Abstract
Multimodal freight transportation moves most goods across the United States, with trucks and railroads as the dominant modes. However, growing freight activity has raised concerns about environmental impacts, infrastructure strain, and public health, especially in communities near major corridors. Existing data systems often lack the spatial and temporal detail as well as timeliness needed to monitor truck and rail freight operations effectively. This dissertation developed advanced sensing and machine learning methods for high-resolution freight monitoring. It introduces approaches that combine infrastructure-based sensors with deep learning for accurate freight vehicle identification. To further reduce reliance on manual annotations, the study investigated an automated prompt refinement technique using vision-language models. These contributions will help fill critical data gaps and advance the development of more sustainable and intelligent freight systems.