Scholarlink

Accurate classification of rice growth stages is essential for effective decision-making in precision agriculture. This study investigates the performance of deep learning models, with a focus on ConvNeXt and SegFormer, in classifying five primary rice growth stages—seedling, tillering, jointing, heading, and ripening—using RGB images captured by unmanned aerial vehicles (UAVs). Three datasets with varying spatial coverage and resolutions were constructed to evaluate model sensitivity to input scale. ConvNeXt consistently achieved the best performance across all datasets, reaching the highest accuracy (0.860) and F1-score (0.857) on the high-coverage dataset. SegFormer also delivered competitive results, particularly when more spatial context was retained. These findings emphasize the critical role of model architecture and spatial coverage in capturing subtle phenological variations among rice growth stages. These findings suggest that integrating multispectral data with RGB imagery in future research may further enhance the accuracy and robustness of UAV-based classification of crop growth stages.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Evaluating Spatial Coverage in UAV-Based Rice Growth Stage Classification with Deep Learning Models

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Van-Hoa Nguyen,
  • Duy-Phuong Nguyen-Ly,
  • Thanh-Nghi Do

摘要

Accurate classification of rice growth stages is essential for effective decision-making in precision agriculture. This study investigates the performance of deep learning models, with a focus on ConvNeXt and SegFormer, in classifying five primary rice growth stages—seedling, tillering, jointing, heading, and ripening—using RGB images captured by unmanned aerial vehicles (UAVs). Three datasets with varying spatial coverage and resolutions were constructed to evaluate model sensitivity to input scale. ConvNeXt consistently achieved the best performance across all datasets, reaching the highest accuracy (0.860) and F1-score (0.857) on the high-coverage dataset. SegFormer also delivered competitive results, particularly when more spatial context was retained. These findings emphasize the critical role of model architecture and spatial coverage in capturing subtle phenological variations among rice growth stages. These findings suggest that integrating multispectral data with RGB imagery in future research may further enhance the accuracy and robustness of UAV-based classification of crop growth stages.