Approaches For Voltammetric Analysis Of Coffee, Ryan Koes

Honors Theses

This thesis investigates deep learning approaches for voltammetric analysis of brewed coffee using a low-cost electrochemical system and screen-printed electrodes (SPEs). Traditional analytical methods, such as high-performance liquid chromatography (HPLC) and gas chromatography-mass spectrometry (GC-MS), provide precise quantification of key compounds but require expensive instrumentation and specialized expertise, limiting accessibility. While SPEs offer a more accessible alternative, they yielded poor results with traditional processing; however, when combined with a neural network, the system proved more effective. In experiments with 132 coffee samples, mean errors for caffeine, CGA, and TDS predictions were 52.98 ppm, 70.48 ppm, and 0.08%, respectively. These findings …

Deep Learning For Wireless Communications, Swarada Ajit Kulkarni

Electrical Engineering Dissertations

The rapid evolution of wireless communication imposes stringent requirements for ultra-reliable, low-latency transmission in dynamic, interference-prone environments. Traditional model-driven signal processing struggles to adapt to nonlinear hardware effects, time-varying channels, and complex interference patterns. Deep learning (DL) offers a transformative, data-driven alternative, enabling end-to-end optimization and robust adaptation under uncertain propagation conditions.

This dissertation investigates deep learning architectures for intelligent and resilient wireless communication through three complementary contributions. The first introduces a Vision Transformer (ViT)-based modulation classification framework that leverages self-attention to capture local and global dependencies in spectrogram representations of Quadrature Amplitude Modulation (QAM) signals. The ViT achieves superior …

Real-Time Lidar-Based Vehicle Detection Using Pointpillars For Connected And Automated Vehicle Applications, Jared Townsend Behr

Graduate Theses, Dissertations, and Problem Reports (ETD)

This work explores the implementation of a real-time LiDAR processing model for autonomous vehicle design. Automotive perception systems are becoming increasingly important as Advanced Driver Assistance Systems (ADAS) and Connected and Automated Vehicle (CAV) technologies continue to progress toward higher levels of autonomy. Reliable perception is paramount for autonomous features, such as adaptive cruise control (ACC), where identifying and tracking a lead vehicle is a safety critical task.

Many production ADAS perception systems primarily rely on radar and camera sensors for object detection and sensor fusion. These sensors work well to achieve classification and accurate distance detections; however, they are …

Ai-Based Smart Proxy Models For Flow Assurance, Integrating Oil Rate Prediction, Pipeline Monitoring, And Leak Detection, Ali Sajedian

Graduate Theses, Dissertations, and Problem Reports (ETD)

Oil and gas pipeline networks operate under complex multiphase flow conditions and dynamic transient regimes that present significant challenges for traditional monitoring, anomaly detection, and operational decision-making methodologies. Many existing monitoring approaches rely on high-fidelity physics-based transient flow simulations to accurately represent system behavior. However, these models are computationally intensive and therefore difficult to implement in real-time operational environments.

This dissertation presents a physics-guided artificial intelligence framework for intelligent pipeline monitoring and integrity management based on the concept of Smart Proxy Modeling. By integrating first principles engineering knowledge with machine learning techniques, the proposed framework develops computationally efficient surrogate models …

Dataless Neural Networks For Boolean Satisfiability And Network Optimization, Andrew Evan Gautier

Graduate Theses, Dissertations, and Problem Reports (ETD)

Combinatorial optimization problems (COPs) require searching over a finite solution space subject to constraints, with the goal of satisfying an objective function. They arise in operations research, scheduling, resource allocation, circuit design, and many other fields. Many problems in combinatorial optimization (including satisfiability and network design) are NP-hard. Traditionally, researchers have built approximate solvers that return near- optimal solutions efficiently by developing increasingly sophisticated heuristics and meta- heuristics. Deep learning has provided new opportunities for improving combinatorial solvers by leveraging neural guidance to prune the search space. Traditional neural networks have distinct drawbacks in this context: separate training and …

Computational Methods For Identification Of Molecular Signatures, Weijun Yi

Graduate Theses, Dissertations, and Problem Reports (ETD)

