Source Code for FedRefiner: Robust Federated Learning with Noisy Label Refinement for Remote Sensing

It supports reproduction of FedRefiner, a federated learning method for remote sensing imagery with noisy labels. The code implements a two-stage training procedure.

The first stage performs robust warm-up with symmetric cross entropy. The second stage uses a dual-view two-component Gaussian mixture model to separate reliable and noisy samples, then combines high-confidence label correction with a learning-to-cluster objective for representation optimization. All sample selection, label correction, and representation optimization are performed locally on each client, and only model parameters are uploaded.