Bayesian Learning for Object Recognition from Noisy Time Series Data.

从嘈杂的时间序列数据中进行对象识别的贝叶斯学习。

• 批准号: 2597698
• 金额: --
• 依托单位: University of Cambridge
• 依托单位国家: 英国
• 项目类别: Studentship
• 财政年份: 2021
• 资助国家: 英国
• 起止时间: 2021 至 无数据
• 项目状态: 未结题
• 来源: https://gtr.ukri.org/projects?ref=studentship-2597698
• 关键词: Bayesian; Learning; Object; Recognition; Noisy

This research aims to capitalize on recent advances in Bayesian modelling and supervised learning to introduce a novel framework for robust target recognition from time-series data, with particular focus on radar, e.g. unmanned air-traffic management applications or robotics. The proposed approach can not only treat noisy sensory observations with intermittent and potentially asynchronous salient features, but also effectively exploit underlying spatio-temporal dependencies to achieve improved sequential target classification results. Amongst the key tackled challenges is maintaining computational as well as training-data efficiency and interpretability of the develop technique. Real data from Aveillant's Gamekeeper radar is expected to be made available to evaluate and benchmark (e.g. versus convolutional neural networks or other standard machine learning classifiers) the performance of the introduced method(s).

本研究旨在利用贝叶斯建模和监督学习的最新进展，引入一种新的框架，用于从时间序列数据中进行鲁棒目标识别，特别关注雷达，例如无人驾驶空中交通管理应用或机器人。该方法不仅可以处理具有间歇性和潜在异步显著特征的噪声感官观测，而且可以有效地利用潜在的时空依赖性来获得改进的序列目标分类结果。解决的关键挑战之一是保持计算以及训练数据的效率和开发技术的可解释性。来自Aveillant的Gamekeeper雷达的真实数据有望用于评估和基准测试（例如与卷积神经网络或其他标准机器学习分类器相比）所引入方法的性能。