In construction safety inspections, visual obstructions often lead to missing features, resulting in dangerous misjudgments. To improve the efficiency of risk identification in construction, a method for occluded feature inference based on amodal completion technology was proposed, using construction fence detection as a case study. First, a dataset of fence detection images with visual obstructions was created using image synthesis techniques. Then, a combination of YOLOv8 instance segmentation and the amodal segmenter based on boundary uncertainty estimation (ASBU) feature completion network was used to infer the visual features of the occluded parts of the fence. The completed features of the occluded construction fences can be applied to various construction safety monitoring tasks, such as closed-loop detection. The approach was validated using fence images from multiple construction sites, achieving precise feature completion for occluded fences (average intersection ratio mIoU >95.5%). The research results provide a framework for feature inference in occluded construction scenes, which enhances the efficiency of intelligent construction safety supervision.