A Light‐Weight and Controllable Attention Mechanism for Interleaved Signal Recognition with High Similarity

ABSTRACT To tackle the challenge of recognizing highly similar interleaved signals, this paper proposes a novel recognition method leveraging a spatial dimensional upgrading transformation squeeze and excitation (SD‐SE) block. Unlike traditional attention mechanisms reliant on internal back‐propagation, SD‐SE block offers enhanced controllability and effectiveness. It introduces SD‐SE, a pre‐weighted training network, using a lightweight, manually controllable approach. The block addresses issues of initial weight randomness and incomplete signal characteristic utilization in attention mechanisms. Furthermore, SD‐SE mitigates CNN's tendency to lose key information during feature extraction. Experiments show SD‐SE achieves superior accuracy in recognizing highly similar interleaved signals, resolving challenges in complex signal recognition. Notably, this lightweight module is compatible with various attention networks, broadening its applicability.