AUV移动探测下基于U-Net网络的本体噪声去除

Self-noise Removal Using U-Net for AUV-based Underwater Target Detection

摘要:

在基于移动 AUV(自主水下机器人)平台的水下机动目标探测场景中，由于拖曳能力和探测孔径的限制，目标噪声极易淹没在本体噪声中。而且，本体噪声与目标噪声具有非常相似的频谱特性和倍频关系，进一步加重了目标噪声分析的难度。为降低本体噪声的影响、提高目标噪声的信噪比，开展了基于 U-Net 深度网络的水下混合噪声信号分离算法研究。通过仿真模型测试了算法在不同转速差、桨叶数差以及目标噪声信噪比条件下去除本体噪声的性能。实验结果初步表明：在目标信号信噪比不低于–10 dB 的条件下，算法可以对本体噪声进行有效去除。

关键词:目标探测;本体噪声去除;盲源分离;U-Net 深度网络;短时傅里叶变换

Abstract:

Due to the limitation of towing capability and detection aperture，the target noise is easy to be submerged by the self-radiated noise under the condition of underwater moving target detection based on AUV ( Autonomous Underwater Vehicle ) platform. Moreover ， the self-radiated noise has very similar spectral characteristics and frequency doubling relationships with target noise，which further aggravated the difficulty of target noise analysis. To reduce the influence of self-radiated noise and improve the signal-to-noise ratio of target noise，a source separation algorithm designed for underwater mixed noise condition based on U-Net is proposed. Performance of our algorithm is tested with different rotational speed difference，blade number difference and target noise signal-to-noise ratio. The experimental results show that the proposed algorithm can effectively remove the self-radiated noise when the signal-to-noise ratio of the target signal is not less than -10dB.

Key words:target detection;self-noise removal;blind source separation;U-Net;STFT