Machine Learning of Behavioural Models for Improved Multi-Sensor Fusion

用于改进多传感器融合的行为模型的机器学习

• 批准号: 2889812
• 金额: --
• 依托单位: University of Liverpool
• 依托单位国家: 英国
• 项目类别: Studentship
• 财政年份: 2023
• 资助国家: 英国
• 起止时间: 2023 至 无数据
• 项目状态: 未结题
• 来源: https://gtr.ukri.org/projects?ref=studentship-2889812
• 关键词: Machine; Learning; Behavioural; Models; Improved

SSensor fusion is increasingly critical in the defense and security sectors, enhancing decision-making by providing operators with integrated data from diverse sources. This capability is particularly valuable on platforms such as unmanned aerial vehicles (UAVs), where it enables less sophisticated sensors to collectively emulate the functionality of more advanced systems, thereby reducing costs and enhancing resilience. The efficacy of sensor fusion heavily relies on the accuracy of behavioral models used to predict the movements of surveillance targets. Current models, predominantly based on simplistic, nearly constant velocity assumptions, fall short in addressing complex behaviors and deception tactics used by targets.This PhD research aims to address these shortcomings by developing advanced behavioral models through machine learning techniques, utilizing asynchronous, historic multi-sensor data. Despite the potential of machine learning in this context, its application has been limited due to the complexity of dynamically modeling target behavior while managing non-linear measurements and handling detection anomalies. Recent advancements in computational techniques, such as extensions to Particle Markov Chain Monte Carlo (Particle-MCMC) that utilize gradient information and leverage modern deep learning compute resources, offer promising solutions to these challenges.This project will explore and adapt these cutting-edge techniques to improve the predictive accuracy of behavioral models derived from historic surveillance data. The research will assess the utility of these models within the framework of generic multi-sensor fusion and in scenarios specific to Leonardo-a leading entity in defense and security-potentially encompassing the integration of electronic surveillance, radar, and electro-optical/infrared (EO/IR) data. Success in this endeavor could significantly advance multi-sensor data fusion research, bolstering Leonardo's capabilities and positioning the researcher as a future leader in data science within a vital sector committed to national security.

SS传感器融合在国防和安全领域越来越重要，通过向运营商提供来自不同来源的集成数据来增强决策。这种能力在无人机(UAV)等平台上特别有价值，在这些平台上，它使不太复杂的传感器能够集体模拟更先进系统的功能，从而降低成本并增强弹性。传感器融合的有效性在很大程度上依赖于用于预测监视目标运动的行为模型的准确性。目前的模型主要基于简单的、近乎恒定的速度假设，不能处理目标使用的复杂行为和欺骗策略。本博士研究旨在通过利用异步的、历史的多传感器数据通过机器学习技术开发高级行为模型来解决这些缺点。尽管机器学习在这种背景下具有潜力，但由于在管理非线性测量和处理检测异常的同时动态建模目标行为的复杂性，其应用受到了限制。计算技术的最新进展，例如利用梯度信息和利用现代深度学习计算资源的粒子马尔科夫链蒙特卡罗扩展(粒子-MCMC)，为这些挑战提供了有希望的解决方案。本项目将探索和调整这些尖端技术，以提高从历史监控数据派生的行为模型的预测精度。这项研究将在通用多传感器融合的框架内，以及针对莱昂纳多--国防和安全领域的领先实体--可能包括电子监视、雷达和光电/红外(EO/IR)数据的集成的特定场景中，评估这些模型的实用性。这一努力的成功可能会极大地推动多传感器数据融合研究，增强莱昂纳多的能力，并将这位研究人员定位为未来数据科学领域的领导者，在这个致力于国家安全的关键部门。