Abstract:

Phosphorus （P） content in honey pomelo leaves is an important indice for accurate diagnosis and quantitative evaluation of growth status. The hyperspectral estimation models of P contents in honey pomelo leaves were established to provide basis for rapid， non-destructive and accurate estimation of P content. Based on the hyperspectral data and the measured data of P contents in pomelo leaves， the characteristic bands and spectral characteristic variables of the original spectrum and the first-order differential spectrum were extracted， and the single variable estimation model， partial least squares regression model and BP neural network regression model were constructed， and the best estimation model of P content was determined. The reflectance spectra of pomelo leaves with different P contents were significantly different at 350~1 050 nm. The correlation between the original spectrum and the first-order differential spectrum and the leaves P contents in the visible light range were significant， and there were multiple extremums. The sensitive wavelengths of the original spectral curve were 549 and 718 nm， and the sensitive wavelengths of the first order differential curve were 528， 703 and 591 nm， respectively. Among the established regression models， the model with higher coefficient of determination was selected for precision test. The fitting R² （0.775 9） of BP neural network model was the largest， followed by the fitting R² （0.749 9） of partial least squares estimation model. The results showed that the BP neural network model was the best one to estimate the P contents of pomelo leaves， the determination coefficient （R²） of modeling and verification were 0.71 and 0.775 9， respectively； the second was the partial least squares estimation model， the R² of modeling and verification were 0.64 and 0.749 9， respectively. Above results provided a theoretical basis for the nutrition monitoring and diagnosis of large-area honey pomelo leaves and the rational fertilization.

Key words: hyperspectrum, honey pomelo, phosphorus, spectral index

摘要：

蜜柚叶片磷素（phosphorus，P）含量是准确诊断和定量评价生长状况的重要指标，为快速、无损、精确地估测磷素含量，需要建立蜜柚叶片磷素含量高光谱估算模型。基于蜜柚叶片高光谱数据和磷素含量实测数据，提取原始光谱及一阶微分光谱特征波段和光谱特征变量，构建单变量估算模型、偏最小二乘回归模型和BP神经网络回归模型，并确定蜜柚叶片磷素含量最佳估算模型。在350~1 050 nm波段，原始光谱和一阶微分光谱与叶片磷素含量在可见光范围内有多波段相关性显著，并出现多个极值。原始光谱敏感波长为549和718 nm，一阶微分的敏感波长为528、703和591 nm。在建立的回归模型中，选择决定系数较高的模型进行精度检验，其中BP神经网络模型的拟合R²（0.775 9）最大，偏最小二乘估算模型的拟合R²（0.749 9）次之。综合建模精度和模型检验精度，确定BP神经网络模型为蜜柚叶片磷含量的最佳估算模型，建模和验证的R²分别为0.71和0.775 9；其次为偏最小二乘估算模型，建模和验证的R²分别为0.64和0.749 9。上述结果可为大面积的蜜柚叶片营养遥感监测诊断和合理施肥提供理论依据。

关键词: 高光谱, 蜜柚, 磷素, 光谱指数