Transfer-Learning-Based Ranging Error Mitigation With Mobile Anchor in Village Cadastral Survey

Abstract

Village cadastral survey is a time-consuming and laborintensive task that plays a key role in guiding economic development. Attempts to automate village cadastral surveys have been carried out for nearly a decade, and the core problem is that the high-precision positioning of boundary points has still not been solved, especially for areas the satellite positioning signal is affected by building obscuration. To facilitate automation of village cadastral survey, Ultra-wideband (UWB) ranging technology with cm-level accuracy was introduced into village cadastral survey, and a novel asynchronous anchor ranging localization mode was proposed to overcome the shortcomings of the traditional synchronous anchor setup. The novel neural network model and transfer learning method were proposed to better mitigate ranging errors. The training dataset was collected in the open outdoor scenario to obtain a pre-trained model and the transfer learning method was used to adjust the pre-trained model to fit the building corner environment. In the algorithm experiment, in the building occlusion environment, the upper limit error of the data is reduced from the original 16.10cm to 12.21cm, and the error median of data is reduced from the original 9.61cm to 4.75cm. In further localization experiments, the maximum positioning error was reduced from the original 0.1902m to 0.1357m, the experimental result shows that the proposed method successfully solves the problem of high-precision positioning of boundary points in the building obscuration environment.