Attentional Cross-Layer Fusion Network for Enhanced Spectrum Sensing

Recently, spectrum sensing algorithms based on convolutional neural networks have achieved excellent performance. However, most existing algorithms rely only on features from the highest convolutional layer, potentially limiting further performance improvements. To address this, this paper proposes an attention cross-layer fusion network for enhanced spectrum sensing. The designed fusion module adaptively combines multilayer features via attention mechanisms, effectively preserving critical details while leveraging high-level semantic information. Simulation results show that the proposed algorithm outperforms classical spectrum sensing algorithms. Notably, the proposed algorithm achieves a detection probability of 98.2% with a false alarm probability of 10% at SNR = -18dB.