Investigation of cavity defects in trees using an enhanced TR-MUSIC algorithm

This study presents an automated enhanced time-reversal multiple signal classification (TR-MUSIC) method for non-destructive detection of internal defects in trees. The proposed approach introduces an automated algorithm for signal subspace dimension optimisation. This algorithm integrates the minimum description length (MDL), Akaike information criterion (AIC), and Bayesian information criterion (BIC) frameworks into a hybrid information-theoretic criterion. It thereby automatically determines the optimal subspace dimension, eliminating the need for manual adjustment of the M-value required in traditional TR-MUSIC algorithms. Additionally, the approach designs an intelligent defect boundary auto-delineation mechanism using adaptive statistical thresholding algorithms based on local feature characteristics, enabling precise morphological reconstruction of complex-shaped defects through relative intensity values and local mean standard deviation analysis. By integrating ultra-wideband (UWB) pulse technology with background elimination strategies, the enhanced algorithm accurately localises and reconstructs defects with dimensions as small as half the central wavelength. Experimental validation demonstrates superior performance for both point-like and complex-shaped defects under varying dielectric conditions, significantly improving detection efficiency while maintaining high accuracy. The automated parameter optimisation enables real-time field deployment capability, providing a practical solution for forestry non-destructive testing (NDT) applications.