Using a deep learning neural network to predict flight path

The subject of this paper is a new approach using a deep learning neural network designed for predicting the flight path of an unmanned aerial vehicle (UAV). The purpose of this study was to improve the accuracy of drone flight path prediction by developing a deep learning-based trajectory forecasting model. The task was to collect and prepare a dataset of video and photo materials for training the neural network, develop and implement a deep learning model for trajectory prediction, and enhance UAV flight trajectory forecasting through model optimization and validation. Methods used included the creation of a synthetic dataset using the 3D modeling tool Blender, which enabled the generation of animations representing various drone flight scenarios. These scenarios include different environmental conditions and urban landscapes, providing a robust training ground for the neural network. To further improve and test the model’s predictive capabilities, real-world data, including eyewitness videos, were used. The architecture of the neural network includes long short-term memory (LSTM) units that can process sequential data, making them ideal for predicting dynamic UAV trajectories. The training process involved several stages, starting with pre-training on general visual features and then fine-tuning to UAV-specific motion patterns. The results of this study show that the neural network achieved high accuracy in trajectory prediction, with the model showing better performance in real-world scenarios compared to traditional trajectory prediction methods. The integration of LSTM enabled efficient learning and generalization of temporal data, capturing complex motion patterns and interactions with the environment. This research not only demonstrates the feasibility of using deep learning to predict UAV trajectories but also offers potential applications in civilian security or delivery logistics, where real-time trajectory prediction can significantly improve the efficiency and speed of decision-making. Conclusions. The scientific novelty of the obtained results lies in the development and training of deep learning models specifically designed for predicting drone flight paths. This study demonstrated the effectiveness of the proposed approach by demonstrating its ability to enhance the accuracy of UAV trajectory forecasting.