Improving buried object localization accuracy in heterogeneous environments using TOA-based IR-UWB and machine learning

Abstract

This paper proposes a buried-object localization method in heterogeneous environments using time of arrival (TOA)-based IR-UWB radar and machine learning techniques. A propagation model is developed to generate a TOA dataset for buried-object locations under multipath and noisy conditions. To improve TOA estimation, a matched-filtering and adaptive-thresholding (MFAT) technique is proposed to enhance reflected pulse detection at low SNR. The extracted TOA features are then used to train Random Forest (RF), XGBoost, and LightGBM (LGBM) models for object localization. Experimental results show that the proposed LGBM-MFAT approach achieves the best performance, with an MAE of 1.26 cm, an RMSE of 2.37 cm, and an R² value of 0.999. The results demonstrate that combining adaptive TOA processing with gradient-boosting models significantly improves localization accuracy and provides an effective solution for penetrating IR-UWB radar systems operating in heterogeneous environments.