Real-Time Radio Signal Classification Based on Spectrograms

This paper presents a cost-effective practice-oriented approach to real-time radio signal classification based on a HackRF One software-defined radio receiver. For model training, a custom dataset comprising spectrograms of radio signals tailored to target various application scenarios is created. The developed system is capable of detecting the presence of a signal and classifying its type, provided that the corresponding class is included in the collected dataset. Laboratory tests confirm the stable performance of the classifier under real-time conditions, with an average accuracy of 99.3%. The results demonstrate the potential of the proposed solution for radio monitoring and spectrum analysis under limited computational resources.