Comparison of Pattern Recognition Techniques for Classification of the Acoustics of Loose Gravel

Road condition evaluation is a critical part of gravel road maintenance. One of the parameters that are assessed is loose Gravel. An expert does this evaluation by subjectively looking at images taken and written text for deciding on the road condition. This method is labor-intensive and subjected to an error of judgment; therefore, it is not reliable. Road management agencies are looking for more efficient and automated objective measurement methods. In this study, acoustic data of gravel hitting the bottom of the car is used, and the relation between these acoustics and the condition of loose gravel on gravel roads is seen. A novel acoustic classification method based on Ensemble bagged tree (EBT) algorithm is proposed in this study for the classification of loose gravel sounds. The accuracy of the EBT algorithm for Gravel and Nongravel sound classification is found to be 97.5. The detection of the negative classes, i.e., non- gravel detection, is preeminent, which is considerably higher than Boosted Trees, RUSBoosted Tree, Support vector machines (SVM), and decision trees.