Blind Detection of Communication Signals Based on Improved YOLO3

Abstract

Blind detection of communication signals is a challenging task. In this paper, a general and novel blind detection method is proposed based on the similarity between communication signal detection and image object detection. We designed an improved YOLO3 model to detect the communication signals contained in the 2D wide-band spectrograms, the main innovates are as follows: 1) in order to reduce the burden of spectrograms labeling, an ingenious and automatic signal object labeling method is proposed; 2) in view of the fact that the communication signals are long and narrow objects in the spectrograms, the corresponding prior anchors are designed to improve the detection probability; 3) in order to improve the training efficiency and detection accuracy, the CIOU cost function and DIOU-NMS inference algorithm are introduced to achieve high-precision signal detection. The simulation results demonstrate that the proposed method can effectively detect the continuous and burst signals in wide-band communication signal data, and its performance is better than the traditional energy detection method.