Climate-robust AI framework restores alfalfa seed maturity assessment across years
· newswise.com · popularity 21/100 · 0 similar stories today · AI & Data Science (Artificial Intelligence)
Read the original at newswise.com →
Summary
A new Earth Mover's Distance (EMD)-guided deep learning framework could make alfalfa seed maturity assessment reliable across climatically different years. The study combines a Multispectral Spatial Attention Network (MSANet) with a diagnose-adapt-finetune strategy. MSANet integrates a three-dimensional convolutional neural network (3D-CNN) backbone, spectral attention, and Transformer-based spatial attention to classify five maturity stages from 19-band multispectral images (365-970 nm). When a model trained in one year is applied to another, accuracy collapses. The proposed framework uses EMD to locate shifted layers, applies EMD-guided adaptive batch normalization (AdaBN), and then performs few-shot fine-tuning. It restores more than 90% accuracy with only 100 labeled seeds per class, offering a data-efficient route to climate-robust seed phenotyping.
Found in: Newswise. Rating basis: single outlet; outlet tier 2 (unlisted); 1 feeds; feed position 1.