用户登录
期刊信息
  • 主管单位:
  • 中国科学技术协会
  • 主办单位:
  • 中国仪器仪表学会、上海光学仪器研究所、中国光学学会工程光学专业委员会
  • 主  编:
  • 庄松林
  • 地  址:
  • 上海市军工路516号上海理工大学《光学仪器》编辑部
  • 邮政编码:
  • 200093
  • 联系电话:
  • 021-55270110
  • 电子邮件:
  • gxyq@usst.edu.cn
  • 国际标准刊号:
  • 1005-5630
  • 国内统一刊号:
  • 31-1504/TH
  • 邮发代号:
  • 单  价:
  • 15.00
  • 定  价:
  • 90.00
基于近红外光谱技术的马铃薯叶片含水率高效预测
Efficient determination of water content in potato leaves based on spectroscopy technology
投稿时间:2020-03-30  
DOI:10.3969/j.issn.1005-5630.2020.04.002
中文关键词:  马铃薯叶片  含水率  光谱  偏最小二乘回归(PLSR)  BP神经网络
英文关键词:potato leaf  moisture content  spectrum  partial least squares regression(PLSR)  BP neural network
基金项目:国家大学生创新创业训练计划(201910681027); 校级研究生核心课程建设项目(YH2018-C04)
作者单位E-mail
于旭峰 云南师范大学 物理与电子信息学院云南 昆明 650000  
李红梅 云南师范大学 物理与电子信息学院云南 昆明 650000  
卓伟 云南师范大学 物理与电子信息学院云南 昆明 650000  
冯洁 云南师范大学 物理与电子信息学院云南 昆明 650000 fengjie_yunnan@126.com 
摘要点击次数: 135
全文下载次数: 36
中文摘要:
      提出了运用近红外光谱技术检测新鲜马铃薯叶片中含水量的方法,并通过预测结果和运算量的对比得出一种高效率的预测方法。采集了900~2100 nm波段范围内110个新鲜马铃薯叶片的光谱反射率信息,经SG(Savitzky-Golay)平滑、多元散射校正(MSC)和标准正态变量变换(SNV)3种预处理后,分别建立偏最小二乘回归(PLSR)模型和BP神经网络模型,再运用回归系数(regression coefficients, RC)法在全波段光谱中提取特征波长,同样经3种预处理后分别建立预测模型。结果表明:在运用光谱全波段信息构建的模型中,经多元散射校正(MSC)预处理建立的BP神经网络模型预测效果最好,预测集决定系数R20.9791,均方根误差RMSE为0.3723;在基于特征波长构建的模型中,经SG平滑预处理建立的神经网络模型预测效果最优,预测集决定系数R20.9658,均方根误差RMSE为0.4759;验证了特征波段结合BP神经网络建立的模型与全波段建立的模型预测结果相差不大,因而能够极大地减少运算量,提高预测效率。
英文摘要:
      The determination of moisture content in potato leaves using spectral technique was studied in this paper. Spectral signatures of one hundred and ten fresh potato leaves in the wavelengths of 900-2100 nm were acquired by the spectral device. Then, the moisture content was measured by the drying method.The near-infrared reflection spectrum information was corrected by the Savitzky-Golay (SG) smoothing, multiplicative scatter correction (MSC) and standard normal variable (SNV) correction. The quantitative relationship between spectral information and moisture was built by partial least squares regression (PLSR) and BP neural network respectively. The effective wavelength was identified by regression coefficients (RC) and corrected by three pretreatment methods. Then the PLSR and BP neural network models were built respectively. The results showed that for full wavelengths-based models, MSC-BP model performed the best with the coefficient of determination (R2) of 0.9791 and the root mean square error (RMSE) of 0.3723 in the prediction. For selected wavelengths-based models, it was the SG-BP model that obtained the optimal result. The R2 value was 0.965 8 and the RMSE value was 0.475 9 in the prediction. This experiment verified that the prediction results of the model established by combining the characteristic band with BP neural network were not different from those of the model established by the whole band, so it could greatly reduce the computation and improve the efficiency.
HTML   查看全文  查看/发表评论  下载PDF阅读器
关闭