Indonesia Toll Road Vehicle Classification Using Transfer Learning with Pre-trained Resnet Models

Abstract

Research in vehicle classification through various methods has become a popular field of study for decades. Mainly, vehicles are categorized based on the model, manufactured, logo, types, and dimensions, and the dataset for it is available publicly and relatively easy to get. However, based on our survey, vehicle classification based on the number of axles using deep learning has not been conducted, and the public dataset for it is not available yet. This paper aims to compose a vehicle classification based on type and number of axles then categorize it into five groups, namely, Group-I, Group-II, Group-III, Group-IV, and Group-V. This vehicle grouping refers to the Indonesia toll road tariff regulation. Nowadays, deep learning as one of the most advanced methods becomes the preferred technique to apply in image classifications due to its high performance, so do this study. Utilizing Convolution Neural Networks (CNN) as image segmentation and classification, Transfer Learning as a technique, Resnet architectures as base models, and fine-tuning as an enhancement, we can achieve accuracy about 99% for the specific vehicle classification in this study.