Gram-Schmidt算法与GRNN融合的加工番茄早疫病高光谱预测

加工番茄早疫病的准确预测,有助于及时采取防治措施,降低产量损失.测定加工番茄早疫病冠层光谱,对380～760 nm进行连续统去除变换,提取波段深度、波段位置、波段宽度、斜率、面积等特征参数,并对原始光谱提取红谷、绿峰、红边及相应波段位置等特征参数.利用Gram-Schmidt算法对特征参数进行成分提取,作为广义回归神经网络(GRNN)的输入变量,对加工番茄早疫病病情严重度进行预测.研究结果表明,与多元线性回归和偏最小二乘法预测模型比较,Gram-Schmidt算法与GRNN融合模型的预测精度相对较高,R2为0.843,RMSE为0.136,该方法能够对加工番茄早疫病病情严重度进行快速、准确的预测.

Highspectral prediction of early blight in processing tomato based on Gram-Schmidt algorithm and GRNN

Accurate prediction of early blight in processing tomato is good for taking active prevention measures and reducing loss of production. The canopy spectrum of early blight in processing tomato was measured, and continuum removal and transformation were conducted over 380-760 nm to get characteristic parameters of band depth, band position, band width, slope and area, and to extract characteristic parameters of red valley, green peaks, red edge and corresponding band position from original spectrum. Then the components were extracted from characteristic parameters by Gram-Schmidt algorithm, and the components were taken as input variable of General Regression Nerve Net (GRNN) to predict severity of early blight in processing tomato. The results show that compared with multiple linear regression prediction mode and prediction mode of partial least squares method, combined model of Gram-Schmidt algorithm and GRNN is more precise with R2 of 0.843 and RMSE of 0.136, which indicates that the model can rapidly and accurately predict the severity of early blight in processing tomato.