In this paper, we propose the application of deep convolutional neural networks for the detection of radio signals in the broadband frequency spectrum. The RFROI–CNN (Radio Frequency Regions of Interests) approach uses a different way of proposing regions than in typical convolutional object detectors in images – instead of proposing regions based on the distinguishing features of specific radio transmissions, the radio signals are treated as masks that obscure the actual noise distribution, and a deep network is used to estimate the noise distribution in the broadband radio frequency spectrum. This approach allows the SNR of the signals to be maximised in signal processing and, consequently, detection characterised by higher values of quality indicators.