Double-layer robust broad logistic regression for detecting weak pulse signal in chaotic interference

Abstract

Owing to the extreme complexity of detecting weak signals submerged in chaotic interference, traditional neural network detection methods based on hypothesis testing are difficult to sufficiently portray the data and chaotic noise evolution characterization for achieving high effective training, and the hypothesis test usually has the shortcomings of poor generalization performance. In this article, a double-layer robust broad logistic regression (DRBLR) is proposed to overcome the aforementioned issues and achieve highly accurate detection. The proposed method reconstructs the observation signal into a high-dimensional phase space with fixed-size tuples, and it extracts both the chaotic evolution trend and nonlinear structure from these spatiotemporal sequences by broad enhancement and manifold embedding operation of the Robust Manifold Broad Learning System. After that, the weights in Logistic Regression with two layers can be trained based on the obtained features and their corresponding pulse signal labels. The strategy of the DRBLR model leads to a robust learning paradigm and obtains better generalization performance. Experiments to detect weak pulse signals are conducted using Lorenz and sunspot datasets. Simulation results illustrate that the proposed method can better capture the characteristics of the chaotic interference evolution and effectively detect the weak pulse signal in the chaotic noise background.