多信息融合的冬小麦地上鲜生物量检测研究  

Monitoring of Winter Wheat Aboveground Fresh Biomass Based on Multi-Information Fusion Technology

作　　者：郑玲[1,2] 朱大洲 董大明[1] 张保华[1] 王成[1] 赵春江[1] 

机构地区：[1]北京农业信息技术研究中心,北京市农林科学院,北京100097 [2]安徽大学电子信息工程学院,安徽合肥230039 [3]农业部食物与营养发展研究所,北京100081

出　　处：《光谱学与光谱分析》2016年第6期1818-1825,共8页Spectroscopy and Spectral Analysis

基　　金：国家自然科学基金项目(31201125);农业部公益性行业科研专项项目(201203026);北京市自然科学基金项目(4142019)资助

摘　　要：将小麦冠层光谱与小麦冠层图像或者侧面图像进行多信息融合,对冬小麦地上鲜生物量进行预测,提高了冬小麦地上鲜生物量无损检测精度,试验获取苗期93个样本的冠层光谱和冠层图像,中后期(拔节期、抽穗期、开花期、灌浆期)200个样本的冠层光谱和侧面图像。将光谱反射率作为光谱特征参数,并通过图像处理提取小麦覆盖度作为图像特征参数,建立苗期和中后期基于光谱特征参数和图像特征参数的地上生物量检测模型,将冠层光谱反射率和小麦覆盖度作为多信息融合的输入,利用多元回归分析(MRA)和偏最小二乘法(PLS)建立地上鲜生物量预测模型并进行验证。结果表明,在苗期和中后期,将光谱信息和图像信息融合,采用PLS所建立的预测模型与单独的图像模型和光谱模型相比精度最高。苗期基于信息融合所建立的PLS模型验证集R^2为0.881,其RMSE为0.015kg;中后期基于信息融合所建立的PLS模型验证集R^2为0.791,RMSE为0.059kg。由此可见,相比单一的光谱模型和图像模型,图像信息和光谱信息融合之后,充分提高了光谱信息和图像信息的利用率,使模型的精度得以提高。The aim was to find a nondestructive way to improve the accuracy of detecting the winter wheat aboveground fresh biomass（AGFB）. In this study, data fusion technology of the spectroscopy technology and the machine vision technology were used to analyze the AGFB and solve the problem that the accuracy of the prediction model of a single technology is not high. In this experiment, canopy spectra and canopy pictures of 93 samples at seeding stage were collected. Canopy spectra and side images of 200 samples at medium and later growth stage were collected. Spectral reflectance as the spectral absorption parameter was used to construct the AGFB prediction models based on the spectra technology at different stages; The wheat coverage were extracted from canopy pictures and side images by using image processing technology to build the AGFB prediction models. Mul- tivariate regression analysis （MRA） and Partial least-squares regression analysis（PLS） were implemented on the feature varia- bles from the spectral information and image information. The results showed that, compared with the individual image model and spectral model, the AGFB prediction models of PLS based on multi-information at different stages shows better performance. At the seeding stage, the determination coefficient （R^2 ） of PLS models based on multi-information was 0. 881, and the RMSE was 0. 015 kg. The R^2 of PLS models based on multi-information was 0. 791, the RMSE was 0. 059 kg at middle and final stages. It demonstrated that the precision of model based on multi-information fusion technology, which increased utilization of image and spectral information, was improved for AGFB detecting, which is than the individual image model and spectral model.

关 键 词：多信息融合 偏最小二乘回归 冠层光谱 机器视觉 冬小麦 生物量 

分 类 号：S132[农业科学—农业基础科学]