Scholarlink

Rice panicle segmentation plays a critical role in precision agriculture. Focusing on its indispensability, in this paper, we propose a multi-stage training pipeline that leverages the customized U-Net with EfficientNet-B3 and pseudo-labeling to improve segmentation accuracy. The datasets are prepared through the acquisition of video clips of rice crops, captured from different fields under diverse environmental conditions. The model is evaluated on internal and external test sets, demonstrating the effectiveness of the staged training and synthetic data augmentation. Comparing the performance of the proposed model with the strongly performed nnU-Net model shows the superiority of our proposed model in terms of Dice score and Intersection over Union. This highlights the impact of pseudo-labeling in improving the segmentation accuracy.

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

Optimizing Rice Panicle Segmentation using a Multi-stage Training Pipeline Aggregated with Pseudo-labeling

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Harnoor Kaur Khehra,
  • Farnaz Sheikhi,
  • Alan Bach,
  • Elijah Mickelson,
  • Farhad Maleki

摘要

Rice panicle segmentation plays a critical role in precision agriculture. Focusing on its indispensability, in this paper, we propose a multi-stage training pipeline that leverages the customized U-Net with EfficientNet-B3 and pseudo-labeling to improve segmentation accuracy. The datasets are prepared through the acquisition of video clips of rice crops, captured from different fields under diverse environmental conditions. The model is evaluated on internal and external test sets, demonstrating the effectiveness of the staged training and synthetic data augmentation. Comparing the performance of the proposed model with the strongly performed nnU-Net model shows the superiority of our proposed model in terms of Dice score and Intersection over Union. This highlights the impact of pseudo-labeling in improving the segmentation accuracy.