Vehicle Model Recognition Based on Convolutional Neural Network

Abstract

Many types of cars and the large computational quantitylead to the low accuracy and low efficiency for vehicle model recognition,to solve this,a vehicle model recognition method based on convolutional neural network (hereinafter referred to as CNN) is proposed in this paper. The method extracts the feature areas of vehicles by using bounding-box annotations. It randomly adjusts the brightness, contrast and hue of images to reduce the impacts of surrounding environments on the training effects. Finally, N images are randomly selected from the training set for batch training. The network performance comparison and optimization are realized by evaluating parameters such as the number of convolutional layers, learning rate, decay rate, momentum and moving average model. The experimental results show that the method proposed has an accuracy of 70% in recognizing the 10 models of the BMW series.