In recent years, artificial intelligence (AI) technology has achieved breakthrough progress in interdisciplinary fields. AI core algorithms have demonstrated considerable professional-level performance in applications such as image recognition, processing, and complex decision-making. Therefore, to promote efficient scheduling and accelerate progress in rail transit construction scenarios, This paper focuses on the unique scene of the construction site. Based on the deep learning object detection algorithm YOLO v5, an end-to-end, real-time, and high-precision image recognition model for construction sites is designed. This model can automatically receive, recognize, classify, and store images captured by construction site workers. Concurrently, in order to deal with the challenging scenario of construction site, this paper significantly improves the verification indicators of YOLO v5 through data augmentation, attention mechanism and knowledge distillation. By identifying key elements of the construction phase, such as dust nets and tower cranes, the system can help assess whether work is proceeding as planned and reduce human error.