Vehicle classification with deep learning

With the development of deep learning methods, image, gene and sound analysis, robotics, autonomous systems, medical field diagnosis has been started to be used in many areas. High accuracy in solving such problems has led to a widespread use. In this thesis, the data set of approximately 20000 different data sets of 10 different brands / models of vehicles are trained by transfer learning method of Faster R-CNN Resnet50, Faster R-CNN ResNet101, R-FCN ResNet101 and SSD mobilenet networks. As a result of the tests, Faster R-CNN ResNet50 model was the most successful with 94.4% accuracy. The training results were observed by using a dataset consisting of unused images. The open source TensorFlow library and Python programming language developed by Google were used during the training phase. In our model, it has been observed how the performance changes by using different parameters. Successful results were obtained as a result of the trainings. Tables and loss graphs related to the results are shown at the end of the section.