An Improved YOLOv8-Based Target Detection Algorithm for Online Rumor “Shoe Print Flow and Blood Flow”

Automatic detection and identification of network rumors are of great significance for maintaining network security and public order, but existing detection methods still have many challenges in identification accuracy, real-time and multimodal data processing capability. In order to effectively improve the detection efficiency and accuracy of the target of network rumor “shoe print flow + blood flow”, this paper proposes a target detection method based on YOLOv8_WRD algorithm. First, we design a multi-scale null DilateWRD architecture to adapt to the differences in the performance of police-related content in different media formats; second, we enhance the model's ability to capture details and its recognition robustness under variable backgrounds by introducing an improved adaptive attention module, the SPD-Conv module; and lastly, we add the WRDFormer two-layer routing attention mechanism and replace the backbone network with the RapidNet to replace the backbone network, which further improves the interaction and fusion efficiency of cross-modal features and optimizes the model's performance in recognizing rumors in complex network environments.