通过小波变换去除了可见光区(350～560 nm)的噪声, 提取出了叶酸的特征波段366 nm和与叶绿素有关的特征波段380, 414, 437, 554 nm。 在560～2 500 nm的波长范围内, 去除噪声后的最大误差低于1.47%； 在特征峰谷处的最大误差不超过0.11%。 用BP神经网络建立了番茄施氮量预测模型。 研究表明, 在用植物探头获取番茄叶片光谱数据并去噪的条件下, 用554, 673, 1 440, 1 940 nm处的吸光度值作为BP神经网络的输入变量建立的番茄施氮量的预测模型有很高的预测精度, 有极大的潜力能够满足实际应用的需要。 对研究大田有效养分的预测模型也有重要的参考价值。

It was successful to denoise the spectrum signal within visual wave band (350-560 nm) by wavelet transformation, to extract the folic acid characteristic wavelength 366 nm and some character wavelengths with relation to chlorophyll at 380, 414, 437 and 554 nm. In the range from 560 to 2 500 nm, after denoising, the biggest error was smaller than 1.47%, while at the peak or vale of character wavelength the biggest error was smaller than 0.11%. Moreover, the model was established based on the denoised data acquired with aid of plant probe. The model was also based on BP neural network and for predicting the nitrogen content in nutrient solution in hydroponic cultivation of tomato. The results showed that the predicting model, which used the values of absorbance at 554, 673, 1 440 and 1 940 nm as input variables of BP neural network, had a very good forecasting accuracy and great potential to be used practically.