基于高光谱数据研究应用近红外相机加装滤光片实现玉米叶片水分测量的关键参数

Key Parameters for Maize Leaf Moisture Measurement Using NIR Camera With Filters Based on Hyperspectral Data

实现玉米等活体作物叶片不同区域以及近地冠层的水分含量检测,可为研究作物生长中的水分胁迫响应机制,以及旱情监测、精准灌溉等提供有效技术手段。大体积、大重量、低光通量的高光谱仪难以在田间实现活体检测,将小体积、小重量、高光通量近红外相机加装滤光片使其具有波长分辨能力,有望实现田间活体叶片水分含量成像检测。基于玉米叶片的近红外高光谱数据,对实现水分测量的特征波长位置及个数,以及特征波长带宽和偏移限度等关键参数进行研究;其中,不同带宽的仿真数据依据滤光片透光分布函数得到;固定带宽条件下的中心波长偏移仿真数据依据插值方法得到。研究结果表明:特征波长分别为1150和1400 nm,带宽小于100 nm时可满足要求,能够找到符合参数条件的滤光片产品;在带宽为25 nm下所建水分含量模型,其建模集的决定系数(R 2)和均方根误差(RMSE)分别达到0.968,1.245%,预测集分别达到0.960,1.298%,可实现玉米叶片水分的准确检测。为推测加装滤光片所建模型受环境温度的影响,应用固定中心波长下的模型预测不同偏移量的仿真数据,当偏移量为0.05 nm时,模型预测误差为不漂移条件下3%左右,可忽略,由干涉滤光片中心波长漂移与温度的关系,相当于环境温度在50℃的范围对检测结果影响不大。该研究对近红外相机加装滤光片形成多光谱成像检测装置提供了重要技术参数支撑和环境工作范围建议,装置的搭建工作已开始,装置的实现可为现代农业作物生理、生产研究提供新型、有效手段。

To provide effective technical means for the study of the crop growth in water stress response mechanism,drought monitoring,and precision irrigation,moisture content detection in different regions of maize live crop leaves and near the canopy is achieved.The large volume,large weight,low luminous flux hyperspectrometer is difficult to achieve live detection in the field.The small volume,small weight,high luminous flux NIR camera with filters to make it wavelength-resolved is expected to achieve live leaf moisture content imaging detection in the field.Based on the near-infrared hyperspectral data of live maize leaves,this study is to investigate the key parameters,including the characteristic wavelength position and number,bandwidth and offset limit.Among them,the simulation data of different bandwidths are based on the filter light transmission distribution function;the simulation data of the center wavelength shift under fixed bandwidths are based on the interpolation method.The research results showed that the feature wavelengths are 1150 and 1400 nm respectively,and the bandwidth is less than 100 nm,which can meet the requirements and find the filter products that meet the parameter conditions.When the bandwidth was 25 nm,the model set’s determination coefficient(R 2)and root mean square error(RMSE)were 0.968 and 1.245%,respectively,and the prediction set was 0.960 and 1.298%,respectively.To infer that the model built with the filters is affected by the ambient temperature,the model within a fixed center wavelength is used to predict the simulation data of different offsets.When the drift was 0.05 nm,the model prediction errorunder non-drift conditions was about 3%,which could be neglected.The relationship between the center wavelength drift of the filter and the temperature is equivalent to that the ambient temperature in the range of 50℃has little effect on the detection results.Thisresearch provides important technical parameter support and working range for the NIR camera with filters to form a multispectral imaging system detection device.The device’s construction has been started,and the realization of the device can provide new and effective means for modern agricultural crop physiology and production research.