Abstract: Aiming at the problem that low signal-to-noise ratio causes the low resolution ability of multi-source target azimuth estimation in the shallow sea complex positioning environment, an algorithm for the deconvolution of the offgrid sparse reconstruction of multi-target underwater acoustic signals based on off-grid sparse Bayesian learning is proposed. First, the algorithm uses the Wiener filter deconvolution algorithm to "de-noise" the signal received by the array, and then performs singular value decomposition of the signal data to reduce the noise and the computational complexity in the signal reconstruction process. Secondly, the off-grid sparse signal model is established, and the maximum posterior probability is obtained by the Bayesian learning algorithm. Finally, the estimates for the direction of arrival (DOA) of multiple target sources are obtained. The proposed algorithm uses Wiener filtering deconvolution super-resolution algorithm to best estimate the signal received by the array, thereby suppressing noise interference, obtaining higher resolution, and improving the detection performance of multiple objects. Simulation analysis and sea trial results show that, compared with MUSIC algorithm and OGSBI algorithm, the root mean square error of azimuth estimation of this method is within 1° when the signal-to-noise ratio is greater than -8 dB, and it has better performances in multi-target positioning accuracy, algorithm robustness and operating speed, which provides a reference for the DOA estimation of underwater multi-target.