A new approach based on convolutional neural network and feature selection for recognizing vehicle types

The number of vehicles used in traffic life has reached enormous dimensions today. increase the day by causes some problems along with it; such as congestion, accidents, pollution, and safety. To overcome all these problems, convolutional neural networks (CNN) methods are one trend recent years due to their success. In this study, a new approach is proposed use power CNN low-power devices. First all, MobileNetv1, MobileNetv2, NASNetMobile models were optimized accuracy performance. Then, an which mobile approaches only feature extractors, combining, selecting, classifying obtained features together. As result classification made approach, increased approximately 5%.