Abstract: To enhance the automation of video data from monitoring taskson subway open-cut construction, and to intelligently survey the geometry of earth berm during the tasks, this paper proposes a method for the intelligent measurement of geometric parameters of earth bermfrom monitoring imageson subway open-cut construction. An AFU network for precise segmentation of subway open-cut construction scenes is established, incorporating skip connections with attention modules and triple cross-scale attention modules to enhance the perception of subtle structures and local features. Based on the principle of spatial stratification of steel supports, a pixel length calculation model is constructed to accurately estimate the width and slope angle of earth berm. Tests on three subway open-cut construction sites show that the AFU network achieves optimal performance in ablation experiments and comparisons with other deep learning networks. The geometric size measurements of the earth berm are closely matched with actual values. The results of this study have theoretical and practical value for risk management during construction processes.

Key words: subway construction, intelligent measurement, construction scenes image segmentation, earth berm, safety monitoring