An Accurate Deep Learning Model for Vehicle-Type Classification

Vehicle-type recognition can provide data support for expressway vehicle information retrieval systems and electronic toll collection. We developed the initial model based on the faster region-convolutional neural network (Faster RCNN) deep learning network, which aims to detect the original vehicle image data for the expressway vehicle face data set generation. To find the best classification model, we developed three different models based on InceptionV3, improved ResNet-50, and Xception, respectively, according to the transfer learning theory. Several measures were employed during the analysis of the experimental results, including accuracy, precision, and recall, for the evaluation. The experimental results showed that our proposed ResNet-50 model outperformed other models with 97.3% classification accuracy. The average accuracy and average recall of our proposed ResNet-50 model were close to 97%. Considering the comparison of the outcomes, we have seen that our improved ResNet-50 model is more appropriate for vehicle-type classification in scenarios that take place on expressways.