Weak Pulse Signal Distributed Fusion Detection with Multidimensional Nonlinear Correlation Subjected to Broad Learning

Abstract

Due to highly complex distributed sensors chaotic environments in sea monitoring, machinery fault diagnosis, and EEG weak signal detection, the data available to realize the effective detecting tasks of neural networks are often insufficient. Compared with machine learning models, a statistical approach with multidimensional nonlinear correlation exhibits a unique and extraordinary signal pattern predicting capacity and has a simple and powerful framework for signal processing. However, the application of multidimensional nonlinear correlation directly to weak pulse signal detection remains limited. To overcome these limitations and reach highly precision signal detection, a novel multidimensional nonlinear correlation (MNC) based on phase reconstructing and manifold broad learning is investigated for distributed sensor fusion detection under chaotic noise. Firstly, the distributed observation data is reconstructed into fixed-size arrays with phase space reconstructing, and the high dimensional sequence of the manifold broad learning of these tuples is used as an input of nonlinear correlation for extracting spatio-temporal features. Subsequently, a multidimensional nonlinear correlation with a QRS detector layer is developed to predict and classify the existence of a weak pulse signal. The MNC is integrated in the high feature enhanced space of the source domain and the detection fusion of distributed sensors is realized through the majority principle. Simulation experiments results show illustrates the effectiveness and robustness of the proposed MNC method combined with phase reconstructing and manifold broad learning strategy for distributed sensors weak pulse signal fusion detection in a chaotic background.