Multimodal data noise processing methods, computer equipment and software products for subway stations

Abstract

This application provides a method, computer equipment, and program product for processing noise in multimodal data of subway stations. Applied to multi-source sensing data processing scenarios at subway stations, it enables noise identification and anomaly repair during multimodal data fusion. First, it performs spatiotemporal alignment processing on modal data from different devices, mapping the multi-source data to the same time slice and station spatial unit to obtain corresponding multimodal data. Based on this, it constructs dynamic fusion features for each station spatial unit in the current time slice. Based on the cross-modal consistency features in the dynamic fusion features and the feature representations of each modal data, it identifies noise embedded in the dynamic fusion features and locates abnormal modalities. Finally, it repairs the target modal data feature representations corresponding to the abnormal modalities and writes the repair results back to the dynamic fusion features for updating, thereby suppressing noise propagation and improving the accuracy of the fusion results.

Images