Deep learning for acoustic detection of submarines

用于潜艇声学探测的深度学习

• 批准号: 531161-2018
• 负责人: Bouchard, Martin
• 金额: $ 1.82万
• 依托单位: University of Ottawa
• 依托单位国家: 加拿大
• 项目类别: Engage Grants Program
• 财政年份: 2018
• 资助国家: 加拿大
• 起止时间: 2018-01-01 至 2019-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=645900
• 关键词: Deep; learning; acoustic; detection; submarines

This project is devoted to the development of a pattern recognition system for underwater acoustic signals used**in submarine detection. 2-D images will be built from the acoustic signals and their features, and image**classification algorithms will be applied to those images. In particular, the use of deep convolutional networks**will be thoroughly investigated, including strategies to reduce their size and complexity. In order to reduce the**computational and memory requirements, the network will be iteratively reduced in size by removing**redundant connections or connections of lower weight. We will also investigate the possibility of reducing the**number of layers. The use of a cascade of classifiers is also envisioned for improving the efficient of the**detector. A system (i.e., executable software code) that uses pattern recognition to identify signatures from**acoustic spectra will be delivered to the partner.

本计画系致力于发展一种可用于侦测潜器之水声讯号模式辨识系统。将从声学信号及其特征构建2-D图像，并将图像 ** 分类算法应用于这些图像。特别是，深度卷积网络 ** 的使用将被彻底研究，包括减少其大小和复杂性的策略。为了减少 ** 计算和内存需求，网络将通过删除 ** 冗余连接或较低权重的连接来迭代地减小大小。我们还将研究减少层数的可能性。还设想使用分类器的级联来提高 ** 检测器的效率。一种系统（即，可执行软件代码）将被递送给合作伙伴，该软件代码使用模式识别来识别来自声频谱的签名。