基于深度学习的水稻表型特征提取和穗质量预测研究
DOI:
作者:
作者单位:

作者简介:

通讯作者:

中图分类号:

基金项目:

国家自然科学基金面上项目(31770397); 国家自然科学基金青年项目(31701317)


Deep learning-based extraction of rice phenotypic characteristics and prediction of rice panicle weight
Author:
Affiliation:

Fund Project:

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 资源附件
  • |
  • 文章评论
    摘要:

    水稻产量与水稻穗数和穗质量密切相关,精确预测水稻产量可以加快育种速度。以盆栽水稻为研究对象,首先利用可见光图像结合图像处理技术进行特征提取,获取整株水稻的51个表型特征。结合深度学习,运用Faster RCNN卷积神经网络训练模型对水稻穗数进行检测,同时使用SegNet网络框架训练得到的模型对水稻稻穗进行分割,得到水稻穗部的二值图像,结合图像处理技术提取穗部的33个表型特征数据。提取了颜色、形态、纹理共85个表型参数,对所有85个数据进行归一化处理,将归一化的85个表型数据与稻穗鲜质量、干质量进行逐步线性回归,挑选相关性高的特征数据。分别使用穗数和33个特征穗部、51个特征整株、所有85个特征中相关性高的特征数据构建盆栽水稻稻穗鲜质量、干质量的预测模型,最后根据模型的决定系数R2、平均相对误差(MAPE)和相对误差绝对值的标准差(SAPE)挑选最优预测模型。预测结果表明穗部特征预测效果最好,其中效果最好的模型鲜质量、干质量预测值与真实值的决定系数R2分别达到0.787±0.051和0.840±0.054。

    Abstract:

    The yield of rice is closely related to the panicle number and the panicle weight of rice. The accurate prediction of rice yield can accelerate the speed of breeding. In order to study the relationship between rice yield and rice phenotypic characteristics,the visible light images combined with image processing technology was used for extracting feature of potted rice. 51 phenotypic traits of whole rice were obtained. Combined with deep learning technology,the Faster R convolutional neural network training model was used to detect the number of rice spikes. At the same time,the SegNet model was trained using the SegNet network framework to segment the rice spikes to obtain the binary image of the rice spikes. 33 phenotypic feature data of the panicle were extracted with image processing technology. A total of 85 phenotypic parameters of color,shape,and texture were extracted,and all 85 data were normalized. The 85 phenotypic data normalized were gradually linearly regressed with the fresh and dry quality of rice panicle,and the correlation was selected. The artificial measurement data in the experiment included the fresh weight and dry weight of potted rice panicle. The models of predicting fresh and dry panicle weight of potted rice were established separately by using panicle number and characteristic panicles,51 characteristics of whole plants and all 85 characteristics of high correlation characteristic data. The prediction model was optimized according to the determination coefficientR2,mean relative error (MAPE) and standard deviation of absolute relative error (SAPE). The optimal prediction model was selected according to the decision coefficient R2,average relative error (MAPE) and standard deviation of relative absolute value (SAPE) of the model. The results of prediction showed that the effect of predicting panicle characteristics is the best. The decision coefficients R2 of the predicted value and the real value of the model with the best effect are 0.787±0.051 and 0.840±0.054,respectively. Combined with deep learning,the number and characteristics of panicle difficult to obtain automatically by traditional methods are extracted. It will provide a new idea and method for predicting rice panicle weight,and further improving the accuracy of predicting rice panicle weight.

    参考文献
    相似文献
    引证文献
引用本文

杨万里,段凌凤,杨万能.基于深度学习的水稻表型特征提取和穗质量预测研究[J].华中农业大学学报,2021,(1):

复制
分享
文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:2020-06-15
  • 最后修改日期:
  • 录用日期:
  • 在线发布日期: 2021-02-10
  • 出版日期: