A method based on bispectrum analysis was proposed to enhance the accuracy of identifying radio frequency (RF) signals in drone flight controls. Given the rapid advancement of drone technology and its expanding application areas, effective detection and identification of drones were crucial for ensuring airspace safety. Special attention has been paid to radio frequency detection methods, as they can directly lock onto the signal source, achieve rapid positioning, and are less affected by external environments, resulting in particularly significant detection effects on non cooperative drones. The basic theory of bispectrum analysis was elaborated, using the publicly available DroneRFa dataset as experimental samples to carry out preprocessing, feature extraction, and other steps to obtain bispectral features of RF signals from different drone models’ flight controls. Furthermore, support vector machine (SVM) was employed for designing the signal recognition algorithm. Through 5-fold cross-validation, various kernel functions were tested, with the Sigmoid kernel function ultimately showing the best performance in recognition accuracy, reaching 96%. This method maintains high recognition accuracy even under varying degrees of noise interference, achieving approximately 70% recognition accuracy at a signal-to-noise ratio as low as -5 dB, demonstrating good noise resistance. These research findings contribute to improving drone surveillance and control capabilities, providing new ideas and methods for technological development in related fields.