分别于返青期、 拔节期、 抽穗期和灌浆期采集不同磷素处理的冬小麦叶片原始高光谱数据; 之后求取其一阶导数(一阶导数光谱)并进行小波去噪处理; 通过分析原始光谱和一阶导数光谱对不同磷素处理水平的响应特征, 确定敏感波长范围并提取四种吸收面积; 将每个叶片磷素含量值对应的四种吸收面积的归一化值, 作为样本空间样本点的位置坐标(4维样本输入矢量), 对应叶片磷素含量的归一化值作为该样本点的目标输出, 二者同时提交给径向基函数神经网络。 结果表明： (1)冬小麦叶片原始光谱对叶片磷素含量变化反应敏感的波长范围为426～435 nm和669～680 nm。 (2)一阶导数光谱的敏感波长范围为481～493 nm和685～696 nm。 (3)训练后的径向基函数神经网络模型能够学习和掌握样本点与目标输出之间的线性/非线性映射关系, 并且具有一定的推广能力。

The hyperspectral leaf reflectance in winter wheat was measured under 4 phosphorus levels at different growth stages, i.e.revival stage, jointing stage, tassel stage and grouting stage.And their first derivative of spectra were calculated and denoised by the threshold denoising method based on wavelet transform.After studying characteristics of the two kinds of spectra resulting from different phosphorus contents levels as well as correlations between leaf phosphorus contents and spectral values, sensitive wavebands and four kinds of absorption areas were extracted.Then the four kinds of absorption areas and their corresponding leaf phosphorus content were normalized and input to RBFNN.Results show that: (1) Sensitive wavebands for monitoring leaf phosphorus contents in original leaf spectra are 426~435 and 669~680 nm.(2) Sensitive wavebands in first derivative of spectra are 481~493 and 685~696 nm.(3) Trained RBFNN can learn and seize the linearity/non-linearity mapping between samples and output targets.