Global Context Attention in YOLO for Broadband Signal Identification

Abstract

Detecting broadband communication signals is crucial for signal processing, as its accuracy directly impacts subsequent processing such as information restoration. However, due to the ever-growing complexity of the electromagnetic environment, traditional broadband detection algorithms are no longer suitable for current application needs. This paper proposes a broadband signal detection model, NLYOLO. We improve the precision of the image by analysing the characteristics of the target signal by selecting suitable parameters. We also select anchors adapted to the special shape of the target signals and introduce attention modules Non-local and SimAM in the feature extraction of NLYOLO and improve the loss function. This enhances the ability of high-level semantic information feature extraction and improves the precision rate and recall of small target signals. The experimental results show that compared with the traditional energy detection algorithm and the baseline model, the detection precision and recall rate are significantly improved, which effectively achieves the target detection task in the field of signal processing in complex scenes.