Аннотация
This article considers the problem of visual discrimination between birds and unmanned aerial vehicles (drones) in connection with the task of automatic recognition of objects in the airspace. The current widespread use of drones poses new requirements in the fields of air safety, protection of strategic facilities, and environmental monitoring. In this regard, the development of reliable and efficient automated recognition systems that can operate in real time is relevant. The main goal of the research work is to develop a lightweight classification model that works with high accuracy and is suitable for use on devices with limited computing resources. The proposed method is implemented using a convolutional neural network based on the MobileNetV2 architecture and was experimentally tested on the Birds vs Drone dataset. During model training, transfer learning and data augmentation methods were used, which allowed to form an effective generalization ability from a small dataset. In addition, these approaches reduced the probability of overfitting of the model and increased its stability and reliability. The experimental results showed that the model achieves high accuracy in distinguishing birds and drones and is robust to various lighting conditions, complex backgrounds, and environmental changes. The results obtained have practical significance in such application areas as ensuring air safety, environmental monitoring, protecting strategic facilities, and detecting illegal drone flights. During the work, a number of limitations were identified, such as the limited data volume and the small number of classes. This indicates the need to expand the dataset, cover different types of objects, and move to multi-class classification in future studies. Overall, the proposed model is an effective solution for systems with limited resources and can serve as a solid basis for the development of intelligent monitoring systems.