Detection and segmentation of underwater CW-like signals in spectrum image under strong noise background

Abstract Aiming at the detection and segmentation of underwater Continuous wave-like (CW-like) signals of under strong noise sea background, this paper introduces a Gaussian mixture model clustering method by analyzing the signal spectrum features. First, the time domain original signal is converted to its frequency domain correspondence by Windowed Fast Fourier Transform. Second, by studying on the feature of target signal, we introduce a spectrum filtering to alleviate the background noise of ocean environment, which is analyzed and calculated with both time and frequency information. Then, the target echo location signals is constructed using a Gaussian mixture mode based binary clustering algorithm. Finally, we use the EM algorithm and adaptive binarization for solving and optimizing the clustering results. Experimental results have shown the accuracy and efficiency of our detection, which can be also simply modified and applied for detecting similar and specific signal from complex background noise.