Comparison of different convolutional neural network techniques for car models classification

Car model classification is crucial as part of recognizing suspicious cars on the road that irresponsible owners have modified to obscure the views of police authorities for criminal purposes. This study presents car model classification using a deep learning state-of-the-art Neural Network architecture called Convolutional Neural Networks (CNN). Rather than being taught the concepts of a car, computers could study images, learn the concepts on their own using CNN, and classify them into their respected models using image processing methods. Different pre-trained CNNs were used in this study, including AlexNet, ResNet, and GoogleNet. The main idea of using different pre-trained CNNs is to compare the performance of the classification task by using various networks. The experimental results show that ResNet performed better than AlexNet and GoogleNet in classifying the modified cars into their actual class.