This work develops computational methods for identifying molecular signatures from high-throughput genomic data and for modeling long non-coding RNA (lncRNA) sub-cellular localization. The response of multiple myeloma to CB-6644, a selective RUVBL1/2 complex inhibitor with potential anti-tumor activity, is analyzed to identify drug-responsive pathways and molecular signatures. Conventional gene set enrichment analysis (GSEA) often excludes low-expression genes. Here, phenotype comparison is reformulated as a supervised machine learning problem: genes most informative for discrimination are first selected using a machine learning approach, and GSEA is then applied to these machine-learning derived gene sets. This framework improves detection of CB-6644-associated pathways. For …

Data-Driven Methodologies For Mapping Cultural Heritage: The Case Of The National Coal Heritage Area, West Virginia, Usa, Hossain Mohammad Nahyan

Graduate Theses, Dissertations, and Problem Reports (ETD)

The objective of this dissertation was to develop a comprehensive, data-driven spatial framework for characterizing the complex cultural landscape of the National Coal Heritage Area (NCHA) in West Virginia. By transitioning away from traditional, heuristic spatial mapping, this research integrates advanced spatial statistics, machine learning, and GIS-based methodologies to objectively quantify the physical, visual, and cultural dimensions of the post-mining environment. The research is structured around three interconnected empirical studies, each addressing a specific scale of the Landscape Character Assessment (LCA) framework to support heritage conservation and sustainable spatial planning. The first paper focused on landform classification, developing an automated …

Multimodal Emotion Detection System, Shubhankar Sameer Munshi

Master's Projects

Trying to understand emotion from speech is a problem that is present in human computer interaction. Nevertheless, there are still some shortcomings in current SER methods. Text-based systems may miss vital vocal cues, such as sarcasm, tone changes, and delivery. On the other hand, purely audio-based systems are prone to noise and unstable acoustic features. The combination of linguistic and acoustic features in multimodal approaches partially solves this problem, but many existing approaches use inflexible multimodal fusion techniques that cannot adjust their behaviors according to the quality of input signals. In this work, we propose a multimodal approach based on …

A Framework For Digital Energy Modulation And Translation In Projectional X-Ray Imaging, Richard R. Wargo

Theses and Dissertations

In projectional X-ray imaging, acquisition energy fundamentally influences subject contrast, image noise, and radiation dose. While dual-energy techniques exploit spectral differences to optimize contrast and enable material separation, acquiring multiple images of the same anatomy often requires additional exposures and increased system complexity.

This dissertation presents a framework for digital energy modulation (DEM), in which computational models are trained to translate projection images between energy domains without requiring additional X-ray acquisitions. DEM is conceived as a generalizable strategy for controlled contrast modulation in projection imaging, with potential applications in image optimization, radiation therapy alignment, and dual-energy applications.

The framework was …

Associations Between Contralesional Neuroplasticity And Motor Impairment Through Deep Learning-Derived Mri Regional Brain Age In Chronic Stroke (Enigma): A Multicohort, Retrospective, Observational Study, Gilsoon Park, Mahir Khan, Justin Andrushko, Nerisa Banaj, Michael Borich, Lara Boyd, Amy Brodtmann, Truman Brown, Cathrin Buetefisch, Adriana Conforto, Steven Cramer, Michael Dimyan, Martin Domin, Miranda Donnelly, Natalia Egorova-Brumley, Elsa Ermer, Wuwei Feng, Fatemeh Geranmayeh, Colleen Hanlon, Brenton Hordacre, Neda Jahanshad, Steven Kautz, Mohamed Salah Khlif, Jingchun Liu, Martin Lotze, Bradley Macintosh, Feroze Mohamed, Jan Nordvik, Fabrizio Piras, Kate Revill, Andrew Robertson, Christian Schranz, Nicolas Schweighofer, Na Jin Seo, Surjo Soekadar, Shraddha Srivastava, Bethany Tavenner, Gregory Thielman, Sophia Thomopoulos, Daniela Vecchio, Emilio Werden, Lars Westlye, Carolee Winstein, George Wittenberg, Jennifer Ferris, Chunshui Yu, Paul Thompson, Sook-Lei Liew, Hosung Kim

Department of Radiology Faculty Papers

BACKGROUND: Stroke leads to complex chronic structural and functional brain changes that specifically affect motor outcomes. The brain predicted age difference (PAD) has emerged as a sensitive biomarker of both sensorimotor and cognitive function after stroke. Our previous study showed a higher global brain PAD associated with poorer motor function after stroke. However, the association between local stroke lesion load, regional brain age, and motor impairment is unclear. This study aimed to investigate the associations between focal lesion damage, regional brain PAD in both hemispheres, and motor outcomes in chronic stroke, and to identify key predictors of motor impairment.

