Machine Learning Assisted Three-dimensional Sparse Point Clouds Filling for Geological Spatial Modeling

Contact detectors can be used in harsh geological environments. In contrast to laser and radar detectors, which are affected by environmental influences such as dust on light and radar waves, contact detectors are not affected by the environment. However, the spatial data obtained by contact detectors is sparse compared to laser and radar detectors. To improve the detection accuracy of contact detectors, polynomial regression models and support vector regression (SVR) models from traditional machine learning are used and compared to the classical mathematical model, Newton interpolation. The data will first be split to simplify the training complexity while improving the model accuracy. At the high sparsity, the polynomial regression model has an $R^{2}$ of 0.90 and an MSE of 0.0078, and SVR has an $R^{2}$ of 0.93 and an MSE of 0.0043, both of which are better than Newton interpolation. Moreover, the longest training time is only 3.136 s, which is acceptable. Because traditional machine learning requires less hardware equipment for the detector, the detector can be equipped with the polynomial regression model for relatively simple spatial detection, while the detector can be equipped with the SVR model for complex spatial detection or when the accuracy requirement is higher.