AAI_2025_Capstone_Chronicles_Combined

Our goal is to create a scalable and modular AI pipeline capable of detecting and classifying wildlife species, with a temporal feedback loop for continuous model updates. Unlike traditional object detection and classification systems—which assess each image independently and lack the ability to recognize evolving trends or shifts—our approach integrates ongoing performance monitoring as new data become available. Our hypothesis is that by combining temporal awareness to object detection and classification, we can develop a more adaptable and accurate model over time (Zhu et al., 2022). The final product will be a containerized, cloud-deployable system that provides real-time wildlife analytics, featuring a dashboard to visualize key metrics such as model accuracy, F1 score, and species distribution across locations (Leorna & Brinkman, 2022).

Dataset Summary & Exploratory Data Analysis (EDA)

The WildScan system is built using the Caltech Camera Traps (CCT20) dataset, which includes over 250,000 motion-triggered images from wildlife trail cameras across more than 140 locations (Tabak et al., 2019). The dataset is structured with metadata such as image IDs, file names, animal species labels,

281