Predicting apple tree leaf nitrogen content based on hyperspectral applying wavelet and wavelet packet analysis

The visible and NIR spectral reflectance were measured for apple leaves by using a spectrophotometer in fruit-bearing, fruit-falling and fruit-maturing period respectively, and the nitrogen content of each sample was measured in the lab. The analysis of correlation between nitrogen content of apple tree leaves and their hyperspectral data was conducted. Then the low frequency signal and high frequency noise reduction signal were extracted by using wavelet packet decomposition algorithm. At the same time, the original spectral reflectance was denoised taking advantage of the wavelet filtering technology. And then the principal components spectra were collected after PCA (Principal Component Analysis). It was known that the model built based on noise reduction principal components spectra reached higher accuracy than the other three ones in fruit-bearing period and physiological fruit-maturing period. Their calibration R-2 reached 0.9529 and 0.9501, and validation R-2 reached 0.7285 and 0.7303 respectively. While in the fruit-falling period the model based on low frequency principal components spectra reached the highest accuracy, and its calibration R-2 reached 0.9921 and validation R-2 reached 0.6234. The results showed that it was an effective way to improve ability of predicting apple tree nitrogen content based on hyperspectral analysis by using wavelet packet algorithm.