DRIFTERS: Deep Radar Interpretation For Tracking and Enhancement of Raw Signal

深度学习在计算机视觉和语音识别等领域非常成功，因为它能够学习能够处理原始信号以实现高级任务的表示。在这个研究项目中，我们将开发基于深度学习的机载海事雷达信号处理方法，其中信号相对非结构化且噪声严重。更具体地说，我们将评估方法1）抑制海杂波和检测静态信号中的目标; 2）从序列雷达信号中检测目标; 3）从序列信号中跟踪目标。此外，深度学习需要覆盖各种背景的大数据集才能获得良好的结果，而公开的数据库通常仅限于特定的环境。因此，我们还将探索基于深度学习的方法来增强从仿真模型中获得的数据，以便增强数据集并使用具有真实特征的信号覆盖更多情况（例如，复杂的噪声分量）。这有可能导致发现更好的方法来处理雷达信号，从而提高基于这种技术的传感设备的性能，并可能对沿海监测产生积极影响，以加强国家安全和海洋环境的安全。

Deep learning has been highly successful in fields such as computer vision and speech recognition, due to its capacity to learn representations able to process raw signals to achieve high level tasks. In this research project, we will develop approaches based on deep learning for airborne maritime radar signal processing, where the signal is relatively unstructured and severely noisy. More specifically, we will assess approaches to 1) suppress sea clutter and detect targets in static signal; 2) detect targets from sequential radar signal; and 3) track targets from sequential signals. Moreover, deep learning requires big datasets covering a variety of contexts to attain good results, while the databases publicly available are generally limited to specific environments. Thus, we will also explore deep learning-based approaches to enhance the data obtained from simulation models, in order to augment the dataset and cover more situations with signals having realistic characteristics (e.g., sophisticated noise components). This has the potential of leading to the discovery of better ways to process radar signals and as such improve the performance of sensing devices based on this technology, with a possible positive impact on coastal surveillance for increasing national security and safety in maritime environments.