Toward the optimisation of the Kalman Filter approach in ground penetrating radar application for detection and locating buried utilities

Ground Penetrating Radar (GPR) stands as a pivotal non-destructive tool for identifying and assessing buried utilities. However, the noisy GPR radargram data requires extensive expert interpretation, making it time-consuming and subjective. While various signal processing techniques, like time-zero correction and background subtraction, help mitigate noise, mathematical estimators like the Kalman Filter (KF) and particle filter present advanced solutions. Notably, KF, known for its efficiency and computational benefits, excels in denoising and decluttering GPR radargrams. This research introduces an innovative KF-based optimisation algorithm tailored to minimise user input and pinpoint buried utilities and anomalies. Leveraging the distinctive KF parameter, Normalised Innovation Squared (NIS), the algorithm aims for enhanced target detection. A genetic algorithm-driven multi-objective optimisation model evaluates the method's efficacy, focusing on the receiver operating characteristic (ROC) and mean of innovations. Cost functions considered encompass noise covariances and optimal Fourier analysis frequency. Preliminary results showed a 23.13% rise in the area under the ROC with optimised parameters (91.4%) compared to user-selected ones (68.27%). This method not only reduces GPR data noise but also augments the detection of buried utilities. Traditional chi-squared hypothesis testing was replaced with an NIS signal function analysis, facilitating more refined anomaly detection. Collectively, this study offers a transformative approach to GPR data post-processing, emphasising efficiency and reduced user dependency.