Identification of Road Surface Defects Using Multiclass Support Vector Machine

The objective of this research is to identify road surface defects on flexible road pavements using the Multiclass Support Vector Machine (M-SVM) method. The method of data collection is done by identifying road defect images on the road surface. There are four types of road defects identified, namely cracking, corrugation, depression, and potholes. The analytical method begins with making a pre-processing step by converting it into a greyscale mode image. Next is to do an analysis using the Gabor filter method and energy value to get data that is in accordance with the expected specifications. Then carry out the Multiclass Support Vector Machine analysis process. The results of the study show that based on energy value analysis, it provides information that potholes have a better pixel homogeneity level than other road defects. Whereas the match accuracy rate test shows that potholes road defects with an average value of 94.68% in the Gaussian kernel, and 93.53% in the Gaussian kernel. Based on the results of the match accuracy rate, the Gaussian test approach is considered to have a better level of accuracy than the Polynomial test.