Unified Deep Learning Techniques For Spatial Detection And Temporal Forecasting Across Visual Domains

Author

• John Olawale Olamofe, Prairie View A&M University

Date of Award

5-2026

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Degree Discipline

Electrical Engineering

Abstract

This dissertation proposed a unified deep learning framework for spatial detection and temporal forecasting across visual domains, designed to address limited supervision, class imbalance, and scale variability. The framework was structured around four complementary principles: Domain-Aware Input Rebalancing, Representation-Centric Learning, Diversity-Driven Robustness, and Transfer Across Scale and Modality, which together enabled robust visual representation learning across heterogeneous data sources.

Domain-Aware Input Rebalancing mitigates non-uniform and sparse data distributions by actively reshaping inputs before learning via class-aware augmentation, sampling, resolution manipulation, and super-resolution. This principle underpins object detection in overhead satellite imagery (xView), dense urban aerial scenes (CADOT), temporally sparse NDVI signals, and low-contrast microscopy images, where sensitivity to small or rare structures is critical. Representation-Centric Learning emphasizes shared feature encoders as the primary mechanism for generalization. YOLO-based spatial encoders serve as the backbone for object detection across satellite, aerial, and microscopy domains, while sequence encoders and pretrained transformers are employed for NDVI forecasting. By prioritizing transferable representations over task-specific heuristics, the framework supports both spatial localization and temporal sequence modeling within a unified learning paradigm. In xView, scale-and class-dependent augmentation effects revealed complementary learning behaviors, while CADOT explicitly exploits multiple YOLOFocus variants through ensemble fusion to improve robustness under clutter and visual ambiguity. For NDVI, model-family comparisons under few-shot settings further demonstrated the stabilizing role of diversity. Finally, Transfer Across Scale and Modality enabled the framework to generalize learned principles beyond individual tasks. Spatial learning strategies inform temporal NDVI forecasting, pretrained sequence models adapt to environmental time series, and super-resolution enhances cavity detection in microscopy, confirming scale-invariant behavior across modalities. Collectively, this dissertation demonstrated that spatial detection and temporal forecasting can be unified as data-efficient representation learning problems, providing a principled framework for robust visual intelligence in data-constrained environments.

Index Terms—Aerial imagery, computer vision, data augmentation, deep learning, ensemble learning, Faster Region-based Convolutional Neural Network (Faster R-CNN), frozen pretrained transformer (FPT), image super-resolution, machine learning, normalized difference vegetation index (NDVI), object detection, Patch Time Series Transformer (PatchTST), reservoir computing, satellite imagery, time series forecasting, transformers, transmission electron microscopy (TEM), You Only Look Once (YOLO).

Committee Chair/Advisor

Lijun Qian

Committee Member

John Fuller

Committee Member

Xiangfang Li

Committee Member

Xishuang Dong

Committee Member

Ram Ray, Anthony Hill

Publisher

Prairie View A&M University

Rights

Date of Digitization

7/13/2026

Contributing Institution

John B Coleman Library

City of Publication

Prairie View

MIME Type

Application/PDF

Recommended Citation

Olamofe, J. (2026). Unified Deep Learning Techniques For Spatial Detection And Temporal Forecasting Across Visual Domains. Retrieved from https://digitalcommons.pvamu.edu/pvamu-dissertations/139