Abstract Due to the rapid development of deep learning techniques, numerous frameworks including convolutional neural networks (CNNs), deep belief networks (DBNs) and auto-encoder (AE), etc., have been established. In this context, advances in marine object recognition have been dramatically boosted, especially in the past decade. In this paper, we exclusively focus on an intensive review on deep-learning-based object recognition for both surface and underwater targets. To facilitate a comprehensive review, key concepts and typical architectures are firstly summarized in a unified framework. Accordingly, popular/benchmark datasets for marine object recognition are thoroughly collected and deep learning methodologies are comprehensively analyzed with intensive comparisons. Moreover, experimental results and futuristic trends in marine object recognition are intensively discussed. Finally, conclusions on state-of-the-art marine object recognition using deep learning techniques are drawn.

中文翻译：

---

海洋物体识别深度学习技术综述：架构和算法

摘要 由于深度学习技术的快速发展，已经建立了包括卷积神经网络（CNN）、深度信念网络（DBN）和自动编码器（AE）等在内的众多框架。在这种背景下，海洋物体识别的进步得到了极大的推动，尤其是在过去十年中。在本文中，我们专注于对基于深度学习的水面和水下目标物体识别的深入审查。为了便于全面回顾，首先将关键概念和典型架构总结在一个统一的框架中。因此，彻底收集了用于海洋物体识别的流行/基准数据集，并通过深入比较全面分析了深度学习方法。而且，深入讨论了海洋物体识别的实验结果和未来趋势。最后，得出了使用深度学习技术进行最先进的海洋物体识别的结论。