西双版纳普洱茶叶片生化参数高光谱估测模型研究

Hyperspectral Estimation Model of Pu'er Tea Leaves Biochemical Parameters in Xishuangbanna

以西双版纳普洱茶为研究对象,利用ASD Field Spec 3地物光谱仪采集叶片高光谱数据,采用导数光谱分析技术对光谱数据进行处理,在实验室测定相应的茶氨酸和氮素含量,分析普洱茶叶片生化参数与原始光谱、光谱一阶微分、光谱对数一阶微分以及高光谱特征变量间的相关性,并利用遗传算法优化的BP神经网络建立了茶氨酸含量和氮素含量的高光谱估测模型。结果表明:普洱茶叶片生化参数含量与高光谱原始反射率间相关性弱,但与光谱一阶微分、光谱对数一阶微分和高光谱特征变量在可见光、近红外波段范围内相关性较强;遗传算法优化下的BP神经网络模型对普洱茶叶片生化参数的估测精度优于普通BP神经网络模型,茶氨酸含量估测精度RMSE为0.21 mg/g,R^2为0.73,氮素含量估测精度RMSE为0.36 g/kg,R^2等于0.88。

Taking Xishuangbanna Pu’er tea as the research object, ASD Field Spec 3 was used to collect the hyperspectral data of Pu’er tea leaves. The spectral data was processed by derivative spectral analysis technology and determine the corresponding nitrogen and theanine content in the laboratory. The correlation between biochemical parameters of Pu’er tea leaves and the original spectrum, 1 st derivative of spectrum, 1 st derivative of the spectrum logarithm and hyperspectral feature variables were analyzed, and a Back Propagation Neural Network model for content of nitrogen and theanine estimation was established. The results show that the correlation between the biochemical parameter content of Pu’er tea leaves and the original reflectance of hyperspectral is weak, but the correlation with the 1 st derivative of spectrum, the 1 st derivative of the spectrum logarithm and the hyperspectral feature variables are strong. The estimation accuracy of BPNN model optimized by genetic algorithm is better than that of the common BPNN model for estimating the biochemical parameters of Pu’er tea leaves. The BP neural network model optimized by genetic algorithm is superior to the common BP neural network model in estimating the biochemical parameters of Pu’er tea leaves. The accuracy of the determination of theanine content is 0.21 mg/g, R^2 is 0.73, the estimation accuracy of nitrogen content is RMSE of 0.36 g/kg, and R^2 is equal to 0.88.