Classification of Vehicles in Aerial Imagery Using Deep Convolutional Neural Networks

Aerial imagery is important for delivery of spatial information from large surface areas. The detection and classification of vehicles in use or parked is important for the analysis of road traffic or agricultural analyzes. Modified VEDAI database with reduced number to 9 classes of vehicles is used. Deep convolutional neural network is tested for the classification purposes and the influence of number of kernels in the first layer is investigated. Achieved results show similarity between kernels for different setups. Most kernels are radial and classification results are related to the number of used classes.