WiFi Emitter Identification Based on Few-Shot Convex Optimizable Network

The method based on deep learning applied in the field of communication signal emitter classification has been fully developed and achieved good classification performance. However, This requires ample labeled data to learn, and data acquisition and processing in the field of communication signals is a costly task. In this regard, based on the method of few-shot learning, this paper proposes Convex Optimized Linear Classifier with Attention Mechanisms(AM-COLC) framework for small samples emitter Identification. SVM classifier is used to solve the classification problem as a convex problem which improves the generalization of the model on new classes. we use a convolutional attention mechanism to improve end-to-end training and obtain better classification performance. In order to discuss the generalization and robustness of the signal identification model used in this article, we designed a comparative experiment to compare the classification performance of the baseline model in the experiment from the number of samples and the size of the model, indicating that the model has a certain inhibition of overfitting effect and is more suitable for signal classification with fewer samples.