基于拓扑结构的无线传感网多模数据稀疏修复

Wireless Sensor Networks (WSN) is widely used in many fields such as environment monitoring. However, data of some sensor nodes may be missing due to various reasons such as sensor failaures. Besides, most autonomous data learning techniques are not robust to incomplete data. Thus, it is very important to recover these missing data. Sparsity is one of the powerful features of the data in practical applications. Sparsity-based algorithms have shown very promising results in data recovery and compression in WSN. Our previous work found that the keys to successful recovery in WSN using sparsity-based approach are designing a suitable dictionary to sparsely represent WSN data and setting up a good recovery model to make full use of the available data information. This project aims to significantly reduce the recovery errors by constructing a topology-based dictionary and designing multimodal missing data recovery models. Research work will be conducted as follows: 1) build a topology-based dictionary by exploiting the physical node connections or the similarities of the node data; 2) propose a jointly sparse representation for the multimodal sensor data by exploiting the correlation among multimodal spatial-temporal data; 3) propose jointly sparse recovery models for multimodal sensor data based on 1) and 2) to reduce the recovery errors. This project will promote the development and applications of network signal processing methods, and enhance the adaptability of WSN to different environments and data mining methods.

无线传感网广泛应用于环境监控等领域，但由于节点失效等多种原因导致部分数据丢失。丢失数据使得很多数据挖掘方法难以直接应用，而节点重传会增加能量消耗，因此修复这些数据非常重要。数据的稀疏性广泛存在，它在传感网数据修复和压缩有着重要应用。我们前期研究发现稀疏修复成功的关键在于合理的稀疏表示字典和充分挖掘数据信息的稀疏修复模型。本项目拟从传感网特有的网络连接和多模式数据特点，提出基于拓扑结构的多模式传感数据稀疏修复模型，提高数据修复质量。研究内容包括：1)利用传感网物理连接或数据相似性得到的拓扑结构，建立表示不规则网格数据的稀疏字典；2)考虑时-空数据多模式间的关联性，建立多模数据的联合稀疏表示方法；3）建立基于1)和2)的多模数据联合修复模型，降低数据修复误差。本项目的实施可以促进网络信号处理方法的发展和应用，增强无线传感网对环境和数据挖掘方法的适应性。

项目执行中，主要进行了传感网数据的Graph拓扑结构表示、稀疏修复模型和最优化算法的研究及其在成像科学中的应用，提出了联合时间-空间的Graph稀疏重建方法；将多模信号建立成关联的多通道信号，利用四元素结构和低秩特性，建立了多模信号的联合低秩重建模型和快速重建算法；提出了多模医学图像融合和医学图像超分辨率重建方法；此外，项目还探索在医学成像等相关领域的信号自适应稀疏表示及稀疏重建模型和快速算法研究。项目基本按计划执行，并取得了丰富的研究结果。