Blind source separation of nonlinearly mixed plant leaf electrical signals using polynomial-mapped FastICA

Due to the nonlinear mixing of electrical activities generated by different cell types within leaf tissues under light induction, independent component analysis (ICA) methods often fail when processing non-invasively extracted plant electrophysiological signals. To overcome this limitation, a FastICA-based approach combined with polynomial mapping is adopted to perform blind source separation (BSS) of plant electrical signals with nonlinear characteristics. To examine these nonlinear properties, surrogate data generated through phase randomization were constructed. Using sample entropy as an indicator, results confirmed at a 95% confidence level that the recorded electrical signals possess intrinsic nonlinear dynamical characteristics. Subsequent simulations under multiplicative and exponential coupling conditions demonstrated that the second-order polynomial-mapped FastICA (Poly2) significantly outperformed conventional linear ICA, its third-order variant (Poly3), and B-spline-based post-nonlinear algorithms. Monte Carlo trials revealed that Poly2 achieved the highest Spearman correlation coefficients (S1: 0.82, 95% CI [0.78, 0.85]; S2: 0.87, 95% CI [0.83, 0.89]), exhibiting clear advantages over all baseline methods. Furthermore, when applied to real non-invasive leaf surface recordings, the algorithm successfully isolated independent components. Based on morphological similarity comparisons with classic literature-derived response curves, the results indicated that the Spearman correlation coefficients were 0.80 (95% CI [0.78, 0.82]) for S1 and 0.82 (95% CI [0.79, 0.87]) for S2, respectively, which presumably reflect the distinct electrophysiological characteristics of guard and mesophyll cells, demonstrating effective signal separation. Compared to conventional methods, the proposed approach demonstrates superior signal decoupling capabilities, significantly advancing the application of BSS techniques in precision agriculture and plant electrophysiology.