We have been working on a deep learning pipeline to detect crop diseases from drone imagery. Using YOLOv8 fine-tuned on our custom dataset of wheat and rice diseases. Our current model achieves 89% mAP on the test set but struggles with early-stage infections. Has anyone worked on similar problems? Would love to discuss: 1. Data augmentation strategies for small lesion detection 2. Transfer learning from ImageNet vs medical imaging datasets 3. Deployment on edge devices (Raspberry Pi / Jetson Nano) Happy to share our dataset and pretrained weights.