Prediction of Total Nitrogen in Soil Based on Random Frog Leaping Wavelet Neural Network

Abstract

Soil total nitrogen is an important information for diagnosing soil fertility levels and guiding accurate fertilization of crops, it is important to establish a near-infrared spectral estimation model of soil total nitrogen and optimize the selection of modeling bands for the rapid acquisition of soil nutrient information and accurate agricultural development. In this paper, near-infrared spectra of 85 field soil samples were measured using a Fourier-NIR spectrometer. First, S-G smoothing filter was applied to the original spectral curve, and then the sensitivity wavelength of soil total nitrogen content was selected by the normal analysis of correlation coefficient and the random frog leaping algorithm. Multiple linear regression models and wavelet neural network models were established using the selected sensitive wavelength and soil total nitrogen content. The modeling results showed that the determination coefficient Rc2 of the soil total content prediction model established based on random frog leaping-wavelet neural network was 0.9428, the prediction verification coefficient Rv2 was 0.9236, and the root mean square error correction RMSEC was 0.0084. The root mean square RMSEP of the prediction error was 0.0099. The accuracy of modeling and forecasting is significantly improving compared with the traditional method, and the wavelet neural network can effectively solve the nonlinear problem of soil absorbance and can be better used in actual production.