| Abstract |
Plant diseases are becoming harder to manage as climate change shifts growing conditions and increases the risk of outbreaks, which can reduce crop yield and threaten food supply. However, building accurate segmentation systems is often limited by the cost and time required to create detailed pixel-level disease masks. LeafSeg addresses this by using a Vision Transformer (ViT) encoder with a U-Net-style decoder to segment diseased regions on plant leaves and quantify severity. By focusing on data-efficient training and strong spatial predictions, the project aims to produce a scalable plant health monitoring approach that can support earlier detection, consistent tracking over time, and future deployment in real agricultural settings.
|
