A Credible Detection Method for UAV Targets Based on Deep Learning of Regularized Evidence

The increasing presence of Unmanned Aerial Vehicles (UAVs) in low-altitude airspace poses security threats such as privacy invasion and air traffic interference. Reliable UAV detection is imperative for airspace safety and to counteract these risks. However, current UAV detection methods prioritize accuracy over reliability, leading to a significant risk of misclassification when confronted with UAVs that resemble other flying objects. This paper introduces a novel UAV detection approach that leverages deep learning and regularized evidence to improve reliability and accuracy. By applying a Dirichlet distribution as a prior, the model not only predicts category probabilities but also assesses the uncertainty of its predictions, thereby enhancing the reliability of detection results. Additionally, a newly designed regularized loss function accelerates the learning process, improving both detection speed and precision. The experimental results show that the proposed method improves the mean average precision(mAP) by 3.1% compared to the baseline method, which is of great engineering significance for real-time low-altitude defense requirements.