Specific Emitter Identification of ADS-B Signal Based on Visual State Space Duality

In recent years, deep learning has developed rapidly, significantly empowering various fields, and specific emitter identification (SEI) is one of them. Performance and efficiency are two key indicators of deep learning tasks. In this paper, the non-causal state space duality (NC-SSD) with superior performance and linear algorithm complexity is taken as the core, and the visual state space duality (VSSD) model where the core was originally located is improved by using pooling depthwise convolution (PDC) downsampling strategy, injecting positive-incentive noise (PN), and adding batch hard triplet loss (TH) function. The improved visual state space dual (IVSSD) model proposed in this paper achieves performance and efficiency improvement compared to the VSSD model in SEI of automatic dependent surveillance-broadcast (ADS-B) tasks, and also holds advantages over models such as ConvNeXt, MaxVit, and TransNeXt, with a lead in classification accuracy of no less than 2%. In terms of features, this article achieves higher identification accuracy compared to a single feature by using a simple and highly scalable channel fusion method for multi feature fusion.