Abstract: For concrete dam construction, traditional computer vision target recognition methods are difficult to meet the requirements for intelligent detection in complex construction sites, as it involves narrow spaces, continuous process transitions, and various other elements such as personnel, machinery, and environment (human-machine-environment or HME). These elements often lead to occlusions, dense overlaps, and variations in size and orientation. This paper describes a new method, YOLOv5-SS, for recognition of the multiple elements in the HME scenarios of such construction. By integrating a CBAM attention module, this method improves the performance of the object detector and enhances its sensitivity to HME elements of different sizes and positions. And, it incorporates the weighted bidirectional feature pyramid network (BiFPN) to enable the object detector to focus on key image information related to real-time HME elements. To validate the recognition capability of this method, a dataset based on image information from a concrete arch dam construction site is used. Comparison of YOLOv5-SS with the YOLOv5 and Faster R-CNN models demonstrates it effectively improves the efficiency and accuracy of target detection in concrete dam construction scenarios.

Key words: concrete dam construction, human-machine-environment, multi-elements recognition, computer vision, YOLOv5-SS