Detection of Wet Road Surfaces from Acoustic Signals using Scalogram and Optimized AlexNet

Wet road surface increases the risk of traffic accidents. Hence, there is a need for automated systems that could detect this and provide early warning to the road users. The study proposes to detect wet road surfaces using acoustic signals and convolutional neural network. Data is acquired from the IDMT-Traffic database. The acoustic measurements are then, converted into scalogram and used to train the AlexNet. Two optimizers and learning rate settings are assessed in this study. The best performance is attained with Adam optimizer and low learning rate, yielding validation accuracy of 89.9%. Generally, implementation of acoustic signals and optimized AlexNet is feasible for detecting wet road surfaces.