欢迎访问《棉花学报》! 今天是

棉花学报 ›› 2015, Vol. 27 ›› Issue (3): 275-282.doi: 10.11963/issn.1002-7807.201503012

• 研究与进展 • 上一篇    下一篇

应用灰板校正提高计算机视觉预测棉花植株含水量的精确度

王娟1,危常州1*,万丹1,王肖娟2,李玮1,顾凯1   

  1. 1. 石河子大学农学院农业资源与环境系,新疆 石河子832000;2. 新疆天业(集团) 有限公司,新疆 石河子832000
  • 收稿日期:2014-07-24 出版日期:2015-05-15 发布日期:2015-05-15
  • 通讯作者: changzhouwei@126.com
  • 作者简介:王娟(1981―),女,博士, symwj @163.com
  • 基金资助:
    国家自然科学基金(31060276) 、农业部行业专项(201103003)和教育部高等学校博士学科点专项科研基金(20106518110001)

Gray Broad Calibration May Increase Precision in the Prediction of Cotton Water Content Based on Computer Version

Wang Juan1, Wei Changzhou1*, Wan Dan1, Wang Xiaojuan2, Li Wei1, Gu Kai1   

  1. 1. Department of Resource and Environment, Agronomy College, Shihezi University, Shihezi, Xinjiang 832000, China; 2. Xinjiang Tianye (Group) Co., Ltd., Shihezi, Xinjiang 832000, China
  • Received:2014-07-24 Online:2015-05-15 Published:2015-05-15
  • Contact: changzhouwei@126.com

摘要: 利用灰板校正以消除棉花不同生育期图片颜色特征值的亮度差异,建立适用于不同生育期预测植株含水量的通用模型,以提高运用计算机视觉技术进行棉花植株含水量预测的精度。研究结果表明,由灰板校正前、后颜色特征值G-B建立的最佳预测模型,决定系数分别为0.746和0.782。有效性检验结果表明,灰板校正前、后计算预测值与实测值的决定系数分别为0.739和0.783;RMSE分别为2.218和2.03,RE分别为2.13%和1.79%。基于计算机视觉提取的冠层图片颜色特征值能够预测植株含水量,应用灰板校正颜色特征值能够提高模型预测精度,可为提高计算机视觉预测植株水分状况的精度提供技术支撑和方法补充。

关键词: 计算机视觉; 灰板校正; 棉花; 植株含水量; 颜色特征值; 模型

Abstract: This paper aimed at establishing a rapid, non-destructive and low-cost model to predict cotton water content for whole growth stages based on computer-vision technology. The data of color characteristic parameters extracted from cotton canopy digital image of different growth periods were adjusted by gray board in order to eliminate differences in brightness and to improve the prediction precision of the model. The best prediction model is established based on color characteristic parameters G-B by original data or gray board corrected data, the coefficient of determination for the two models was 0.746 and 0.782, respectively. Validation test indicated that the prediction accuracy of models based on gray broad calibrated data was improved comparing to model established by original data. The coefficient of determination between measured plant water content and predicted value was 0.739 and 0.783, respectively; the root mean square error (RMSE) was 2.218 and 2.030; the relative error (RE) between predicted values and measured values were 2.13% and 1.79%. Our results showed that the application of computer vision may predict cotton water content and the color value adjusted by gray board may improve the model prediction accuracy. This research provided a simple, higher precision technical support and new method for diagnosis of plant water status based on computer vision.

Key words: computer vision; gray broad corrected; cotton; plant water content; color characteristic parameters; model

中图分类号: 
  • S126:TP391