基于真值发现的分布式多传感器协同融合研究

The multisensor collaborative sensing and information fusion in sensor networks are mainly faced with scientific problems including distributed sensor selection, the credibility of information fusion and the dynamics of fusion weight. This project aims at the above problems and utilizes truth discovery algorithm to study distributed multisensor collaborative fusion..The importance degree of monitoring area, dynamic network and heterogeneous data are considered to establish a distributed fusion weight estimation model, which is based on truth discovery. According to such fusion weight, a sensor selection algorithm is designed to select “high quality” sensor for collaborative fusion; Aiming at unreliable weight estimation problem caused by less measurement data of some sensors, the sensor-target distance and confidence interval of the measurement data are utilized to propose a confidence-aware fusion method in order to improve the credibility of information fusion; Based on Bayesian probability, the spatio-temporal correlation of the sensors and the truth discovery cumulative error are utilized, and a collaborative fusion method based on dynamic data is proposed to dynamically update the fusion weight and reduce the information fusion error; Performance evaluation metrics are established, and convergence analysis,complexity analysis and performance analysis are carried out. Experiment verification of the algorithms are conducted on the existing verification platform. The distributed multisensor collaborative fusion strategy proposed in this project will lay a foundation for the sensor network to achieve high performance and energy-efficient information fusion.

传感器网络中的多传感器协同感知与信息融合主要面临着：分布式传感器选择、信息融合的可信性及融合权重的动态性等科学问题。本项目针对上述问题，借鉴真值发现算法对分布式多传感器协同融合进行研究。.考虑监测区域重要度、动态网络及异构数据，建立基于真值发现的分布式融合权重估计模型，并依此设计传感器选择算法，选择“高质量”的传感器进行协同融合；针对部分传感器量测数据少导致融合权重估计不可信的问题，利用传感器与目标的距离和量测数据的置信区间，提出基于置信度的可信融合方法，以提高信息融合的可信度；基于贝叶斯概率，结合传感器时空相关性及真值发现累计误差，提出面向动态数据的协同融合方法，动态更新融合权重，降低信息融合误差；建立性能评价指标，进行收敛性、复杂度和性能分析，在已具有的测试验证平台上完成算法的实验验证。本项目提出的分布式多传感器协同融合策略将为传感网实现高性能和高能效的信息融合奠定基础。

本项目针对动态不确定环境下传感器网络协同融合问题开展了深入研究。已取得的研究成果主要包括：1）针对感知数据的异构性、传感网能量受限及分布式特性，基于真值发现算法，建立面向分布式多传感器的信息融合权重估计模型，在提高信息融合性能的同时，最大化资源利用率。2）针对感知数据的long-tail现象对目标检测精度的影响，将节点-目标距离和量测值-真实值误差的置信水平纳入真值发现优化解决方案框架来推断传感器节点的融合权重，提出了基于真值发现的可信信息融合框架，有效的提升了目标检测准确率；3）提出了一种注重能量调度的动态传感网信息融合策略，根据动态变化的网络拓扑结构、节点剩余能量和监测区域的重要程度计算节点重要性，设计并实现了感知因子的自适应过程，以匹配实际情况，并根据检测到的数据自动变化。该策略可以在很大程度上减少唤醒节点的数量，同时通过信息融合保持高可靠性；4）还进一步对非合作空间目标的相对测量进行研究，提出了一种基于点云特征的非合作目标的自主识别和姿态跟踪方法。该方法可以有效地找到非合作翻滚目标的特征，摆脱依赖于目标模型和数据库，实现非合作目标的位姿跟踪。在本项目的支持下，已在国际、国内期刊和会议上发表或录用学术论文8篇，包括本领域重要期刊Information Fusion论文1篇，IEEE Sensors Journal期刊论文1篇，已投稿在审论文1篇，获得国家发明专利3项，同时荣获2020年湖北省自然科学奖三等奖。

内容获取失败，请点击重试