Comparison of Current Convolutional Neural Network Architectures for Classification of Damaged and Undamaged Cars

Abstract

Deep learning algorithms, one of the most popular topics of recent years, are frequently encountered in many areas. One of these areas is traffic. Many applications such as vehicle identification and damage detection using images obtained by various imaging methods in traffic are successfully carried out by deep learning algorithms. In this study, damaged and undamaged car images are detected using current convolutional neural network architectures. In this study, the publicly published data set named damaged car dataset (DCD)-2 was used. This data set includes 80 normal and 80 accident vehicle images. The low number of images in the preliminary analyzes made it cause us to increase the data (image augmentation). With the increase in data, the number of images was increased to 400 damaged vehicles and 400 solid vehicles. VGG-16, VGG-19, Inception V3, NasNetLarge, DenseNet201 were applied from the transfer learning algorithm to the data incremented dataset. At the end of the application, the classification accuracy was 84.11% with VGG-16, 81.96% with VGG-19, 94.11% with Inceptionv3, 98.39% with NasnetLarge and 97.14% with Densenet201. © 2022, The Author(s), under exclusive license to Springer Nature Switzerland AG.