Article Detail

İrem Hatice DOĞAN *

Keywords

Earthquake Survey Incorrect Information Disaster Iskenderun

Doi :

Abstract

The accurate and rapid classification of flying objects is of great importance in the field of aviation for security, traffic management, and defense applications. In this study, pre-trained deep learning models were utilized for the classification of objects such as birds, airplanes, helicopters, parachutes, and drones. The models used include AlexNet, EfficientNet-B0, MobileNetV2, and ResNet18. These models were structured using the transfer learning method, where only the final classification layer was trained, and they were trained for 10 epochs. This approach integrated the powerful feature extraction capabilities of the pretrained models into the classification process.e performance of the models was evaluated on the test dataset, and their accuracy rates were compared. The highest test accuracy of 99.10% was achieved by the EfficientNet-B0 model, followed by MobileNetV2 with 98.74% accuracy, and ResNet18 and AlexNet both with 96% accuracy. Furthermore, additional performance analyses were conducted by classifying the feature vectors obtained from the intermediate layers of the pretrained models using traditional machine learning algorithms (K-Nearest Neighbors, Support Vector Machines, Multi-Layer Perceptron, Random Forest, and Decision Tree). The dataset used in the study is an original dataset created from images collected from Kaggle, Roboflow, Google, and various search engines.

References

• [1] Ramesh, S. (2023). Improvements in Aerial Object Detection: Comparing YOLOv7 with YOLOv5 for Fine Drone and Bird Detection in Volatile Environments (Doctoral dissertation, Dublin, National College of Ireland).
• [2]Zitar, R. A., Kassab, M., El Fallah, A., & Barbaresco, F. (2023). Bird/Drone Detection and Classification using Classical and Deep Learning Methods. Authorea Preprints
• [3]Liu, J., Xu, Q. Y., & Chen, W. S. (2021). Classification of bird and drone targets based on motion characteristics and random forest model using surveillance radar data. IEEE Access, 9, 160135-160144.
• [4]Rahman, S., & Robertson, D. A. (2020). Classification of drones and birds using convolutional neural networks applied to radar micro‐Doppler spectrogram images. IET radar, sonar & navigation, 14(5), 653-661.
• [5]Sim, J., Jahangir, M., Fioranelli, F., Baker, C. J., & Dale, H. (2019, September). Effective ground-truthing of supervised machine learning for drone classification. In 2019 International Radar Conference (RADAR) (pp. 1-5). IEEE.
• [6]Narayanan, R. M., Tsang, B., & Bharadwaj, R. (2023). Classification and discrimination of birds and small drones using radar micro-Doppler spectrogram images. Signals, 4(2), 337-358.
• [7]Reis, D., Kupec, J., Hong, J., & Daoudi, A. (2023). Real-time flying object detection with YOLOv8. arXiv preprint arXiv:2305.09972.
• [8]Schumann, A., Sommer, L., Klatte, J., Schuchert, T., & Beyerer, J. (2017, August). Deep cross-domain flying object classification for robust UAV detection. In 2017 14th IEEE International Conference on Advanced Video and Signal Based Surveillance (AVSS) (pp. 1-6). IEEE.
• [9]Atik, M. E., Duran, Z., & Özgünlük, R. (2022). Comparison of YOLO versions for object detection from aerial images. International journal of environment and geoinformatics, 9(2), 87-93.
• [10] Drone ve Kuş: Hava Nesnesi Sınıflandırma Veri Seti:
• https://www.kaggle.com/datasets/muhammadsaoodsarwar/drone-vs-bird
• [11] Ticari Uçak Veri Seti
• https://www.kaggle.com/datasets/nelyg8002000/commercial-aircraft-dataset
• [12] new-workspace dataset. 0k81p. f lying object dataset https://universe.roboflow.com/ new-workspace-0k81p/flying_object_ dataset, mar 2022. visited on 2023-05-03. 2, 5
• [13] AhmedMohsen. drone-detection-new dataset. https: //universe.roboflow.com/ahmedmohsen/ drone-detection-new-peksv, apr 2022. visited on 2023-05-03. 2, 6
• [14] Coluccia, A., Fascista, A., Schumann, A., Sommer, L., Dimou, A., Zarpalas, D., ... & Rajashekar, S. (2021). Drone vs. bird detection: Deep learning algorithms and results from a grand challenge. Sensors, 21(8), 2824.