Classification of UAVs Utilizing STFT Features of RF Fingerprints and ASPConv-ResNet34 Network

Classification of unmanned aerial vehicles (UAVs) plays an important part in spectrum monitoring. This paper investigates the classification of UAVs from radio frequency (RF) fingerprints on the basis of using ResNet network. Firstly, we converts the RF signals captured from UAV controllers to Short-time Fourier transform (STFT) images. Then, we proposes an adaptively spatial parallel convolution (ASPConv) module to enable neurons to adaptively obtain the corresponding spatial information through multi-scale receptive fields, thereby improving the feature extraction ability. And ASPConv module replaces Conv1 in ResNet34 to form a new network called ASPConv-ResNet34, which applies Hardswith activation function. Using the proposed ASPConv- ResNet34, we classify a total of 17 different commercially available UAVs with an average testing accuracy of 99.97%, which significantly outperforms the existing approaches to our best knowledge.