Small-Sample Jamming Signal Recognition Based On Similarity Calculation

The neural network based classifier can automatically perform feature extraction and achieve high recognition accuracy in the communication jamming signal recognition. However, it typically requires a large amount of labeled data for pre-training the model, which is frequently unavailable in real-world situations. In this paper, a small-sample recognition method based on similarity calculation is proposed to deal with the jamming signal recognition under the constraint of limited labeled samples. The proposed method uses two identical twin networks to design a similarity calculation network, which can compute the similarity between two samples for type recognition. To harness additional prior knowledge, we have constructed a similarity calculation training set based on the MINIST samples, which is employed for the pre-training of the similarity calculation network. Utilizing the pre-trained network, we calculate the similarity between unknown jamming signals and known signals. The results indicate that the algorithm can effectively improve recognition accuracy when there are a limited number of samples.