Abstract: The aim of this study is to investigate the changes in the microstructure of loess under dry-wet cycles.The gray-scale image texture features of Xi’an loess were extracted based on the gray-level co-occurrence matrix,and the area of cracks and pores in the loess was calculated based on the percentage of the area of cracks and pores.The relationship between the gray-scale texture characteristics and the proportion of cracks and pore areas was established through the deep learning time-series regression prediction model,and the damage degree of loess under dry-wet cycles was determined by calculating the damage factor.Our study revealed that within two dry-wet cycles,the edge structure of soil aggregates was destroyed,and cracks and pores increased sharply;within five dry-wet cycles,the texturing trend of soil structure became more obvious,approaching the direction of parallel water migration;after four times of dry-wet cycle,the damage ratio of loess under dry-wet cycle reached 93.10%;after six dry-wet cycles,the damage no longer increased,which means that the microstructure texturing of the loess stabilized.

Key words: loess, dry-wet cycle, microstructure, gray-level co-occurrence matrix, deep learning