Paper Info
Title
Performance Analysis of SVM, ANN and KNN Methods for Acoustic Road-Type Classification
Abstract
In the study, a low-cost acoustic system which classifies different roads using acoustic signal processing tool is proposed (group1 road types: asphalt, gravel, stony and snowy road; group2 road types: asphalt data with car pass noise, asphalt data with rain noise, asphalt data with tire squeal noise). Thus it is aimed to estimate road/tire friction forces using slip ratio/friction curve in the active safety systems of the automobiles. Because friction forces cannot be measured directly and it can be only observed or estimated. In the study, acoustic data features which are linear predictive coding (LPC), power spectrum coefficients (PSC) and mel-frequency cepstrum coefficients (MFCC) are used for the acoustic signal processing methods with minimum variance and maximum distance principle. The features are extracted using time windows 0.1 second as the best representative window of signal properties. The classification process is also executed by support vector machine (SVM), artificial neural network (ANN), K-nearest neighbors (KNN) algorithms and compared to different road types. The most important difference of this study from our previous studies is that it compares performances of these three classification methods for different feature vectors obtained from different road conditions and indicates that the KNN is better method than SVM and ANN methods for the acoustic road type classification. According to the results, the KNN method classifies group1 road data with %90 accuracy rate and group2 road data with % 100 accuracy rate.
Year
DOI
Venue
2019
10.1109/INISTA.2019.8778247
2019 IEEE International Symposium on INnovations in Intelligent SysTems and Applications (INISTA)
Keywords
Field
DocType
acoustic signal processing,classification,artificial neural network,support vector machine,K-nearest neighbors
Mel-frequency cepstrum,Signal processing,Feature vector,Pattern recognition,Noise measurement,Computer science,Support vector machine,Cepstrum,Artificial intelligence,Artificial neural network,Linear predictive coding
Conference
ISBN
Citations 
PageRank 
978-1-7281-1863-5
0
0.34
References 
Authors
2
2
Name
Order
Citations
PageRank
Daghan Dogan100.34
S. Bogosyan212510.53