With the advancement of AI-Generated Content (AIGC), generated unrealistic images increasingly distort visual perception, posing serious challenges to distinguish between reality and fabrication. To alleviate this challenge, we propose a topological anomaly detection method. Firstly, topological connectivity anomaly phenomenon refers to regions that appear continuous on the plane but actually disconnected. Based on this phenomenon, we construct an innovative dataset containing plentiful scene images of both topological anomalous and topological normal objects and corresponding depth maps in Unity3D. Secondly, we construct an improved version of YOLOv8 integrated with depth estimation module, enabling more efficient in detecting breakpoints of pseudo-connected objects. Finally, our method is evaluated comprehensively in different experimental settings, achieving a final mean average precision(mAP) of 89.2 \(\%\) that is superior to the latest general YOLOv8 models. This research breaks through the ability of visual models to recognize situations that violate physical laws and provides a feasibility foundation for the novel field of image anomaly detection.