Dual-Guided Asymmetric MP-Former Advances Rice Root Instance Segmentation
· newswise.com · popularity 18/100 · 0 similar stories today · Machine Learning Research (Artificial Intelligence)
Read the original at newswise.com →
Summary
<img src="https://www.newswise.com/legacy/image.php?image=https://media.newswise.com/pio/articles/858174/6af05dcc7f9d1_27_858174.jpg&width=100&height=150" alt="Newswise image" />Accurate root phenotyping is essential for understanding crop growth, resource use, and yield, but rice roots are slender, overlapping, and frequently occluded, making individual roots hard to separate. This study introduces Dual-Guided Asymmetric MP-Former (DGA-MP-Former), an instance segmentation framework for rice root phenotyping. Built on MP-Former (Mask-Piloted Transformer), the model adds a Guided-Enhancement Pixel Decoder (GEPD) and an Asymmetric Dual-Query Decoder (ADQD). It also presents the Rice Root Segmentation Dataset (RRSD), containing 343 high-resolution, instance-level annotated images. DGA-MP-Former reached 57.2% average precision (AP) at Intersection-over-Union (IoU) thresholds 0.50-0.95 and 87.4% AP0.5, enabling reliable area and perimeter measurement for high-throughput root phenotyping.
Found in: Newswise. Rating basis: single outlet; outlet tier 2 (unlisted); 1 feeds; feed position 10.