基于多信息融合的番茄冠层水分诊断  

Water moisture diagnosis of tomato canopy based on multi-information fusion

作　　者：高洪燕[1] 毛罕平[1] 张晓东[1] 孙俊[1] 

机构地区：[1]江苏大学现代农业装备与技术省部共建教育部重点实验室,镇江212013

出　　处：《农业工程学报》2012年第16期140-144,共5页Transactions of the Chinese Society of Agricultural Engineering

基　　金：国家自然科学基金项目(61075036,31101082);江苏省高校自然科学基础研究重大项目(10KJA210010);江苏省科技支撑计划项目(BE2008380);江苏省"333高层次人才培养工程"专项基金资助项目;江苏高校优势学科建设工程资助项目(苏财教﹝2011﹞8号)

摘　　要：为精确、快速、稳定地测定番茄植株含水率并能全面检测其水分胁迫程度,该文通过将番茄冠层的反射光谱、多光谱图像、冠层温度特征及环境温湿度进行多信息融合,评判其水分胁迫状况。通过去除冠层光谱噪声波长,并进行相关性分析,确定950～1080nm、1170～1300nm、1370～1500nm、1600～1730nm和1860～1980nm为5个相关性较高的波段,并利用逐步回归,进一步筛选出1026、1247、1431、1640、1910nm为特征波长。对获取的中心波长为810nm近红外图像和560nm可见光图像进行3×3窗口的中值滤波,采用最大类间方差法进行图像分割,计算图像的平均灰度。获取番茄冠层温度,并结合环境温湿度信息,建立番茄的冠气温差模型和水分胁迫指数模型。将5个特征波长、单通道的近红外和可见光图像的灰度均值和CWSI作为多信息融合的输入,利用偏最小二乘-神经网络回归分析并进行了验证,最终得到实测值与预测值的相关系数为0.9364,验证均方根误差为10.6713,平均误差为7.6714%,拟合方程斜率为0.9615。多信息融合模型的各项评价指标均好于单一信息模型。In order to accurately, rapidly and stably determine the moisture content of tomatoes and fully detect water stress degree, the multi-information fusion of the reflection spectrum, multi-spectral images, canopy temperature, environmental temperature and humidity were used to judge the water stress. After removing the noise wavelengths and correlation analysis, 950-1 080, 1 170-1 300, 1 370-1 500, 1 600-1 730 and 1 860-1 980 nm wave bands were selected. Bands of 1026, 1247, 1431, 1640 and 1910 nm were as the features using the stepwise regression. The 3x3 window median filter and the Otsu segmented were applied by IR and G images, finally the average gray was calculated. Combined with canopy temperature, environmental temperature and humidity, the CWSI was established. The five characteristic wavelengths, IR and G images average gray and CWSI as multi-information fusions parameters, the analysis and verification results obtained by PLS-ANN were correlation coefficient 0.9364, the root mean square error 10.6713, mean error 7.6714%, fitting equation slope 0.9615. It showed that the evaluations of the multi-information fusion model were better than those of the model represented by single sensor.

关 键 词：信息融合 温度 湿度调控 反射光谱 多光谱图像 冠层温度 番茄 

分 类 号：S123[农业科学—农业基础科学]