Blind source separation and unmanned aerial vehicle classification using CNN with hybrid cross-channel and spatial attention module.

The widespread misuse of drones has become a pressing concern for both public safety and privacy in recent years. The accurate identification and classification of these devices has become a practical imperative, albeit an intricate and demanding challenge. Aiming to effectively separate mixed signals from unmanned aerial vehicles (UAVs), an improved Fast Independent Component Analysis (FastICA) algorithm is proposed. By enhancing the whitening process during the data preprocessing step, the blind source separation process becomes more stable and reliable. The experimental results indicate that the mean absolute error (MAE) of the original algorithm was 14.58%, while the MAE of the improved FastICA decreased by 10.31%, dropping to 4.27%. In order to accurately and effectively identify different types of UAVs, a UAV category recognition method based on improved convolutional neural network (CNN) is proposed. The benchmark network has been optimized by embedding a new attention module, modifying the down-sampling module, and adjusting the classifier module. The experimental results show that the improved network recognition accuracy reaches an impressive 96.30%, which is 3.01% higher than the baseline model. The experiment validates the model's performance through a series of evaluation metrics. The effectiveness of each network improvement is further corroborated by ablation experiments. This study underscores the potential and efficacy of applying convolutional neural networks in the field of UAV recognition, offering a potential technical solution for blind source separation and UAV type identification.