Researchers have developed a new AI model called AT-ViT for recognizing plant traits in images of herbarium specimens. The model uses a combination of cross-attention and multi-scale patching to focus on the plant's morphology rather than background elements. This approach improves accuracy and robustness compared to other models, particularly when dealing with synthetic background perturbations. --- Why it matters: This matters because accurate plant trait recognition is essential for plant sciences, but current methods often rely on non-plant cues due to shortcut learning. AT-ViT's ability to focus on plant morphology could improve generalization and interpretability in plant classification tasks. Source: https://arxiv.org/abs/2608.21067