基于SVM的加工番茄细菌性斑点病氮素含量反演

氮素是作物需要最大量的营养元素,严重影响作物的生长发育和产量品质。高光谱遥感对氮素含量进行反演,具有快速、低耗及非损伤性等优势。提出了一种多光谱指数和SVM模型相结合的方法,选取不同生育期和不同病害严重度的加工番茄细菌性斑点病的病叶,同时测定病叶的氮素含量。通过相关分析、线性回归的绝对系数R2和F值,优选了光谱指数PSSRb、ND705、GMI-2和PTBSc,作为SVM模型的输入变量,反演氮素含量和不确定性分析。结果表明:SVM模型反演氮素含量的真实值与预测值的相关系数为0.849,均方误差MSE为0.012,平均相对误差为0.007。与单光谱指数PSSRb和GMI-2构建的指数模型比较,SVM模型的预测能力更强,真实值与预测值拟合方程的绝对系数R2最大(R2=0.720)。说明多光谱指数的SVM模型,提高了加工番茄细菌性斑点病氮素含量的反演精度,同时为多波段协同反演氮素含量提供了新的思路。

Inversion of Nitrogen Content of Bacterial Speck of Processing Tomato based on Support Vector Machine

Nitrogen is the largest amount of nutrition elements needed by crop.At the same time,it seriously affect growth and quality of crop.There are rapid,low consumption and non|invasive advantages of Hyperspectral remote sensing on nitrogen content inversion.This paper proposed the method of using multi|spectral index and SVM model,which could improve the nitrogen content inversion accuracy.Using Bacterial Speck of processing tomato actual spectral measurement and nitrogen content in the different growth stages and different severity level.Through the correlation analysis and the absolute coefficient R2 and F value of linear regression,we choose spectral index:PSSRb,ND705,GMI|2,PTBSc.They are acted the input variable of SVM model,and invented the nitrogen content,and analyzed uncertainty.The result show:the correlation coefficient between real value and predict value is 0.711 and the value of MSE is 0.021;and the value of average relative error is 0.007.Compared with exponential model of the single spectral index:PSSRb and GMI|2,the predict ability of SVM model is tetter.Then the correlation coefficient between real value and predict value is 0.720,which is maximum.So the multi|spectral index and SVM model prediction has a better fitting effect than the single spectral index.