Oak wilt, caused by the fungal pathogen Bretziella fagacearum, severely affects oak ecosystems across North America, resulting in extensive ecological and economic losses. Manual field inspections remain the primary method for detection; however, their labor intensity and delayed diagnosis hinder early intervention. This chapter presents a UAV-based imaging and lightweight convolutional neural network (CNN) framework enhanced with transformer-based attention and reinforcement learning from human feedback (RLHF) for scalable, early detection of oak wilt. The model was trained and validated on a total of 1,051 UAV-acquired RGB images across four Michigan parks and a publicly available hyperspectral dataset (Sapes et al. 2022), achieving an overall accuracy of 87.4% and an inference latency of 121 ms per image on CPU hardware, confirming real-time deployability. The hybrid EfficientNet-Lite + Vision Transformer (ViT) model demonstrated consistent generalization across multi-park and multi-season datasets, with mean accuracy exceeding 83% under cross-park testing and it sustains 121 ms/image CPU inference for real-time field use, trading modest top-line accuracy for deployability and energy efficiency. Integration of RLHF improved oak wilt recall by 5.3% through iterative feedback cycles. A web-based visualization platform, developed using VueJS and Flask, enables interactive map-based disease monitoring, expert validation, and continuous retraining from field data. By emphasizing generalization, computational efficiency, and adaptive learning, this framework transitions oak wilt detection from controlled laboratory experiments to a reproducible, field-ready monitoring system suitable for large-scale forest health management and state-level conservation initiatives.