Open Internet by MindsNet

Improving Self-Supervised Learning for Hyperspectral Crop Stress Classification

Achieving low accuracy (~50%) with self-supervised learning (SSL) methods (BYOL, MAE, VICReg) on hyperspectral crop stress data for nitrogen deficiency detection. The challenge is to enhance the performance of SSL methods for hyperspectral data, which seems to be stuck due to potential issues with class separability, augmentation strategies, and model architectures.

Computing & Technology, Computer Science, Machine Learning