Scholarlink

<p>The quantification of crop phenotyping traits is essential to understand growth patterns and increase production. Traditional methods of monitoring such properties, particularly in white radishes, are notably labor-intensive and inefficient, necessitating the development of more effective techniques. Moreover, there is a lack of radish dataset to enable monitoring of radish plant growth that combines radish roots and leaves. Addressing these challenges, the current study proposes a radish dataset combining both radish roots and leaves. In addition, we propose an automated approach through the implementation of deep learning and mathematical model that leverages high-resolution imagery for the measurement of white radish phenotype traits. The study utilized a modified Mask Region-based Convolutional Neural Networks (R-CNN) algorithm to accurately segment radish components, facilitating the measurement of leaf and root dimensions. The traditional backbone was improved by introducing a local–global attention mechanism in the feature extraction block. The feature pyramid network (FPN) is also improved by integrating a self-attention mechanism in the top layer. Moreover, we utilize Geometrical Morphological Analysis and the medial axis transform method to measure the height and width of the white radish phenotype traits. Extensive experiments revealed that our proposed modified Mask R-CNN model acquired a mean average precision of 96.3% for segmentation and a mean absolute error (MAE) of 0.51mm for phenotype traits measurement. Our proposed framework demonstrates a significant advancement in the measurement process in agricultural studies, offering a reliable alternative to traditional, time-consuming methods. </p>

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

Deep learning based radish and leaf segmentation for phenotype trait measurement

  • 期刊论文

  • 发表日期：

    2025年1月3日

查看全文
  • Nur Alam,
  • A. S. M. Sharifuzzaman Sagar,
  • L. Minh Dang,
  • Wenqi Zhang,
  • Han Yong Park,
  • Moon Hyeonjoon

摘要

The quantification of crop phenotyping traits is essential to understand growth patterns and increase production. Traditional methods of monitoring such properties, particularly in white radishes, are notably labor-intensive and inefficient, necessitating the development of more effective techniques. Moreover, there is a lack of radish dataset to enable monitoring of radish plant growth that combines radish roots and leaves. Addressing these challenges, the current study proposes a radish dataset combining both radish roots and leaves. In addition, we propose an automated approach through the implementation of deep learning and mathematical model that leverages high-resolution imagery for the measurement of white radish phenotype traits. The study utilized a modified Mask Region-based Convolutional Neural Networks (R-CNN) algorithm to accurately segment radish components, facilitating the measurement of leaf and root dimensions. The traditional backbone was improved by introducing a local–global attention mechanism in the feature extraction block. The feature pyramid network (FPN) is also improved by integrating a self-attention mechanism in the top layer. Moreover, we utilize Geometrical Morphological Analysis and the medial axis transform method to measure the height and width of the white radish phenotype traits. Extensive experiments revealed that our proposed modified Mask R-CNN model acquired a mean average precision of 96.3% for segmentation and a mean absolute error (MAE) of 0.51mm for phenotype traits measurement. Our proposed framework demonstrates a significant advancement in the measurement process in agricultural studies, offering a reliable alternative to traditional, time-consuming methods.