摘要
针对油菜水分胁迫的的无损探测,研究了利用多光谱图像对油菜含水率进行定量分析的方法。采用中值滤波法对图像进行预处理;利用二维最大信息熵阈值分割法对多光谱图像进行背景分割;对油菜冠层多光谱图像的均值和比值特征进行了提取,发现560nm处的可见光图像,960、810nm处的近红外图像均值特征和960nm/810nm图像的比值特征在整个发育期与油菜含水率的相关性均较高。考虑到多光谱变量间存在的多重共线性影响,利用逐步回归法建立了不同发育期油菜含水率的多光谱图像特征预测模型。结果表明,该方法能够实现对油菜水分胁迫的定量分析,其中,油菜苗期含水率预测模型的预测值与实测值的相关系数为0.83,均方根误差为4.52%,平均相对误差小于8%,可为科学精确地指导灌溉提供依据。
Multi-spectral image analysis method was utilized to quantitatively analyze the rape moisture content for the nondestructive testing of rape water stress.Median-filtering method was used to preprocess the images.Two dimensional maximum entropy segment approach was used to complete background segmentation of multi-spectral images.The mean ratio features of multi-spectral images of rape canopy were extracted.It was found that the features of the image mean value at 560,960,810 nm and the 960 nm/810 nm ratio were highly correlated with the rape moisture content during the rape＇s whole growth period.With the consideration of the existence of the multi-collinearity among the multi-spectral variables,the prediction model of moisture content of rape in different growth phases was built by stepwise regression method.The result showed that the multi-spectral image prediction method can be used to quantitatively analyze the rape moisture content.The correlation coefficient between the predicted value and the measured one was 0.83,and the RMSE was 4.52%.The average relative error was less than 8% in the seeding stage.The prediction model in this study may provide scientific evidence for water-efficient irrigation.
出处
《农业工程学报》
EI
CAS
CSCD
北大核心
2011年第3期152-157,共6页
Transactions of the Chinese Society of Agricultural Engineering
基金
国家自然科学基金项目(61075036)
中国博士后科学基金(20100481097)
江苏省农业装备与智能化高技术研究重点实验室项目(BM2009703)
国家"863"高技术研究发展计划项目(2008AA10Z204)
江苏大学高级专业人才基金项目(10JDG081)
江苏大学博士后基金项目
关键词
图像处理
水分
回归分析
多光谱图像
image processing
moisture
regression analysis
multi-spectral image
rape
