Classification of unmanned aerial vehicles using a lightweight deep learning method

In recent years, unmanned aerial vehicles (UAVs) have started to be used in civilian areas, apart from military and security zones, in many areas such as pesticides and surveillance in agriculture, detecting and extinguishing forest fires, transportation, observing natural life, taking aerial photographs and images, detecting and reconnaissance of damages after earthquakes. Today, the produced UAVs can have very different sizes, shapes, configurations and characteristics. This situation makes it necessary to detect and classify UAVs in important and critical places. In this thesis, a convolutional neural network model consisting of dense block and attention mechanism is proposed to classify UAVs and birds. Firstly, a dataset obtained as online was collected and data augmentation process was applied to the images in this dataset. Then, comparison studies were carried out to evaluate the classification performance of the proposed deep learning model. In these studies, pre-trained models namely AlexNet, ResNet-101, MobileNet-v2, ShuffleNet and GoogleNet were preferred. In the studies, the accuracy value of the proposed deep learning model was calculated as 98.04%, while the accuracy values of AlexNet, ResNet-101, MobileNet-v2, ShuffleNet and GoogleNet were obtained as 91.50%, 90.85%, 92.81%, 93.46% and 94.77%, respectively. From these values, it was evaluated that the proposed method not only has an effective classification performance but also can guarantee more reliable results.