Optimising Ground Penetrating Radar data interpretation: A hybrid approach with AI-assisted Kalman Filter and Wavelet Transform for detecting and locating buried utilities

Abstract

Ground Penetrating Radar (GPR) is widely used for detecting buried utilities, but data interpretation remains challenging due to noise and clutter. Although various methods exist for processing GPR data, the Kalman Filter (KF) has been underutilised despite its strength as an estimator. Traditional KF-based algorithms in GPR studies often rely on chi-squared hypothesis testing, which requires expert-defined thresholds and can lead to biased or uncertain outcomes. This paper introduces a novel KF-based framework that addresses these limitations. The framework employs Kalman Filters for noise reduction, with an optimisation algorithm based on a genetic algorithm to fine-tune KF input parameters. A Normalised Innovation Squared (NIS) parameter is used to generate an NIS signal function for identifying anomalies. Additionally, discrete wavelet transforms are applied to the NIS signal function for anomaly detection and localisation, using varying decomposition levels and vanishing moments. Results demonstrate a proportional relationship between wavelet decomposition levels, selected wavelets, and the detection rates of true and false positives. Statistical analysis using receiver operating characteristic curves shows that the optimal detection rate for all tested wavelets occurs at decomposition levels 5 and 6. This framework enhances GPR data interpretation with minimal user interaction, representing a step forward toward autonomy in GPR data processing and interpretation.