Adaptive Convolution Kernels Construction Based on Unsupervised Learning for Underwater Acoustic Detection
In the field of Direction of Arrival (DOA) estimation, due to the complexity of ocean noise and the limitations of using array apertures, the bearing time record (BTR) obtained by CBF typically exhibits low signal-to-noise ratio (SNR) and wide mainlobe width. To address this issue, we propose an adaptive 2D convolution kernel construction method that utilizes an improved k-means clustering algorithm to extract adaptive mainlobe visual patterns from historical BTR data as convolution kernels. Experimental results show that our method can effectively reduce noise levels within multiple target environments.
Authors
- Zongyi Qin (ORCID: https://orcid.org/0000-0002-6329-4005)
- Fan Yin (ORCID: https://orcid.org/0000-0002-8028-2217)
- Chao Li (ORCID: https://orcid.org/0000-0001-8950-9361)
- Chi Wang (ORCID: https://orcid.org/0000-0001-6991-9398)
- Hao Yin (ORCID: https://orcid.org/0000-0001-5666-7208)
- Haibin Wang (ORCID: https://orcid.org/0000-0002-6481-8712)
- Jun Wang (ORCID: https://orcid.org/0000-0002-6278-0677)
Institutions
- Chinese Academy of Sciences (CN)
- Institute of Acoustics (CN)
- University of Chinese Academy of Sciences (CN)
Publication Details
- Journal
- Journal of Marine Science and Engineering
- Published
- 2025-06-06
- DOI
- https://doi.org/10.3390/jmse13061136
- Citations
- 1
- Primary Topic
- Underwater Acoustics Research
- Type
- article
- Field-Weighted Citation Impact
- 1.79
Funders
- National Natural Science Foundation of China
- Chinese Academy of Sciences
- China Scholarship Council