General signal detection models typically feature large parameter sizes and high computational complexity, making them difficult to deploy directly on resource-constrained edge devices. Consequently, model lightweighting has become one of the active research topics in this field. To address this challenge, this paper proposes a lightweight communication signal detection algorithm named ISR-YOLO(I-Shufflenet Rec-SPPF(Spatial Pyramid Pooling-Fast)-You Only Look Once). A lightweight network, IShufflenet, is designed as the backbone feature extraction network, where the feature extraction modules are reconstructed to improve signal detection accuracy. In the neck network, a feature fusion module called Rec-SPPF is designed based on signal geometric characteristics to further enhance the integrity of signal bounding boxes. Additionally, the Convolutional Block Attention Module(CBAM) is incorporated to improve the sensitivity of the lightweighted network to elongated signals. Furthermore, the Enhanced Intersection over Union(EIoU) Loss, which separately calculates the height and width of bounding boxes, is adopted as the new bounding box loss function to improve the network's localization accuracy. Experiments conducted on the IEEE SPAW2021 public dataset and simulated datasets demonstrate that ISR-YOLO achieves performance almost identical to YOLOv5 while reducing computational complexity and model size by 87.34% and 82.64%, respectively, significantly decreasing the dependency on high-computing-power hardware at the edge.