Objective In subway train 360° exterior fault image detection system, the high proportion of invalid alarms is a significant issue due to factors such as environmental interference and dynamic changes of components. To address this, the UFBIAS (user feedback-based invalid alarm suppression) method is proposed, aiming to reduce manual review costs, improve O&M(operation and maintenance) efficiency and user experience, and provide technical reference for intelligent detection in urban rail transit.

Method First, the alarm position is precisely located using the CIoU (complete intersection over union) algorithm. Rules for alarm image similarity evaluation and user alarm review behavior are extracted via the ViT-L/32 model to comprehensively evaluate the alarm validity. Second, efficient suppression is achieved through "user feedback closed-loop mechanism + multi-dimensional feature fusion", to suppress pre-emptively invalid alarms and create an iterative mechanism of "detection-suppression-optimization", continuously improving the method's robustness. After improving the system's technical framework, the method effectiveness is validated by comparing the onsite detection results with the post-testing results.

Result & Conclusion The UFBIAS method, by integrating the precise positioning, comprehensive features, and user feedback closed-loop mechanism, can reduce the number of invalid alarms by over 40%, decrease manual review workload by 50%, exhibiting strong adaptability to lighting variations, component rotations, and dynamic interferences. In complex dynamic scenarios, the invalid alarm suppression indicators method significantly outperforms other existing methods.