For the rapid detection of leaf nitrogen content of summer corn, visible and near infrared (Vis/NIR) spectra of summer corn leaves, with different nitrogen levels at spinning stage, were measured by an ASD FieldSpec. Discrete approximation wavelet coefficient vectors of the second-scale were obtained via logarithmic transformation and multi-scale wavelet decomposition of the spectra data within “near infrared spectrum platform” (760 ～ 1000nm) and “green peak” (450 ～ 680nm ). Then principal components (PCs) were selected from these vectors by principal component analysis (PCA), and used as inputs of a generalized regression neural network (GRNN). The model was employed for the prediction of leaf nitrogen content of summer corn. Results show that logarithmic transformation can highlight the differences in the spectral response of summer corn leaves with different level of nitrogen within “near infrared spectrum platform” and “green peak” at spinning stage. The wavelet-based PCs can manifest the changes in the spectra of summer corn leaves with different nitrogen levels. Trained GRNN model with wavelet-based PCs as inputs can predict leaf nitrogen content of summer corn. The model is reliable and practicable.%使用高光谱仪ASD Field Spec于吐丝期采集不同氮素处理的夏玉米叶片光谱,并进行对数变换处理;通过对"绿峰"(450～680nm)和"近红外反射平台"(760～1000nm)谱段光谱数据进行多尺度小波分解,获取第二层离散近似小波系数向量;采用主成分分析,从第二层离散近似小波系数向量中提取特征作为输入参数,建立对叶片氮素含量的广义回归神经网络估算模型.结果表明:对数变换显著地增强了"绿峰"和"近红外反射平台"谱段夏玉米叶片光谱对不同氮素处理的响应差异;从第二层离散近似小波系数向量中提取的小波主成分能够反映夏玉米叶片光谱在不同氮素处理下的整体变化趋势;以小波主成分作为输入参数的广义回归神经网络能够较为准确地预测夏玉米叶片氮素含量,并且具有一定的推广能力.
展开▼
