A Time Domain Overlapping Signal Recognition Method for Multiple Modulation Types Radio Fuze Based on Deep Residual Neural Network
A method based on deep residual neural network for automatic modulation recognition of multi modu-lation type time-domain aliasing radio fuze signals in scenarios of multiple incoming ammunition from different directions was proposed to achieve accurate recognition of multi modulation type time-domain aliasing signals un-der low signal-to-noise ratio.A denoising preprocessing method for aliased signals based on feedforward denois-ing convolutional neural network(DNCNN)encoding decoding was proposed,laying the foundation for effective signal recognition under low signal-to-noise ratios;A multi modulation type fuze time-domain aliasing signal multi label classification model was established for matching target signals,and deep semantic feature extraction was performed on the denoised time-frequency images to achieve automatic classification and recognition of time-domain aliasing signals;An incremental small sample learning method was proposed for mismatched target sig-nals.By adding an additional incremental learning structure without affecting the original model parameters,in-cremental learning and online recognition of new mismatched fuze signals are achieved.The simulation results showed that this method could achieve accurate recognition of time-domain aliasing signals of different modula-tion types at low SNR,and the average recognition rate could reach 90%at-10 dB SNR.
radio fuzemulti modulation type time-domain aliasing signaldeep residual neural networkmismatched target signal
万方数据
维普
