Combining CNN Deep Learning and SVM for Vehicle Classification

Vehicle classification plays an important role in an intelligent transportation system. The development of research in image processing, pattern recognition, and deep learning has encouraged more research interest in categorizing vehicles using deep learning technology. In everyday life, Automatic vehicle classification can be found in vehicle search, real-time traffic monitoring, automatic toll gates, and others. The use of vehicle classifications as a vehicle search problem is commonly used by a police officer in the case of car robbery crimes. The use of a human expert, in this case, is completely accurate but requires time and money. In this case, automatic categorization of a vehicle using deep learning technology can be one solution. Vehicles can be categorized into four categories: Minivans, Pickup Trucks, Sedans, and SUV. The method used in this research is to use a combination method of Convolution Neural Networks (CNN) as a feature extractor with Support Vector Machine (SVM) as a classifier to increase accuracy. The deep learning architecture used is CNN (AlexNet, VGG16, and Resnet50). The dataset used in this research is the AI Standford Car Dataset which consists of 16,185 images and 196 vehicle classes. From 196 classes, four classes were selected (Minivan, Pickup Truck, Sedan, and SUV). The accuracy obtained is AlexNet + SVM 38.50 %, VGG16 + SVM 82.93 % and ResNet50 + SVM 90.87 %.