基于时频图特征压缩的低复杂度CNN干扰识别方法OA

In broadband communication jamming recognition,certain algorithms that employ time-frequency analysis,e.g.the short-time Fourier transform,to convert signals into time-frequency images(TFIs)and then process them through convolutional neural networks(CNNs)for jamming recognition of-ten suffer from high complexity.To address this issue,we propose a low-complexity CNN-based jamming recognition method utilizing TFI feature compression.The proposed algorithm first exploits the inherent redundancy and non-essential information in typical jamming TFIs by jointly characterizing interference signal features through time-domain mean filtering,frequency-domain mean filtering and peak-value fil-tering.The filtered outputs yield three sets of one-dimensional time-frequency feature sequences,which are then fed into a CNN for jamming recognition.While ensuring recognition performance,the proposed approach significantly reduces complexity by decreasing the input data volume and the CNN dimension.Experimental results show that for the common seven types of oppressive jamming,compared with the tra-ditional CNN recognition method based on time-frequency diagrams,the proposed method reduces the number of network parameters by 98.78%,decreases the network computation amount by 93.57%,and increases the recognition accuracy by 2 dB at low jamming-to-noise ratio conditions.Furthermore,the proposed method outperforms the low-complexity schemes,such as depth-wise separable convolution,network pruning,and time-frequency graph size compression,in both recognition accuracy and network complexity.The proposed method is particularly suitable for real-time jamming recognition in resource-constrained equipment such as unmanned aerial vehicles(UAVs)and portable communication equip-ment,offering a high precision and low complexity solution for jamming recognition in complex electro-magnetic environment.