Spectral feature extraction and modeling of soil total nitrogen content based on NIR technology and wavelet packet analysis

It is a non-destructive and real-time method to detect the soil nutrient content by using spectroscopy analysis technology. In order to isolate the effective spectral for TN content from the soil spectra effectively, the NIR model predicting TN was developed based on wavelet packet analysis. 100 soil samples were collected for calibration and validation from the field. First, using the high-precision NIR detecting instrument to scan the target and obtaining the continuous spectra of soil samples in the laboratory. Secondly, with three different orthogonal wavelets (bior4.4, db4, sym4) as the generating functions, the original signal of each soil sample was decomposed and reconstructed based on the respective wavelet packet. Then the multiple linear regression (MLR) models for TN were established based on each drawn characteristic spectrum. Finally, three models were compared and analyzed, and the model with the highest forecasting accuracy was obtained based on db4, which determined R 2 reached 0.904. The research concluded that wavelet packet analysis could eliminate or substantially reduce the factors outside the parameters to the spectrum directly or indirectly, and the obstacles in establishing linear models for soil parameters were removed. It is feasible and potential to the real-time prediction of TN content.