UAV Signal Classification Methods Design Based on Neural Network Encoder

Unmanned Aerial Vehicles (UAVs) are becoming increasingly prevalent in various sectors such as healthcare, agriculture, environmental monitoring, disaster management, and more. Consequently, there arises a critical need to develop a robust system capable of accurately discerning UAV signals in complex environments. This paper intends to present a novel UAV signal recognition system predicated on the classification of signals emitted by both the UAV itself and its controller. Leveraging the Wavelet transform, this system extracts pertinent features, subsequently employing a neural network encoder to differentiate between UAV signals and controller signals, as well as distinguishing among various UAV and controller models. The proposed model adopts an autoencoder architecture and introduces innovative enhancements in two key frameworks. The initial framework integrates skip connections within residual layers, thereby ensuring the preservation of essential information during convolutional operations. Simultaneously, the second framework harnesses this retained information to extract supplementary features, thereby enhancing the capability of the model to discern intricate patterns within the data. This amalgamation of the two key frameworks in the proposed model facilitates a more profound comprehension of the embedded feature information, and thereby enhances the efficiency of the system. The proposed model has achieved remarkable results in the experiments of UAV signal classification task, indicating that the optimization of neural network structure in this model is effective for RF signal processing.