Federated Learning for Wheat Head Segmentation

Federated learning pipeline for wheat head instance segmentation using a modified U-Net.

A thesis project exploring distributed computer vision training across multiple clients without centralizing data.

Highlights:

  • Benchmarked against centralized baselines with IoU and Dice scores up to 95%.
  • Built data transformation and cleaning pipelines for heterogeneous wheat imagery.
  • Used a model-agnostic design to compare FL strategies.

Tools: PyTorch, U-Net, Distributed ML