Road Surface Defects Detection Based on IMU Sensor

To solve the problems of low efficiency, high cost, and limited detection time in road surface defect detection, this paper proposes a fully connected neural network (FCNN) based on the inertial measurement unit (IMU) because of the characteristics of IMU with fewer data but carrying more information. Different signals (acceleration, velocity, and Euler angle) were processed by the data processing method proposed and made into databases with various features. Four methods, including random forests (RF), support vector machines (SVM) with linear and RBF kernel functions, light gradient boosting machines (LightGBM), and the FCNN, were used to compare detection accuracy. The results show that the accuracy of most methods increases with the number of sample features. The FCNN proposed in this research has the best performance, and RF shows the worst performance. The FCNN achieved an accuracy of 99.4% on the test set, which is much lower than other machine learning algorithms.