METHODS: …

On-Device Artificial Intelligence Solutions With Applications To Smart Environments, Fabrizio De Vita, Dario Bruneo, Sajal K. Das

Computer Science Faculty Research & Creative Works

Recent advances in Artificial Intelligence (AI) and the increasing availability of computational power have accelerated the diffusion of Intelligent Cyber-Physical Systems (ICPSs), enabling smart applications with reasoning capabilities. However, the limited resources of embedded and Edge devices significantly constrain the complexity of deep learning models that can be effectively deployed. Traditional approaches rely on cloud-based training and edge-only inference, a paradigm that becomes inadequate when low latency, privacy, security, and high customization are required. In this context, On-device AI is emerging as a new paradigm in which both training and inference are performed directly on the device, avoiding data transfer …

Real-Time Isolated Asl Recognition: Evaluating Spatial-Temporal Networks And Multimodal Llms, Raga Mouni Batchu

West Chester University Graduate Theses, Dissertations, and Final Projects

This thesis investigates the deployment of high-accuracy Isolated ASL Recognition (ISLR) in resource-constrained edge environments. We train a lightweight Spatio-Temporal Attention Network (SSTAN,∼2.7 M parameters,∼10 MB) on the WLASL-100 benchmark, achieving 75.25% Top-1 and 88.24% Top-5 accuracy with 139 ms CPU-only inference. A systematic comparison against frontier multimodal LLMs (Gemini 3 Flash, Gemini 3.1 Pro, Qwen 3 VL) shows SSTAN outperforms the best LLM baseline by∼1.85×in accuracy while being 22–230×faster and up to 40×cheaper annually. The LLMs’ core limitation is a lack of fine-grained temporal perception; they impose English-language semantic priors rather than learning the articulatory distinctions that define ASL …

A Unified Framework For Evaluating Training Efficiency In Deep (Bayesian) Neural Networks: Metrics, Overtraining, Stopping Criteria, And Grokking Computer Science, Eduardo Cueto Mendoza

Doctoral

Measuring training efficiency for artificial neural networks is an open research problem, current literature reports several attempts to define measures or create reporting frameworks. Current methods lack generality as they require measurements of the hardware or software thus, comparing efficiency between different systems can be difficult. Similarly, current metrics or frameworks generally do not propose the use of the metrics to directly improve training efficiency. This thesis presents three main contributions: (1) a novel framework that quantifies the training efficiency of a neural architecture on a learning task as the average ratio of model accuracy to total energy consumption during …

Identification Of Neural Crest And Melanoma Cancer Cell Invasion And Migration Genes Using High-Throughput Screening And Deep Attention Networks., J C Kasemeier-Kulesa, S Martina Perez, R E Baker, Paul M. Kulesa

Manuscripts, Articles, Book Chapters and Other Papers

BACKGROUND: Cell migration and invasion are well-coordinated in development and disease but remain poorly understood. We previously showed that the neural crest (NC) cell migratory wavefront shares a 45-gene panel with other cell invasion phenomena. To rapidly and systematically identify critical genes, we performed a high-throughput siRNA screen and statistical and deep learning analyses to determine changes in NC- versus non-NC-derived human cell line behaviors.

RESULTS: We find 14 out of 45 genes significantly reduced c8161 melanoma cell migration; four of the 14 genes altered leader cell motility (BMP4, ITGB1, KCNE3, and RASGRP1). Deep learning identified marked disruptions in cell-neighbor …

Deep Learning Assisted Time-Resolved Optical Imaging For Surgical Guidance And Radiotherapy, Shiru Wang

Dartmouth College Ph.D Dissertations

Image-guided therapy enables clinicians to monitor treatment progress, verify accurate delivery, and make immediate adjustments. Among various imaging modalities available for this purpose, optical imaging techniques offer distinct advantages for real-time guidance through their non-ionizing nature and high temporal resolution. However, light propagation in biological tissue fundamentally limits these imaging modalities by obscuring spatial information and introducing noise artifacts. These physical constraints make it hard to accurately interpret subsurface structures and compromise the reliability of real-time guidance.

This thesis develops deep learning methods to overcome these limitations in two distinct optical imaging modalities. The first approach aims to improve time-of-flight …