Application of an Oversampling Method for Improving Road Surface Classification

This paper presents the application of oversampling methods to improve the road surface classification performance. A large number of multimodal experiments were implemented with the following configuration: three accelerometer vibration signals from acceleration in the X-, Y-, and Z-axis; three microphone acoustic signals; two speed channels; and the torque and position of the handwheel. Data acquisitions were made under different settings: three worm-gear interface configuration; hands on or off the wheel; vehicle speed (constant speed of 10, 15, 20, 30 km/h, or accelerating from 0 to 30 km/h); and road surface (smooth flat asphalt, stripes or cobblestones). The objective was to identify the different kinds of road surfaces where the car travels during the experiments. Testing of the system considered small training datasets augmented using two oversampling methods: Synthetic Minority Oversampling Technique (SMOTE) and Generative Adversarial Network Synthesis for Oversampling (GANSO). It is demonstrated the capabilities of data augmentation to reduce the classification requirements for the size of training datasets in the proposed application.