AAI_2025_Capstone_Chronicles_Combined

WildScan: A Semi-Automated AI Pipeline for Wildlife Detection, Classification, and Continuous Learning

Tyler Clinscales, Geoffrey Fadera, Edwin Merchan University of San Diego August 13, 2025

Introduction

WildScan addresses a critical operational challenge in wildlife conservation: the overwhelming amount of image data generated by motion-triggered trail cameras and the inefficiencies associated with manually reviewing these images. The problem we are solving is how to automate and continuously improve the detection and classification of animals in trail camera images to support large-scale conservation efforts. Specifically, our system detects animals in raw images through object detection models, then classifies each detected object by species using Convolutional Neural Networks (CNNs) (Morris, 2023; Liu et al., 2024). This project is crucial because it automates a time-consuming essential task for biodiversity monitoring, habitat analysis, and anti-poaching efforts. As human encroachment and climate change continue to threaten ecosystems, tools like WildScan can help researchers and conservationists to respond rapidly and effectively with data-driven insights. By automating the detection and classification process, WildScan provides near real-time analysis, reduces human labor, and minimizes errors in wildlife monitoring workflows (Dolan et al., 2020; Liu et al., 2024). The primary end users of the WildScan AI system include wildlife researchers, field ecologists, conservation scientists, and environmental policymakers. These individuals rely on accurate, large-scale monitoring to inform conservation strategies and assess the effectiveness of interventions across protected areas or wildlife corridors (Leorna & Brinkman, 2022; Böhner et al., 2023).

279