Low-SNR Digital Modulation Signal Detection Based on an Improved YOLOv11

To address the challenges of low signal detection accuracy under low signal-to-noise ratio (SNR) conditions and suboptimal performance in identifying small targets within timefrequency representations, this paper proposes a conventional modulation signal detection approach based on an improved version of the YOLOv11 model, termed YOLOv11-LD. The proposed method first applies wavelet denoising to the sampled signals, followed by converting the denoised signals into timefrequency plots, which are then fed into the improved YOLOv11 network for target detection. Experimental results demonstrate that, compared to the original YOLOv11 model without denoising preprocessing, incorporating wavelet denoising significantly enhances detection performance, with mAP50 and mAP50-95 increasing by 3.1% and 5.3%, respectively. Furthermore, with the integration of the proposed architectural improvements, the YOLOv11-LD model achieves an additional 2.3% and 2.1% improvement in mAP50 and mAP50-95, respectively, over the denoised baseline YOLOv11 model. Notably, the method maintains high detection accuracy under low SNR conditions ranging from -6 dB to 0 dB, demonstrating strong robustness in challenging communication environments.