Exploring the mysteries of plant disease severity through multimodal learning
Peer-Reviewed Publication
Below are the latest research news from KeAi Communications Co., Ltd..
Updates every hour. Last Updated: 20-Oct-2026 18:16 ET (20-Oct-2026 22:16 GMT/UTC)
Researchers in China developed AgriMamba, a text-guided multimodal framework for plant disease severity grading in complex field environments. The framework combines disease images with pathological descriptions of lesion color, shape, edge characteristics, and spatial distribution. It follows a coarse-to-fine process that localizes the target leaf, segments lesion regions, and quantifies disease severity using lesion-to-leaf area ratios. Experimental and visualization results showed that AgriMamba achieved competitive performance compared with several representative deep learning methods. The model also progressively focused on target leaves and lesions despite background interference, leaf occlusion, and small or scattered disease regions. These findings provide a new technical approach for automated plant disease assessment in complex field environments.
Researchers in China functionally characterized the branching habit regulatory gene AhPILS in peanut through MutMap analysis. This gene encodes a PIN-like protein that mediates polar auxin transport, thereby establishing an auxin gradient at the base of lateral branches. The formed auxin gradient potentially promotes cell elongation, which facilitates the morphological transition of peanut lateral branches from a spreading growth habit to an upright growth habit. This study identifies a novel functional gene, reveals key insights into the molecular regulatory mechanism of peanut branching habit, and provides elite germplasm resources for the molecular breeding of peanut cultivars with ideal plant architecture.
Researchers in China uncovered how cold priming leaves a lasting molecular memory in wheat, helping plants cope better when chilling returns. They found that the recovery phase is a critical window in which stress-responsive genes are reprogrammed and histone modifications are reshaped, while primed plants maintain better photosystem performance and show a reduced need for photosystem repair during later chilling. The findings reveal a key role for post-stress recovery in maintaining cold memory and provide new insight into how wheat adapts to repeated low-temperature stress.
Researchers have identified rainbow trout blood as a promising platform for monitoring the biological hazards of biodegradable microplastics. By combining advanced fluorescence imaging with single-cell analysis, the study reveals how these particles interact with circulating blood cells and disrupt immune-related processes. The findings also highlight an overlooked concern: environmental aging may alter and potentially intensify the harmful effects of plastics designed to biodegrade.
Subcutaneous immunization with live influenza virus in mice provided durable protection against lethal flu, outperforming intramuscular or inactivated vaccines. The virus remained restricted to the injection site and lymph nodes without causing lung damage. A single dose gave 100% protection for 277 days through strong germinal center responses.
Researchers developed SwiVirScan, a screening platform containing 36,416 peptides from 140 swine-associated viruses. In 127 adult breeding pigs, every serum sample showed antibody reactivity to peptides assigned to multiple viruses. The study also identified a persistent association between antibody signals for porcine bufavirus and porcine circovirus 3. The findings can guide follow-up surveillance, but antibody signals alone do not establish active infection.
This review evaluates China's first basin-scale aquatic ecosystem monitoring pilot in the Yangtze River Basin and demonstrates that shifting from chemical water-quality compliance to integrated, classification-based ecosystem assessment can support measurable ecological recovery. The review shows how tailoring indicators and targets to distinct rivers, lakes, and reservoirs across the basin can help guide targeted restoration and produce tangible results.
Dogs spontaneously follow the walking direction of a human figure represented by just a few moving dots, yet show no such directional bias when the dots depict another dog. In a study published in Cognitive Science Research, researchers found that dogs respond to directional cues from human biological motion as readily as they do to human head turns, suggesting that domestication may have tuned canine social perception to human movement.