This study presents a computer vision–based system for automatically classifying trailer loading depths and assessing manifest compliance for Roadrunner Transportation using a limited dataset of images. To address data constraints, we employed a transfer learning framework based on DenseNet121 pretrained on ImageNet, enabling effective feature extraction and improved generalization. Input images were standardized and preprocessed prior to model training. We deployed the resulting model through a FastAPI backend with a Gradio-based interface and hosted the system on Railway to ensure scalability and accessibility. The proposed framework evaluated shipping manifests according to configurable business criteria, producing automated pass/fail assessments. Experimental results demonstrated strong predictive performance despite limited training data, underscoring the effectiveness of the approach and the importance of standardized image acquisition. We further discuss recommendations for future work, including structured labeling at the point of data collection and the implementation of continuous feedback mechanisms to enhance model accuracy and generalizability.
Watch the team present this project at 07:45 in the session recording here.
Keywords: computer vision, deep learning, transportation, logistics, image processing, transfer learning, trucking, model deployment
Faculty Advisor
Dr. Agar’s work spans applied AI and data science, with focus areas in supply chain, digital manufacturing, and physical AI. Her work focuses on translating emerging technologies into practical applications for complex operational environments.
She brings more than a decade of industry experience in analytics, optimization, digital transformation, and AI product development. Her experience includes leadership roles at Caterpillar, where she has led initiatives across digital manufacturing, supply chain, enterprise data, and analytics, including AI-enabled simulation and digital twin applications.
Dr. Agar has taught at the University of Chicago since 2019 and currently teaches Machine Learning, Python for Analytics, Supply Chain Optimization, and Capstone courses. Her teaching focuses on connecting technical foundations with real-world problems through hands-on work, industry applications, and practical decision-making.
Dr. Agar holds a Ph.D. and M.Sc. in Industrial Engineering from the University of Oklahoma, as well as B.Sc. degrees in Industrial Engineering and Computer Engineering from Cankaya University. She is also an inventor on a U.S. patent and has received Caterpillar’s CEO Award and Outstanding Achievement in Analytics Award.