Adversarial Image Detection in Agriculture
- Classifies high-fidelity adversarial plant images; attributes source models (StyleGAN2/3, DS8, BLIP, Pix2Pix) alongside health status.
- Trains EfficientNet-B0 / ResNet-50 / CLIP-ViT on binary health, source (Real/GAN/Diffusion), and detailed 10-way crop–health–generator tasks.
- Designed to scale as new generators emerge; targets cyber-biosecurity for Agriculture 4.0 systems.
- Authors: Md Nazmul Kabir Sikder, Mehmet Oguz Yardimci, Trey Ward, Shubham Laxmikant Deshmukh, Feras A. Batarseh (Preprint, 2025).
