Wheat Lodging Mapping from UAV Imagery Using WheatSegNet with Lodging-Aware Copy-Paste Augmentation

Accurate mapping of wheat lodging from UAV imagery is important for yield-loss assessment and agricultural damage evaluation, but remains challenging due to high spectral similarity, ambiguous boundaries, and complex texture variations between lodged and healthy wheat. To address these issues, we propose WheatSegNet, a novel semantic segmentation network featuring a Wheat Lodging Perception Enhancement (WLPE) block to improve discriminative feature learning for lodging identification. We further introduce a Lodging-Aware Copy-Paste (LACP) augmentation strategy to alleviate class imbalance by selectively transplanting lodging patches based on lodging-pixel ratio. Extensive experiments show that the proposed framework achieves strong performance for UAV-based wheat lodging mapping. In particular, integrating LACP with WheatSegNet improves mIoU by 2.25% and increases the IoU of the lodging category by 12.28%, demonstrating its effectiveness and robustness.