Data Exploitation and processing for multi-sensor radar big data

情报、监视和侦察（ISR）系统由几种传感器组成，如雷达、声纳、光电、高光谱和红外。改进的传感器技术产生具有增强的能力的传感器，诸如可以实现上述改进的监视能力的增加的分辨率和灵敏度。然而，它们也导致传感器数据量以超过使用传统工具分析的速度增加。该项目旨在为多传感器和雷达大数据开发信号和图像处理。大数据有可能通过充分利用现有数据来扩展传统监测和监视系统的视野。传统的信号处理技术大多是基于最优性理论开发的，这对于大数据是不实用的。在大数据分析中，信号和图像处理可以采用两种通用方法来处理大量数据。第一种是利用分布式处理技术，将任务分解到分布式计算单元上，以减少单个计算单元的负载，提高计算速度。基于Hadoop的MapReduce框架是一种广泛使用的大数据存储和分析方法。第二种方法使用了近似的概念。它使用通常基于哈希的随机投影来降低维度，并试图获得接近基于完整数据集的结果。在这个项目中，新的信号处理和机器学习技术将基于近似，稀疏表示，分布式处理和集成方法开发，可以充分利用感知大数据中的信息来识别模式的变化，用于监视目的。这些技术将应用于由加拿大国防研究与发展部（DRDC）和工业合作伙伴提供的机载雷达、光电（EO）和红外（IR）数据。

Intelligence, Surveillance, and Reconnaissance (ISR) systems comprise of several sensors, such as radar, sonar,electro-optical, hyperspectral, and infrared. Improved sensor technologies generate sensors with enhanced capabilities such as increased resolution and sensitivity that can achieve the improved surveillance capability stated above. However, they also result in increasing amount of sensor data at a rate beyond using traditional tools to analyze. This project proposes to develop signal and image processing for multi-sensor and radar big data. Big data has the potential to stretch the horizon of traditional monitoring and surveillance systems by fully exploiting the data available. Conventional signal processing techniques are mostly developed on optimality theory which are not practical for big data. There are two general approaches in big data analytic that can be adopted by signal and image processing to process massive amount of data. The first one uses distributed processing to decompose a task on distributed computing units to reduce the load of individual computing units and to enhance computing speed. The Hadoop-based MapReduce framework is a widely used approach for big data storage and analytics. The second one uses the concept of approximation. It uses random projection usually based on hashing to reduce the dimensionality and tries to obtain results close to those based on the complete data set. In this project, new signal processing and machine learning techniques will be developed based on approximation, sparse representation, distributed processing and ensemble approach that can fully exploit the information in the sensory big data to identify change of patterns for surveillance purpose. These techniques will be applied to the airborne radar, electro-optical (EO) and infrared (IR) data provided by Defence Research and Development Canada (DRDC) and the industrial partners.