Abstract: Studying the strain localization of granular materials from the meso-scale is helpful for deeply understanding the physical and mechanical natures of strain localization, and the occurrence and evolution of shear bands are the typical characteristics of strain localization. In the numerical simulations of discrete element method (DEM), observing and measuring the evolution of strain localization of sand (such as the width, inclination angle and distribution type of shear bands, etc.) greatly rely on the manual, visual and estimation ways. Towards this problem, many numerical experiments on the dense sands in biaxial compression are carried out by using PFC^2D. Based on the numerical experiments, an identification method based on "α-shape" and "Delaunay triangulation" algorithms is developed to automatically identify the shear bands and their widths. Furthermore, based on some groups of biaxial compression tests on the dense sands, the effectiveness of the proposed automatic identification method is validated.