A Novel Lightweight and High‐Performance Detection Framework for Signal Detection in Wideband Spectrogram

ABSTRACT With the explosive growth of Internet of Things (IoT) devices, broadband signal detection in low‐signal‐to‐noise ratio (SNR) non‐cooperative environments faces severe challenges. Although deep learning has been widely applied to broadband spectrum signal detection, existing research on low‐SNR environments remains insufficient. In such scenarios, the time‐frequency representations of signals are highly susceptible to noise interference, resulting in difficulties in feature extraction and high false alarm and miss detection rates for existing deep learning models. Meanwhile, due to their high computational complexity, most deep learning models fail to meet the real‐time processing requirements for massive data. Therefore, achieving high precision and fast detection of broadband signals under low SNR has become an urgent problem that needs to be solved. This letter introduces two lightweight modules, namely LFSI‐DH and EOMFE, and integrates them into the YOLOv8 framework. Experiments demonstrate that the framework reduces complexity and improves accuracy compared to the YOLOv8, offering a viable edge device spectrum monitoring solution. Comparative analysis further validates its superiority in detection accuracy.