A new method for determining loess particle size distribution based on 'core plus clothes' structure

The particle size distribution (PSD) of loess particles and agglomerates is the key factor determining the microstructure of loess. The PSD largely determines many physical and mechanical properties such as mechanical strength, deformation characteristics and permeability of loess. In this paper, we combined image processing, geometric measurements and statistical methods to quantitatively characterize the scanning electron microscopy (SEM) images of loess. A new method based on the 'core + clothes' structure is proposed to determine the PSD. The developed Gaussian mixture model is applicable to describe the overall distribution of the image pixel density. Determining pixel density distribution and particle boundary allows for the measurement of various geometric parameters. The comparison analyses suggest that the sieve-hydrometer method has a well-developed test procedure; however, the measured particle size is limited to 0.075 mm. The laser diffraction method is simple and fast, yet only weak agglomerates can be dispersed, and very fine particles are influenced by size data conversion theory and light source parameters. X-ray microtomography has the advantage of a non-destructive sample preparation and intuitive process. However, the results are greatly affected by the parameter settings and segmentation algorithm, and the segmentation effect is poor for particles with a size smaller than 12 mu m. The new image processing technique developed in this study improves the limitation of the measurement results via changes in the loess state (water content, stress level). Moreover, different PSDs can be determined according to the variation of functional units. This work is of great significance for the study of numerous properties related to the loess state.

Automated summary

Particle size distribution is a key factor in determining the microstructure and properties of loess. This study proposes a new method combining image processing and statistical techniques to measure geometric parameters and improve the limitations of previous measurement methods.