多光谱作物生长传感器温度特性试验

Test on temperature characteristics of multi-spectral sensor for crop growth

基于光谱的作物无损监测技术可实时地获取作物生长信息，进而为作物的生长调控提供数据支持，为推广该技术在农业生产中的应用，南京农业大学国家信息农业工程技术中心研制了低成本的多光谱作物生长传感器。传感器监测作物720和810nm光谱反射率，根据反射率反演作物的叶层氮含量、叶层氮积累量、叶面积指数和叶干重等作物生长信息。为提高传感器田间应用的温度稳定性，该文研究了温度对传感器输出特性的影响，并利用符号回归技术构建了传感器反射率的温度补偿。试验于恒温恒湿试验箱中进行，试验温度分别设定为6、11、15、20、25、30、35、40，44、49、54、62℃，相对湿度保持为40%不变，以40%和60%反射率标准灰度板为传感器的监测对象。研究结果表明，在光强不变的情况下，传感器输出电压随温度的升高呈增加趋势，反射率则表现为下降趋势。构建的温度补偿模型使反射率因温度影响产生的波动由1%～2.6%下降到0.45%以下，成对t检验结果表明反射率的温度补偿模型可显著降低温度对传感器反射率的影响（P=0.015＆lt;0.05）。该文传感器温度补偿模型的构建方法具有一定的普适性，可为其他传感器的温度补偿研究提供参考。

In modern agriculture, the application of non-destructive spectroscopic techniques is very useful for estimating crop growth status. Non-destructive monitoring techniques based on spectral reflectance can provide the real-time information required for crop growth regulation. Thus, these techniques have significant application value in crop production. Despite being highly precise, the existing non-destructive spectroscopic techniques such as FieldSpec Pro FR250, GreenSeeker, and Crop Circle ACS-470 are expensive and complicated, hence their application is not suitable for agricultural production especially in China where the average per capita landholding is about 0.1 ha. In order to promote the use of non-destructive monitoring spectrum technology in agriculture, the National Engineering and Technology Center for Information Agriculture, Nanjing Agricultural University has developed a multi-spectral sensor for crop growth monitoring. The sensor monitor has a spectral reflectance of 720 nm and 810 nm to access growth indexes of leaf nitrogen content （LNC）, leaf nitrogen accumulation （LNA）, leaf area index （LAI）, and plant dry matter （PDM）. Under field conditions, the seasonal variations in temperature and sunlight can affect the internal temperature of the sensors from 10℃to 60℃. Temperature compensation is required to minimize the impacts of internal variations in temperature on the output signal of the sensor. Hardware compensation and software compensation are the two methods of temperature compensation. Hardware compensation methods mainly use electric circuits such as the thermal bridge compensation method and the double electric bridge compensation method to eliminate the influence of temperature. However, these methods are complex, expensive, and have low accuracy. Software compensation methods eliminate the influence of temperature on sensors by building a temperature compensation model, such as an interpolation method, least squares polynomial curve fitting method, least squares support vector machine method, or artificial neural network method, etc. The software compensation methods are simple, cheap, and have high accuracy as compared to hardware compensation methods. A temperature compensation model for reflectivity was constructed by studying the effect of temperature on the sensors in order to improve the temperature stability of the sensor for field applications. The experiments were carried out in a temperature and humidity control chamber. In experiment 1, the experiment temperature was set to 6℃, 11℃, 15℃, 20℃, 25℃, 30℃, 35℃, 40℃, 44℃, 49℃, 54℃, and 62℃. and 40%relative humidity was maintained. Under constant light intensity, a standard gray board with 40% reflectivity was used as the monitoring object of the sensor. Experiment 2 was carried out under the same experimental temperature conditions as that in experiment 1. However, the light intensity was changed, and 40%and 60%reflectance standard gray boards were used as the monitoring object of the sensor to examine the suitability of a newly constructed temperature compensation model for the reflectivity. The results indicated that under constant light intensity, the sensor output voltage increases while the reflectance decreases with an increases in temperature. A temperature-based sensor output voltage prediction model with the coefficient of determination R2 was 0.9998, and relative root mean square error RRMSE of 2.31%was constructed to predict sensor output voltage at different temperatures on the basis of output voltage at 25℃. The transformed prediction model in the present study can obtain 25℃ sensor output voltage according to the output voltages at different temperatures. The temperature compensation of reflectance was well implemented by using the 25℃ output voltage obtained by the transformed model to calculate reflectance. After temperature compensation, the reflectance fluctuation range decreased, and the max range was less than 0.45%. The t test results （P=0.015〈0.05） showed that a temperature compensating model could significantly reduce the effect of temperature on the reflectivity.