Machine Learning Road Recognition Based on Wavelet Decomposition

Abstract

Driving road conditions have a great impact on vehicles and personnel. According to road conditions, reasonable adjustment of components such as suspension and electronic control unit parameters can effectively improve ride comfort and handling stability. This paper studies the method of road Recognition from the perspective of machine learning, and compares the Recognition effect of BP neural network and SVM. Using wavelet decomposition, a time-frequency analysis method, the original signal is decomposed into different frequency band signals, and the difference between different road surfaces can be amplified. The commonly used statistics are screened by Fisher’s criterion to obtain excellent data of each dimension of the sample. The method can achieve an Recognition accuracy of nearly 100% in the simulation experiment. In the vehicle experiment, the four kinds of road surfaces are well distinguished, and the comprehensive accuracy is about 82%.