Classification of main components in airports from remote sensing images by efficient network

Abstract

In this thesis, we classify the type of main components in airports from the remote sensing images. This datasets is considered as an interesting information since the same type of component may have different shape, size, and color. EfficientNet architecture is the deep learning architecture to use in this research due to the small number of parameters and computational time compared to the other architectures with similar accuracy. In our experiment, we apply the EfficientNet in versions B0, B1, B2, B3, and B4 to classify four types of components in the airport; the passenger terminal, the radio tower, the runway, and the car park. We collect the RGB format datasets of 322 airports in Asia with a resolution of 560x560 pixels. Then, the datasets is partitioned into 70% for the training set, 10% for the validation set, and 20% for the test set. From the experimental results, we conclude that EfficientNet-B4 is a suitable architecture for our datasets, which provides a high accuracy up to 90%.