基于集成级联学习的多无人平台跨域协同多目标识别算法

Ensemble cascades learning algorithm for multi-platforms cross-domains cooperative multi-Targets recognition

针对目前各类平台获取的目标数据质量参差不齐且存在信号丢失情况,进而导致基于单一平台数据源的多目标识别准确率低的问题,提出一种基于集成级联学习的多无人平台跨域协同多目标识别算法.通过使用多个级联深度网络对多无人机(Multi-UAVs)平台获取的同一目标的多源异构数据进行特征学习,使得同一目标的多源异构数据能够根据数据特性进行有效的分层学习,并将学习到的多层高维特征在决策层通过集成学习实现多目标识别。首先,根据各类数据特性,分别构建不同层级的特征学习网络抽取不同维度的有效信息,为后续的融合提供同一目标更多的多源有效信息,弥补了单一数据源信息不足的问题;然后,将多维度的有效信息通过集成学习进行融合,实现同一目标的多源信息联合,与其他目标差异最大化,从而提升多目标识别的准确性。此外,本文提出的算法通过跨域多源数据融合学习能够解决单一数据源信号丢失后无法识别目标的问题。

Aiming at the problem of low recognition accuracy on multi-targets caused by uneven quality and signal loss of target data obtained by various platforms, a ensemble cascades learning algorithm for multi-platforms cross-domains cooperative multi-targets recognition is proposed. For efficient utilization of multi-source heterogeneous data from multi-UAVs for one target, multiple cascade deep network model is designed to extract multi-layer high-dimensional features according to all kinds of data characteristics and provide more effective information from multi-source data for following fusing.All extracted features are integrated though a ensemble learning classifier to achieve multi-source information fusion for one target and maximize the difference between one target and other targets, which enables better recognition accuracy. Furthermore,the problem of inability to identify targets caused by signal loss of single data-source can be alleviated by cross-domain multi-source data fusion learning.