Comparison of Deep Learning Frameworks For Rice Disease Mapping From UAV Multispectral Imaging
Researchers compared various deep learning frameworks for mapping rice disease severity using UAV multispectral imagery. The study evaluated architectures like U-Net, U-Net++, DeepLabV3+, and SegFormer, testing them with different input configurations including vegetation indices. U-Net++ with EfficientNet-B3 demonstrated the highest performance with a 97.62% mIoU, suggesting that lightweight CNNs are more reliable for operational disease monitoring. AI
IMPACT Lightweight CNNs show promise for operational disease monitoring, potentially improving agricultural efficiency.