Developing Novel Bayesian Track Before Detect Approaches for Maritime Big Data Challenges

本课题的目的是研究一种新的信号处理方法，从海杂波背景中分离出低可观测的运动目标。在静止和运动的海监雷达中，海杂波的返回峰值往往远远大于小目标的峰值。通过观察10秒到100秒的场景，多扫描处理允许检测和跟踪目标10到20分贝的“次杂波”，即使在杂波频谱内也是如此。该项目将研究通过从雷达数据中提取更多细节可以实现的改进。特别是，从杂波频谱中提取次杂波运动目标的技术，如先跟踪后检测，是本项目感兴趣的一个关键领域。例如，海面分析可用于调整这类算法，以将该问题作为检测场景中偏离波场预期反射率行为的异常之一来处理。性能改进、处理负荷和算法或交互算法集的复杂程度之间的关系将在该项目下进行探索。以前的研究已经确定了新的有效的蒙特卡罗算法，该算法利用基于梯度的MCMC建议，可以在项目早期确定的场景中提供有用的增强目标提取性能。该项目的范围预计还包括雷达传感器的特征，包括Hensoldt UK生产的雷达传感器，深入探索海杂波的特征和感兴趣的目标，探索计算技术和机器学习，以利用上述计算范式处理传感器数据。该项目可能会得到相关学术伙伴(由Hensoldt提供部分资金)和Hensoldt工作人员的研究成果的补充，以及与海洋研究和技术界的联系，从2023年秋季开始的重点数据收集活动，自2023年初以来在克兰菲尔德什里文纳姆校区的博士学位，以及由Hensoldt UK协调的用户社区的背景。

This project aims to research novel signal processing methods to separate low observable moving targets from a sea clutter background in static and moving maritime surveillance radar, which frequently see sea clutter return peak amplitudes that are far greater than those from small targets. By observing scenes for 10s to 100s of seconds, multi-scan processing allows detection and tracking of targets 10 to 20dB 'sub-clutter', even within the clutter spectrum. This project will study improvements achievable by extracting further detail from the radar data. In particular, techniques such as track before detect to extract sub-clutter moving targets from within the clutter spectrum is a key area of interest for this project. Sea surface analysis could be exploited in the tuning of such algorithms, for example, to approach the problem as one of detecting anomalies in a scene that deviate from expected reflectivity behaviour in a wave field. The relationship between performance improvements, the processing load and level of complexity of algorithms or interacting algorithm set, is to be explored under the project. Previous research has identified novel effective Monte Carlo algorithms which utilise gradient-based MCMC proposals which can provide useful enhancements to target extraction performance in scenarios that will be determined early in the project. The scope of the project is expected to also include characterisation of radar sensors including those produced by Hensoldt UK, deep exploration of the characteristics of sea-clutter, and of targets of interest, and exploration of computational techniques and machine learning to leverage the above mentioned computational paradigm with sensor data.The project may be complemented by research outcomes from associated academic partners, part-funded by Hensoldt, and Hensoldt staff, as well as liaison with Oceanographic Research and technology community, a focussed data collection campaign commencing autumn 2023, a PhD at the Cranfield Shrivenham campus since early 2023 and context from a user community coordinated by Hensoldt UK.