Abstract:The accurate classification of road surface conditions plays a vital role in ensuring road safety and effective maintenance. Vibration-based techniques have shown promise in this domain, leveraging the unique vibration signatures generated by vehicles to identify different road conditions. In this study, we focus on utilizing vehicle-mounted vibration sensors to collect road surface vibrations and comparing various data representation techniques for classifying road surface conditions into four classes: normal road surface, potholes, bad road surface, and speedbumps. Our experimental results reveal that the combination of multiple data representation techniques results in higher performance, with an average accuracy of 93.4%. This suggests that the integration of deep neural networks and signal processing techniques can produce a high-level representation better suited for challenging multivariate time series classification issues.

What problem does this paper attempt to address?

### Problems the Paper Attempts to Solve This paper aims to address the issue of road surface condition classification. Specifically, it classifies road surface conditions by utilizing road surface vibration data collected from in-vehicle vibration sensors and comparing various data representation techniques. The study categorizes road surface conditions into 4 types: normal road, pothole, bad road, and speed bump. The paper points out that accurately classifying road surface conditions is crucial for ensuring road safety and effective maintenance. Vibration-based technology shows potential in this field by using the unique vibration characteristics generated by vehicles to identify different road conditions. Experimental results indicate that the combination of various data representation techniques can improve performance, achieving an average accuracy rate of 93.4%. This suggests that the integration of deep neural networks with signal processing techniques can produce higher-level representations, thereby better addressing the complex multivariate time series classification problem.

Evaluation of data representation techniques for vibration based road surface condition classification