Computer vision for improved single-sensor spectrum sensing

Spectrum sensing is the technology for determining spectrum usage as seen from a single sensor. This paper presents an improved approach to spectrum sensing. The technique is based on processing the spectrogram using computer vision methods, including morphological operations and computing connected components. The use of these non-linear methods improves performance. For a fixed false alarm rate of 10%, the technique presented improved detection accuracy from 50% to near 90% over a time adaptive method based on cell-averaging.