Vehicle Type Classification in Surveillance Image based on Deep Learning Method

Vehicle type classification is an important part of intelligent traffic. With the development of research in the field of classification, especially in deep learning, many Convolutional Neural Network (CNN) architectures have been created. This becomes very challenging because increasing theaccuracy of the CNN architecture in classifying vehicle types will contribute to the field of intelligent traffic systems. The method we propose is to improve the existing CNN architecture, ResNet-50, by replacing the Global average pooling (GAP) function with a flatten layer and adding a hidden layer before softmax activation. We set the number of filter on the residual block so that the parameters used are smaller than ResNet50. Our research focuses on the vehicle front view image datasetfrom surveillance cameras for the training and testing process. From the experimental results, our proposed method in the vehicle type classification outperforms ResNet-50, VGG16 Network and CNN in previous studies by yielding accuracy of 96.26. © 2021 Elsevier B.V., All rights reserved.