Deep Convolutional Neural Network-based Recognition and Extraction Classification Research for UAV RF Signal Detection and Classification
In order to realize the implementation of the detection and identification of UAV signals in the high-altitude space environment and the task scheduling of multi-moment signal spectrograms,a UAV RF signal identification scheme based on the deep convolutional neural network algorithm is proposed,in which the UAV detection base station searches for and intercepts the RF signals within the frequency range of 300KHz～30GHz,and applies the short-time Fourier transform function to classify the UAV time-frequency domain signals with windowing,squaring,and extraction.Function of the UAV time-frequency domain signal to make a window division,square amplitude of the signal spectrogram calculation,the use of Haar wavelet transform filtering analysis method,filtering out low-frequency or high-frequency noise signals within the spectrum of the signal,and data preprocessing of the signal into the multi-branch convolutional neural networks(Convolutional Neural Networks,CNN)model,reasonable setup of multiple size convolutional kernel,global convolutional layer,pooling layer and other layers,to make downsampling recognition training set extraction calculation of UAV RF signal features,in order to realize the improvement of the detection and recognition efficiency,accuracy and environmental anti-interference ability of the UAV signal spectrogram detection and recognition in the complex electromagnetic environment.
deep convolutional neural networkUAV RF signaldetection and identificationtask scheduling
万方数据
