The rapid detection of undisturbed soil moisture content based on BPNN

As soil moisture prediction model with the whole near-infrared spectral regions is complex and the single-band prediction model is susceptible to environmental impact, a three-wavelength method for measuring undisturbed soil moisture content (MC) rapidly was proposed based on BP artificial neural networks in this study. A total of 115 soil samples were collected and the NIR reflectance spectra of all soil samples were measured. The spectral data were transformed to new spectral data with logarithmic transformation and logarithmic of reciprocal transformation. The spectral wavelengths that is sensitive and insensitive to soil MC were selected by correlation coefficient method. The models for prediction of undisturbed soil MC were developed based on BP neural networks(BPNN) with the sensitive spectral wavelengths and insensitive spectral wavelength. The results show that the prediction precision of the models was high and the determination coefficients(R²) of the best prediction accuracy of the models reached 0.982 with the root mean square error of prediction( RMSEP) of 1.0402%. Thus, it is concluded that the methods used in this paper are available for rapid detection of undisturbed soil MC and also provide theoretical basis for developing a low-cost portable near-infrared moisture meter.