Airport Buildings Classification through Remote Sensing Images Using EfficientNet

The classification of buildings in airports using remote sensing images is a challenging problem. The challenge is to classify the same types of buildings in each airport that may differ in terms of shape, size, and color. In order to overcome these problems, we use the EfficientNet architecture, which achieves high accuracy with low numbers of parameters and relatively inexpensive computational time compared to other architectures. In this work, the proposed problem is handled by utilizing the EfficientNet concept in five versions (B0-B4) by classifying four types of buildings in the airport; the passenger terminal, the radio tower, the runway, and the car park. The dataset covers 322 airports in Asia in the RGB format, with a resolution of 560×560 pixels. The experimental validation of the used method is performed through a partition of the dataset; 70% in the training set, 10% in the validation set, and 20% in the test set. We found that EfficientNet-B4 is a tremendously applicable architecture. It provides a high accuracy level of up to 90%, which is considerably more than the other architectures.