Full-polarimetric Ground Penetrating Radar Underground Objects Classification Using Random Forest

In recent years, some people have applied the full-polarimetric GPR into the classification of underground objects and obtained good results. However, with the increase of the volume of data, the efficiency of calculation will decrease. Random forest is a method for large volume data and has been widely used in remote sensing. In this paper, we used the random forest for the classification of underground objects. For testing the feasibility, we performed laboratory experiment and used the decomposition parameters of Freeman-Durden 3-component as the features of classification. Finally, the accuracy of classification 78.57% was obtained and the final figures of classification further demonstrated the efficiency of the random forest for classification of underground objects. This paves the way for further practical research of machine leaning algorithm for full-polarimetric GPR classification.