ABSTRACT
Accurately classifying vehicle axle groups plays an important role in the fields such as intelligent traffic management and Structural Health Monitoring (SHM). Currently, There are many sensors that can collect vehicle information to classify vehicle axle groups, but many of them have limitations such as the inconvenient installation of inductive loop sensors, the limited perspective and occlusion of image sensors and the poor durability of acoustic sensors. In this paper, we propose an approach for classifying vehicle axle groups based on Convolutional Neural Network (CNN) with a single strain sensor. The strain sensor is arranged under the monitored lane, and uses a one-dimensional Cross-Correlation Filter (CCF) to eliminate environmental noise. Then the axle clustering is applied for obtain the vehicle samples. Furthermore, by training a one-dimensional CNN based on GhostNet, the vehicle axle group is classified. Finally, a field test of vehicle classification was carried out on a steel suspension bridge. The corresponding results validate its high reliability and efficient in vehicle classification. The approach proposed in this paper can effectively eliminate environmental noise and classify vehicles for a long time, and effectively improve the quality of urban road network supervision.
