Highway Anomaly Detection Based on Deep Learning

In modern transportation systems, highways play a crucial role as a vital component of the traffic network. Anomalies on highways have a significant impact on road traffic. Therefore, real-time monitoring and accurate identification of highway anomalies are of utmost importance for enhancing traffic management efficiency and reducing accident rates. In this study, based on road monitoring data from highways in Jiangsu Province, DINO object detection algorithm and EfficientNetV2 image classification algorithm are applied for detecting abnormal elements on highways. Normalized wasserstein distance is introduced to replace GIoU in DINO, and random image cropping and patching method is used to further improve EfficientNetV2. DINO algorithm and EfficientNetV2 algorithm both demonstrate good performance.