This paper investigates and develops a novel model based on the combination of chaos analysis and transformer architecture for drone detection from audio signals. The main goal is to improve the accuracy of drone, helicopter, and sound field detection in complex and noisy conditions. In this study, complex and nonlinear audio features were obtained using chaos-based feature extraction methods and then analyzed with a transformer-based model. This approach results in more accurate profiles and better learning than conventional models. Experiments were conducted on a dataset of 90 audio signals from the open-source Drone Detection Dataset project. The model results showed that all test samples were classified with a perfect accuracy of 1, indicating the high efficiency of the model. This performance is even more outstanding compared to the results of similar research with an average accuracy of about 0.93. This study shows that the combination of chaos and transformer has a powerful ability to accurately identify drones and distinguish them from other sounds. This research can pave the way for the development of Drone acoustic identification systems with higher accuracy and efficiency in real and complex environments and play an effective role in air security and surveillance.