Publication Date
Spring 2026
Degree Type
Master's Project
Degree Name
Master of Science in Computer Science (MSCS)
Department
Computer Science
First Advisor
Nada Attar
Second Advisor
Vuthea Chheang
Third Advisor
Philip Heller
Keywords
Computer Vision, Deep Learning, Faster R-CNN, Image Enhancement, Mean Average Precision
Abstract
Environmental coastal erosion poses significant risks to archaeological landscapes, necessitating the need for a robust preservation system. However, the manual effort required to classify and digitize archaeological artifacts is too time-consuming and limits the scalability of heritage preservation. This project displays the development of a computer vision pipeline used to detect and classify artifacts for coastal site documentation. The dataset consists of images collected across the 28 landscape sections along the southeastern coast of Nevis Island, located in the Caribbean Sea. A Faster R-CNN model is trained on six artifact categories such as Bone, Pottery, Lithic, Building, Glass, and Shell. Due to variation in outdoor lighting conditions on archaeological sites, the pipeline explores image preprocessing techniques to improve performance. Additionally, Mean Average Precision (mAP) at different detection thresholds is used as an evaluation metric to determine model accuracy. The results show improved performance when using image preprocessing techniques, particularly for smaller objects, demonstrating an effective approach for monitoring climate-vulnerable archaeological landscapes.
Recommended Citation
Pandey, Mithi, "Computer Vision Pipeline for Archaeological Artifact Detection Using Faster R-CNN" (2026). Master's Projects. 1750.
DOI: https://doi.org/10.31979/etd.cjjj-dsv8
https://scholarworks.sjsu.edu/etd_projects/1750
