Study on a fast EC measurement method of soda saline-alkali soil based on wavelet decomposition texture feature.

• Uniform standard developed for the saline soil samples with surface cracks. • Wavelet texture features extracted from cracked patterns of soil samples. • Quantitative analysis carried out between EC and wavelet texture features. • A new quick and non-destructive method developed for EC measurement. Wavelet texture features can well describe the cracking of salt-affected clayey soils on different decomposition scales since the desiccation cracking is considered as a common phenomenon and mainly determined by the salt content. To establish the relationship between wavelet texture features and the salinity of soda saline-alkali soils, 200 soil samples were selected in Songnen Plain of China and the crack images of the samples were processed uniformly. The 4-levels orthogonal wavelet decompositions were performed based on coiflet-1 wavelet basis function using the grayscale images. After that, correlation analysis was carried out between electrical conductivity (EC) and wavelet texture features (energy and L1 norm) under different decomposition levels. The results indicate that the poor relationship between low-frequency texture features and EC is hardly affected by the decomposition levels, but the correlation coefficients between high-frequency wavelet texture features and EC of soil samples increases significantly with decomposition levels. Besides, 100 calibration samples were used to establish the regression models and the results show that both energy and L1 norm are exponentially correlated with EC of soil samples especially for those from 90° high-frequency decomposition results with R2 of the exponential models of 0.90 and 0.84. A fast EC measurement method for predicting soil EC was then proposed and verified by the other 100 soil samples. The fitting results show very high prediction accuracy when EC was calculated by the energy from 90° with R2 of 0.91 and ratio of performance to deviation (RPD) of 4.73 and L1 norm from 90° with R2 of 0.85 and RPD of 3.53. Moreover, the fitting results based on the mean texture features calculated from 0°, 90° and 135° high-frequency decomposition results are also relatively good (R2 of 0.88 and RPD of 4.13 for energy, R2 of 0.81 and RPD of 3.36 for L1 norm). [ABSTRACT FROM AUTHOR]