Computer vision and bi-directional neural network for extraction of communications signal from noisy spectrogram

Extraction of communication signals from noisy spectrograms is a challenging problem which has not been explored extensively from an intelligent signal processing and computer vision based perspective. In this paper we propose a novel technique of extracting the communications signal from a noisy spectrogram using a combination of fuzzy neighborhood thresholding based self organizing neural network and morphological operations. We show that about 98% detection is achieved at 5% false alarm of a particular scenario outperforming traditional energy detection.