An Application of Deep Learning YOLOv5 Framework to Intelligent Radio Spectrum Monitoring
Thanh Le Truong, Ngoc Thien Le, Watit Benjapolakul · 2022
In this paper, we propose an intelligent radio spectrum monitoring system based on spectrogram image detection. The system utilizes the You Only Look Once 5thversion (YOLOv5) as the framework's core. YOLOv5 is a widely-known, powerful, and efficient deep learning framework for object detection. We use Electrosense devices as the spectrum sensor to collect the dataset for training YOLOv5 model. The spectrum sensor connects to the Electrosense server and retrieves the Signal to Noise Ratio (SNR) to present the spectrogram. The trained YOLOv5 then detects the frequency bands from spectrogram images by bounding boxes. The trained YOLOv5 performance achieves 99.3% precision and 100% recall (sensitivity) on the training dataset. Compared with paper [1], which has 99.6% accuracy, the proposed model seems a little less accurate, but this is an object detection model with more complexity than classification.