A Comparative Study of Deep Convolutional Neural Networks for Car Image Classification

Abstract

This paper aims to present the result of a comparative study of Deep Convolutional Neural Networks (CNN) using the AlexNet architecture to use the car image classification of a small dataset. We have proposed the experiment result from a comparative study dropout value using Cuckoo Search (CS), of the optimization techniques for a small data set solving problem of overfitting. The car images for the experiment are different in color, size, and position. As a result, the training time average of $\sim 59.16$ minutes, and the model accuracy of 91.41%.