Oak Wilt UAV Aerial Imagery Dataset for Machine Learning Classification (Oak Wilt vs. Not Oak Wilt)

This dataset was developed to support research in the Early Detection of Oak Wilt using Machine Learning and Unmanned Aerial Vehicles (UAVs). It contains two image classes: Oak Wilt (366 images): aerial RGB images of oak trees showing early-stage symptoms of oak wilt infection, collected and validated with expert input from the Michigan Department of Natural Resources. Not Oak Wilt (396 images): aerial RGB images of healthy oak trees and other environments such as roads, grass, cemeteries, and non-oak vegetation. The dataset was curated from UAV flights conducted in Michigan state parks (Lake Forest Cemetery, Mulligan’s Hollow, P.J. Hoffmaster State Park and Warren Dunes State Park) in 2024, supplemented by imagery from Minnesota collections (2022). UAV flights were carried out at 70–100 m altitude using multispectral/RGB sensors, providing high-resolution (5280×3956 px) imagery suitable for canopy-level analysis. This dataset was used in training lightweight convolutional neural networks (CNNs) that achieved 86.72% accuracy in detecting oak wilt, as reported in the 2025 ICMLC proceedings paper (Bismoy et al., 2025). The dataset is intended for machine learning applications in forestry, plant pathology, and computer vision and supports reproducibility of research on detecting forest diseases using UAVs.If you like this research or you are using the dataset, here's the DOI : 10.5281/zenodo.17109752