Rank-Aware Quantile Activation for Motion-Robust Crop Segmentation in UAV Imagery
Researchers have developed a new technique called Dual Quantile Activation (QAct) to improve semantic segmentation in drone imagery affected by motion blur. This method replaces standard magnitude gating with instance-level rank normalization, enhancing the identification of rare, texture-dependent crop classes. Evaluations on the Agriculture-Vision 2021 dataset showed QAct consistently improved mean Intersection over Union (mIoU) scores compared to traditional ReLU activation, particularly in zero-shot and blur-supervised scenarios. AI
IMPACT Improves accuracy in drone-based agricultural monitoring by enhancing segmentation of critical crop features.