Every land moving object accelerates or decelerates based on the fictional coefficient of the road surface. It has been known that this coefficient on the road is determined by the type of road surface. In this work, we propose a simplistic, machine-learning based solution to estimate the road type using the reflected ultrasonic signals paired with ultrasonic transmitter and receiver. Since the reflected signal contains the material information of the surface due to the difference in the surface roughness and acoustic impedance, different characteristics can be observed for each frequency of the reflected signal. To exploit such characteristics, the signals are transformed into the frequency domain using short-time Fourier transform. In addition, a deep convolutional neural network is applied as the road identifier due to its well-known representational power. In order to verify the aforementioned ideas, the ample database consisting of eight types of road surfaces are obtained with the ultrasonic sensors. And then, the database is used to train the model, as well as to evaluate the accuracy of the trained model. It can be seen that the proposed method makes it easier and more accurate to identify the type of road surface than the conventional methods.

每个地面移动物体都会根据路面的虚拟系数进行加速或减速。众所周知，道路上的该系数由路面的类型确定。在这项工作中，我们提出了一种简单的基于机器学习的解决方案，以使用反射的超声波信号与超声波发射器和接收器配对来估算道路类型。由于表面粗糙度和声阻抗的差异，反射信号包含表面的材料信息，因此对于反射信号的每个频率，可以观察到不同的特性。为了利用这种特性，使用短时傅立叶变换将信号变换到频域。此外，深卷积神经网络由于其众所周知的表示能力而被用作道路标识符。为了验证上述想法，利用超声波传感器获得了由八种路面组成的充足数据库。然后，该数据库用于训练模型以及评估训练模型的准确性。可以看出，与传统方法相比，所提出的方法使识别路面类型更加容易和准确。