Integrating Three-Class Classification Strategy and Cross-Domain Feature Sets for UWB Nonline-of-Sight Identification and Error Correction

Aiming at the problem that it is difficult to accurately identify Ultra-Wide-Band (UWB) Non-Line-of-Sight (NLOS) signals and there exists ranging deviation, this paper proposes a UWB NLOS identification and error correction method that integrates a three-class classification strategy and a cross-domain feature set. Firstly, the UWB signal propagation states are divided into Line-of-Sight (LOS), NLOS, and High Confidence Non-Line-of-Sight (HCNLOS), improving the positioning accuracy and efficiency in different occlusion scenarios. A cross-domain feature set is then constructed by combining features from the signal and spatial domains, and a three-class classification model is built using Light Gradient Boosting Machine (LightGBM) to accurately identify the UWB signal states. To further enhance the accuracy of the model, a hybrid algorithm based on Differential Evolution Adaptive (DEA) and Particle Swarm Optimization (PSO) is constructed to optimize the hyper-parameters of LightGBM. Finally, for the identified NLOS and HCNLOS data, the ranging deviation is reduced using the LightGBM error-correction model, resulting in accurate positioning results. Experimental results show that the proposed method achieves a Macro-F1 score of 79.86% for three-class classification in typical indoor scenarios, with positioning root-mean-square errors of 0.0121 m and 0.0142 m in the X and Y directions, respectively.